(Ultra‐)long‐acting insulin analogues for people with type 1 diabetes mellitus

Summary of findings 1. Summary of findings: insulin detemir versus NPH insulin

Outcomes	NPH insulin	Insulin detemir	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin detemir compared with NPH insulin for T1DM
Patients: people with T1DM Settings: outpatients Intervention: insulin detemir Comparison: NPH insulin
All‐cause mortality Follow‐up: 24‐104 weeks	See comment		Peto OR4.97 (0.79 to 31.38)	3334 (9)	⊕⊕⊕⊝ moderate^a	All 5 deaths reported in 2 studies including adults occurred in the insulin detemir group
Health‐related quality of life Description: diabetes health profile; insulin therapy‐related quality of life at night (scale not specified) Follow‐up: 26‐48 weeks	See comment			870 (3)	⊕⊕⊝⊝ low^b	No study reported health‐related quality of life in a format making it suitable for meta‐analysis 1 study including adults reported higher scores in the insulin detemir group vs the NPH insulin group (Kobayashi 2007) 2 studies did not show evidence of a difference between intervention groups (NCT00595374 included children; Standl 2004 included adults)
Severe hypoglycaemia (n/N) Definition: hypoglycaemia requiring third party assistance (Bartley 2008; Kobayashi 2007; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003); episodes where the children were semi‐conscious, unconscious or in a coma, with or without convulsions (Thalange 2013) Follow‐up: 24‐104 weeks	115 per 1000	79 per 1000 (60 to 106)	RR 0.69 (0.52 to 0.92)	3219 (8)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between 0.34 and 1.39 5 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
Non‐fatal myocardial infarction/stroke Definition: myocardial infarction Follow‐up: 24 months	See comment			495 (1)	⊕⊕⊝⊝ low^d	1/331 participants in the insulin detemir group vs 0/164 participants in the NPH insulin group experienced a non‐fatal myocardial infarction (Bartley 2008) Stroke was not reported Study included adults
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring 23:00‐06:00 (Bartley 2008; NCT00605137; Russell‐Jones 2004; Standl 2004; Vague 2003); occurring 22:00‐07:00 (Robertson 2007; Thalange 2013) Follow‐up: 24 weeks ‐ 24 months	54 per 1000	36 per 1000 (21 to 64)	RR 0.67 (0.39 to 1.17)	2925 (7)	⊕⊕⊕⊝ moderate^e	The 95% prediction interval ranged between 0.16 and 2.87 4 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
Serious adverse events (n/N) Follow‐up: 24‐104 weeks	82 per 1000	78 per 1000 (62 to 100)	RR 0.95 (0.75 to 1.21)	3332 (9)	⊕⊕⊕⊝ moderate^e	The 95% prediction interval ranged between 0.71 and 1.27 6 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
HbA1c (%) Follow‐up: 24 weeks ‐ 24 months	The mean HbA1c ranged across the NPH insulin groups from 7.3% to 8.6%	The mean HbA1c in the insulin detemir groups was 0.01% higher (0.1% lower to 0.1% higher)	—	3122 (8)	⊕⊕⊕⊝ moderate^e	The 95% prediction interval ranged between ‐0.1% and 0.1% 5 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; CSR: clinical study report; HbA1c: glycosylated haemoglobin A1c; n/N: number of people experiencing an event; NPH: neutral protamine Hagedorn; OR: odds ratio RR: risk ratio; T1DM: type 1 diabetes mellitus.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by one level because of indirectness (insufficient time frame) ‐ see Appendix 1. ^bDowngraded by two levels because of overall risk of bias ('some concerns') and imprecision (few studies) ‐ see Appendix 1. ^cDowngraded by one level because of inconsistency (95% prediction interval consistent with benefit and harm) ‐ see Appendix 1. ^dDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 1. ^eDowngraded by one level because of imprecision (CI consistent with benefit and harm) ‐ see Appendix 1.

Summary of findings 2. Summary of findings: insulin glargine versus NPH insulin

Outcomes	NPH insulin	Insulin glargine	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin glargine compared with NPH insulin for T1DM
Patients: people with T1DM Settings: outpatients Intervention: insulin glargine Comparison: NPH insulin
All‐cause mortality Follow‐up: 24‐52 weeks	See comment		Peto OR0.14 (0.00 to 6.98)	2175 (8)	⊕⊕⊕⊝ moderate^a	1 study including adults reported 0/1207 participants died in the insulin glargine group vs 1/1068 participants in the NPH insulin group 4 studies included adults, 4 studies included children (the test for subgroup differences could not be performed)
Health‐related quality of life Scales: Well‐Being Enquiry for Diabetics; General Well‐being; Diabetes Quality of Life for Youth and Parents' Diabetes Quality of Life Follow‐up: 24‐28 weeks	See comment			1013 (4)	⊕⊕⊝⊝ low^b	1 study including adults (Bolli 2009) reported greater improvements in the insulin glargine group compared with NPH insulin in one domain (diabetes‐related worries) There was no evidence of a difference in 3 studies (Chase 2008 included children; Home 2005 and Ratner 2000 included adults)
Severe hypoglycaemia (n/N) Definition: symptomatic hypoglycaemia requiring third party assistance, with either a blood glucose level < 2.8 mmol/L or prompt recovery after administration of oral carbohydrate, iv glucose or glucagon (Fulcher 2005; Home 2005; Schober 2002); requiring third party assistance and associated with either blood glucose < 2.0 mmol/L or prompt recovery after oral carbohydrate, iv glucose, or intramuscular or subcutaneous glucagon administration (Chase 2008); hypoglycaemia requiring third party assistance or involving a seizure, coma, unconsciousness or the use of glucagon (Liu 2016); hypoglycaemia requiring third party assistance (Porcellati 2004; PRESCHOOL; Ratner 2000) Follow‐up: 24‐52 weeks	125 per 1000	105 per 1000 (84 to 130)	RR 0.84 (0.67 to 1.04)	2350 (9)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between 0.65 and 1.09 5 studies included adults, 4 studies included children (the test for subgroup differences did not indicate interaction)
Non‐fatal myocardial infarction/stroke Definition: myocardial infarction/cerebral ischaemia Follow‐up: 28 weeks	See comment			585 (1)	⊕⊕⊝⊝ low^d	No participant experienced a non‐fatal myocardial infarction 1 study including adults reported 0/292 participants in the insulin glargine group vs 1/293 participants in the NPH insulin group experienced cerebral ischaemia (Home 2005)
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring 23:00‐07:00 (PRESCHOOL); severe hypoglycaemia occurring after the evening insulin injection and before the morning insulin dose (Fulcher 2005); severe hypoglycaemia occurring during sleep between bedtime and rising in the morning, or before the morning pre‐breakfast self‐blood glucose measurement and the morning insulin injection (Home 2005); severe hypoglycaemia occurring while asleep after the bedtime insulin dose and before the morning insulin dose and before the morning blood glucose measurement (Ratner 2000); severe hypoglycaemia while the participant was sleeping between bedtime and after the evening injection and before getting up in the morning (Schober 2002); severe hypoglycaemia occurring 00:00‐06:00 (Chase 2008) Follow‐up: 24‐28 weeks	87 per 1000	72 per 1000 (54 to 97)	RR 0.83 (0.62 to 1.12)	1893 (6)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between 0.54 and 1.27 3 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
Serious adverse events (n/N) Follow‐up: 24‐30 weeks	100 per 1000	108 per 1000 (63 to 184)	RR 1.08 (0.63 to 1.84)	2229 (8)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between 0.22 and 5.21 4 studies included adults, 4 studies included children (the test for subgroup differences did not indicate interaction)
HbA1c (%) Follow‐up: 24 weeks ‐ 1 year	The mean HbA1c ranged across the NPH insulin groups from 7.1% to 7.3%	The mean HbA1c in the insulin glargine groups was 0.02% higher (0.1% lower to 0.1% higher)	—	2285 (9)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between ‐0.5% and 0.5% 5 studies included adults, 4 studies included children (the test for subgroup differences did not indicate interaction)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). a.m.: ante meridiem; CI: confidence interval; HbA1c: glycosylated haemoglobin A1c; iv: intravenous; n/N: number of people experiencing an event; NPH: neutral protamine Hagedorn; RR: risk ratio; T1DM: type 1 diabetes mellitus.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by one level because of indirectness (insufficient time frame) ‐ see Appendix 2. ^bDowngraded by two levels because of overall risk of bias ('some concerns') and imprecision (few studies) ‐ see Appendix 2. ^cDowngraded by one level because of imprecision (CI consistent with benefit and harm) ‐ see Appendix 2. ^dDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 2.

Summary of findings 3. Summary of findings: insulin detemir versus insulin glargine

Outcomes	Insulin glargine	Insulin detemir	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin detemir compared with insulin glargine for T1DM
Patients: people with T1DM Settings: outpatients Intervention: insulin detemir Comparison: insulin glargine
All‐cause mortality Follow‐up: 26 and 52 weeks	See comment			763 (2)	⊕⊕⊝⊝ low^a	No participant died 2 studies included adults
Health‐related quality of life	Not reported
Severe hypoglycaemia (n/N) Definition: hypoglycaemia requiring third party assistance Follow‐up: 26 and 52 weeks	116 per 1000	68 per 1000 (15 to 304)	RR 0.59 (0.13 to 2.63)	763 (2)	⊕⊝⊝⊝ very low^b	2 studies included adults
Non‐fatal myocardial infarction/stroke Definition: non‐fatal myocardial infarction/stroke Follow‐up: 52 weeks	See comment			443 (1)	⊕⊕⊝⊝ low^a	1 study including adults reported 1/299 participants in the insulin detemir group vs 1/144 participants in the insulin glargine group experienced a non‐fatal myocardial infarction One study including adults reported 2/299 participants in the insulin detemir group vs 0/144 participants in the insulin glargine group experienced a non‐fatal stroke
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring from 11 p.m. to 6 a.m. Follow‐up: 26 and 52 weeks	50 per 1000	27 per 1000 (3 to 253)	RR 0.55 (0.06 to 5.12)	763 (2)	⊕⊝⊝⊝ very low^b	2 studies included adults
Serious adverse events (n/N) Follow‐up: 26 and 52 weeks	59 per 1000	102 per 1000 (54 to 195)	RR 1.72 (0.91 to 3.28)	763 (2)	⊕⊕⊝⊝ low^c	The fixed‐effect statistical model showed an RR of 1.79 (1.04 to 3.08) in favour of insulin glargine 2 studies included adults
HbA1c (%) Follow‐up: 26 and 52 weeks	The mean HbA1c ranged across the insulin glargine groups from 7.6% to 8.2%	The mean HbA1c in the insulin detemir groups was 0.01% lower (0.1% lower to 0.1% higher)	—	763 (2)	⊕⊕⊝⊝ low^c	2 studies included adults
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). a.m.: ante meridiem; CI: confidence interval; HbA1c: glycosylated haemoglobin A1c; n/N: number of people experiencing an event; p.m.: post meridiem; RR: risk ratio; T1DM: type 1 diabetes mellitus.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 3. ^bDowngraded by three levels because of inconsistency (point estimates varied widely, non‐consistent direction of effect) and serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 3. ^cDowngraded by two levels because of serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 3.

Summary of findings 4. Summary of findings: insulin degludec versus insulin detemir

Outcomes	Insulin detemir	Insulin degludec	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin degludec compared with insulin detemir for T1DM
Patients people with T1DM Settings: outpatients Intervention: insulin degludec Comparison: insulin detemir
All‐cause mortality Follow‐up: 26 weeks	See comment			802 (2)	⊕⊕⊝⊝ low^a	No participant died 1 study included adults, 1 study included children
Health‐related quality of life Scale: Short‐Form 36 version 2 (higher values mean better health‐related quality of life) Follow‐up: 26 weeks	Physical health score: the mean score in the insulin detemir group was 52.5 Mental health score: the mean score in the insulin detemir group was 52.5	Physical health score: the mean score in the insulin degludec group was 0.60 points lower (1.83 points lower to 0.63 points higher) Mental health score: the mean score in the insulin degludec group was 3.00 points lower (4.44 points lower to 1.56 points lower)	—	454 (1)	⊕⊕⊝⊝ low^b	Physical health score: MID is 2‐3 points Mental health score: MID is 3 points Study included adults
Severe hypoglycaemia (n/N) Definition: hypoglycaemia requiring third party assistance (Davies 2014) or altered mental status and cannot assist in their own care, is semiconscious or unconscious, or in a coma ± convulsions and may require parenteral therapy (glucagon or iv glucose) (BEGIN Young) Follow‐up: 26 weeks	122 per 1000	143 per 1000 (99 to 207)	RR 1.17 (0.81 to 1.69)	802 (2)	⊕⊕⊝⊝ low^c	1 study included adults, 1 study included children (the test for subgroup differences did not indicate interaction)
Non‐fatal myocardial infarction/stroke Definition: non‐fatal myocardial infarction/stroke Follow‐up: 26 weeks	See comment			453 (1)	⊕⊕⊝⊝ low^a	No participant experienced a non‐fatal myocardial infarction or stroke Study included adults
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring 00:01‐05:59 (Davies 2014) or 23:00‐07:00 (BEGIN Young) Follow‐up: 26 weeks	31 per 1000	34 per 1000 (16 to 75)	RR 1.12 (0.51 to 2.46)	802 (2)	⊕⊕⊝⊝ low^c	1 study included children, 1 study included adults (the test for subgroup differences did not indicate interaction)
Serious adverse events (n/N) Follow‐up: 26 weeks	73 per 1000	92 per 1000 (56 to 150)	RR 1.25 (0.76 to 2.05)	802 (2)	⊕⊕⊝⊝ low^c	1 study included children, 1 study included adults (the test for subgroup differences did not indicate interaction)
HbA1c (%) Follow‐up: 26 weeks	The mean HbA1c in the insulin glargine groups was 7.3%	The mean HbA1c in the insulin detemir groups was 0.05% lower (0.1% lower to 0.2% higher)	—	802 (2)	⊕⊕⊝⊝ low^c	1 study included children, 1 study included adults (the test for subgroup differences did not indicate interaction)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). a.m.: ante meridiem; CI: confidence interval; HbA1c: glycosylated haemoglobin A1c; iv: intravenous; MID: minimal important difference; n/N: number of people experiencing an event; p.m.: post meridiem; RR: risk ratio; T1DM: type 1 diabetes mellitus.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 4. ^bDowngraded by two levels because of overall risk of bias ('some concerns') and imprecision (few studies) ‐ see Appendix 4. ^cDowngraded by two levels because of serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 4.

Summary of findings 5. Summary of findings: insulin degludec versus insulin glargine

Outcomes	Insulin glargine	Insulin degludec	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin degludec compared with insulin glargine for T1DM
Patients: people with T1DM Settings: outpatients Intervention: insulin degludec Comparison: insulin glargine
All‐cause mortality Follow‐up: 26 ‐ 52 weeks	3 per 1000	4 per 1000 (0 to 36)	Peto OR 1.34 (0.15 to 11.93)	973 (3)	⊕⊝⊝⊝ very low^a	A total of 3/646 participants in the insulin degludec group vs 1/327 participants in the insulin glargine group died 2 studies included adults 1 study included children
Health‐related quality of life Scale: Short‐Form 36 version 2 (higher values mean better health‐related quality of life) Follow‐up: 32 and 52 weeks	Physical health score: the mean score ranged across the insulin glargine groups from 50.6 to 51.8 Mental health score: the mean score ranged across the insulin glargine groups from 49.9 to 50.4	Physical health score: the mean score in the insulin degludec groups was 0.04 points lower (1.21 points lower to 1.13 points higher) Mental health score: the mean score in the insulin degludec groups was 0.09 points lower (1.03 points lower to 0.85 points higher)	—	1042 (2)	⊕⊝⊝⊝ very low^b	Physical health score: MID is 2‐3 points Mental health score: MID is 3 points 2 studies included adults
Severe hypoglycaemia (n/N) Definition: hypoglycaemia requiring third party assistance (BEGIN Flex T1; BEGIN Young) or an event associated with impaired consciousness or seizure (Urakami 2017) Follow‐up: 24 and 52 weeks	102 per 1000	124 per 1000 (83 to 185)	RR 1.22 (0.82 to 1.82)	970 (3)	⊕⊕⊝⊝ low^c	2 studies included adults 1 study including children reported no child experienced severe hypoglycaemia (Urakami 2017)
Non‐fatal myocardial infarction/stroke Definition: non‐fatal myocardial infarction/cerebral ischaemia Follow‐up: 24 and 52 weeks	See comment			970 (3)/970 (3)	⊕⊕⊝⊝ low^d	2 studies including adults reported 1/637 participants in the insulin degludec group vs 0/315 participants in the insulin glargine group experienced a non‐fatal myocardial infarction; there were no events in 1 study including children (Urakami 2017) 2 studies including adults reported 1/637 participants in the insulin degludec group vs 0/315 in the insulin glargine group experienced cerebral ischaemia; there were no events in 1 study including children (Urakami 2017)
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring from 22:00 to 06:59 h Follow‐up: 24 ‐ 52 weeks	25 per 1000	35 per 1000 (15 to 83)	RR 1.39 (0.59 to 3.27)	970 (3)	⊕⊕⊝⊝ low^c	2 studies included adults 1 study include children
Serious adverse events (n/N) Follow‐up: 24 and 52 weeks	77 per 1000	71 per 1000 (45 to 113)	RR 0.92 (0.58 to 1.46)	970 (3)	⊕⊕⊝⊝ low^c	2 studies included adults 1 study including children reported no child experienced a serious adverse event (Urakami 2017)
HbA1c (%) Follow‐up: 24 and 52 weeks	The mean HbA1c ranged across the insulin glargine groups from 6.9% to 7.8%	The mean HbA1c in the insulin degludec groups was 0.1% higher (0% lower to 0.2% higher)	—	1388 (4)	⊕⊕⊝⊝ low^c	The 95% prediction interval ranged between ‐0.1% and 0.3% 3 studies included adults, 1 study included children (the test for subgroup differences did not indicate interaction)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; CSR: clinical study report; HbA1c: glycosylated haemoglobin A1c; MID: minimal important difference; n/N: number of people experiencing an event; OR: odds ratio; RR: risk ratio; T1DM: type 1 diabetes mellitus:
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by three levels because of indirectness (insufficient time frame) and serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 5. ^bDowngraded by three levels because of overall risk of bias ('some concerns') and serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 5. ^cDowngraded by two levels because of serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 5. ^dDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 5.

Background

Description of the condition

Onset of type 1 diabetes mellitus (T1DM) can occur at any age and accounts for about 5% to 10% of all diabetes mellitus cases (Daneman 2006). It is a metabolic disease caused by an autoimmune destruction of pancreatic β‐cells which results in a deficiency of insulin secretion. What causes the pathological autoimmune response is not yet fully understood but includes genetic susceptibility in combination with an environmental trigger (Field 1997; Maahs 2010; van der Werf 2007). The incidence of T1DM varies geographically, being highest in Northern Europe (Karvonen 1993). Over the years, a worldwide increase in incidence has been observed, the reasons for which are not yet clear (Onkamo 1999; Pitkaniemi 2004).

Description of the intervention

For people with T1DM, the goal of insulin therapy is to provide insulin that mimics physiologic insulin secretion. The most commonly used administration of insulin is by subcutaneous injection (ADA 2019). Insulin is usually applied through insulin syringes, insulin pens or insulin pumps. In order to control blood glucose levels in periods of fasting and to enable cells to incorporate glucose for production of energy, basal or background insulin is needed, which can be given by means of daily or twice‐daily injections of an intermediate‐acting or (ultra‐)long‐acting insulin preparation. Bolus insulin is taken at mealtime (prandial insulin) to control blood glucose levels following a meal and is given by means of short‐acting or rapid‐acting insulin, usually before meals (ADA 2019). With insulin pump‐based treatments, a continuous delivery of rapid‐acting insulin is administered through the pump, with the addition of mealtime insulin bolus (basal‐bolus regimen). The aim for most people with T1DM is to achieve near‐normal glycaemic levels (ADA 2019) and to allow flexibility regarding time, type and amount of food intake which can best be mastered through structured patient‐education programmes (Pillay 2015).

Since the early 1920s, people with diabetes were treated with insulin, which was purified from bovine or porcine pancreas (animal insulin). Recombinant 'human' insulin was first produced in Escherichia coli in 1978 by combining the expressed insulin A‐ and B‐chains (Chance 1993). In 1982, the first insulin utilising recombinant deoxyribonucleic acid (DNA) technology was marketed. At present, insulin is being produced predominantly in Escherichia coli and yeasts (Chance 1993).

The choice of basal insulin depends upon patient and prescriber preferences, 'lifestyle' and economic and health system considerations. Historically, intermediate‐ and long‐acting insulin preparations were obtained by crystallising either protamine (Neutral Protamin Hagedorn (NPH) type, also known as isophane insulin) or zinc (Lente type). Most insulins have a concentration of 100 units per mL (U100) but more concentrated insulin formulations (U200, U300, U500) are currently available (Heinemann 2019). Soluble human insulin consists of different oligomers (monomers, dimers and hexamers). When administered subcutaneously, insulin monomers and dimers are readily absorbed by blood capillaries. Before dissociation of hexamers into dimers and monomers, the crystalline structures need to dissolve, and this process prolongs the absorption phase and contributes to pharmacokinetic variability between injections. Hence, the rate of insulin absorption is fastest for monomers followed by dimers and hexamers, respectively (Gradel 2018). Treatment with intermediate‐acting human insulins has drawbacks: NPH is associated with a pronounced insulin peak following injection, which seems to be associated with variable absorption (Heinemann 2000; Lepore 2000) and an increased risk of hypoglycaemia (Tricco 2014).

In order to achieve the potential benefits of near‐normal glycaemic control with a reduced risk of hypoglycaemia, new insulins have been introduced to the market. In an effort to provide insulin with a more suitable physiological time course to persons with diabetes mellitus, insulin analogues have been developed. Insulin analogues are insulin‐like molecules, engineered on the basis of the molecular structure of human insulin by changing the amino acid sequence and physiochemical properties. Four main (ultra‐)long‐acting insulin analogues are currently available on the market: two long‐acting insulin analogues (insulin detemir and insulin glargine U100), and two ultra‐long‐acting insulin analogues (insulin degludec and insulin glargine U300). The glargine U300 formulation has a more extended time‐action profile than glargine U100 and is thought to achieve a more stable glycaemic control (Yale 2018).

Because the patent of insulin glargine has expired, biosimilar insulins have become available on the market. Biosimilar insulin glargine is a biological copy of the original insulin glargine which is believed to have comparable quality, efficacy and safety. Biosimilar insulin glargine is cheaper than the original insulin glargine (Soldatov 2019).

Adverse effects of the intervention

The risk of developing hypoglycaemic episodes varies among studies depending on the definition of hypoglycaemia and the desired glycaemic target (Kahler 2014). Due to a more sustainable molecule structure of insulin analogues, studies have indicated a reduced risk of severe hypoglycaemia compared with NPH insulin (Tricco 2014). However, data are conflicting (Laranjeira 2018). Targeting lower glycosylated haemoglobin A1c (HbA1c) levels is often difficult to achieve and leads to a higher incidence of hypoglycaemic events (Kahler 2014). However, targeting near‐normal glucose levels in order to avoid detrimental long‐term consequences of hyperglycaemia is currently recommended in most people with type 1 diabetes (ADA 2019).

Compared to human insulin, some insulin analogues have shown higher mitogenic potency and insulin‐growth factor binding affinity in in‐vitro and animal studies (Grant 1993; Jorgensen 1992; King 1985; Kurtzhals 2000). These effects differ depending on the insulin analogue, but results provided in these studies are unable to clarify their relevance for people with diabetes mellitus. The American and European pharmaceutical registration agencies, the Food and Drug Administration (FDA) and the European Medicines Agency (EMA), have commented on the mitogenic and carcinogenic potency of long‐acting insulin analogues and concluded that there appear to be few detrimental effects (EMA 2003; EMA 2004; EMA 2012; FDA 2000; FDA 2005). Observational studies have shown conflicting results regarding cancer risk with insulin analogues compared with human insulin (Hemkens 2009; Ruiter 2012).

The insulin analogues are usually more expensive than NPH insulin (Ewen 2019). While price differences may not be a major problem for health services in high‐income countries, they may be important in low‐ and middle‐income countries.

How the intervention might work

Based on the altered time‐action profiles of (ultra‐)long‐acting insulin analogues, a number of possible advantages in the therapy of people with T1DM have been suggested. For instance, it has been hypothesised that the longer action and the less pronounced insulin peak will enable both improved glycaemic control and reduced risk of hypoglycaemia (Tricco 2014).

Why it is important to do this review

Although their pharmacokinetic profiles appeared to indicate that (ultra‐)long‐acting insulin analogues improve the insulin therapy of people with diabetes mellitus, their superiority in a clinical setting has still to be demonstrated (Hemmingsen 2019). Systematic reviews comparing the benefits and harms of insulin analogues with NPH insulin exist, but they have methodological deficiencies due to lack of identification of all relevant studies, missing analysis of clinical study reports (CSR) and poor 'Risk of bias' assessment (Laranjeira 2018; Tricco 2014).

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs).

Types of participants

Non‐pregnant people withT1DM.

Types of interventions

We planned to investigate the following comparisons of intervention versus comparator.

Intervention

Long‐acting insulin analogues (insulin glargine U100 or insulin detemir) and their biosimilar insulins.
Ultra‐long‐acting insulin analogues (insulin glargine U300 or insulin degludec).

Comparisons

Long‐acting insulin analogue or its biosimilar insulin versus human NPH insulin.
Ultra‐long‐acting insulin analogue or its biosimilar insulin versus human NPH insulin.
(Ultra‐)long‐acting insulin analogue versus another (ultra‐)long‐acting insulin analogue.

Concomitant interventions had to be the same in both the intervention and comparator groups to establish fair comparisons.

Only studies reporting on subcutaneously administered insulin were be considered for inclusion in this review.

If a study included multiple arms, we included any arm that met our inclusion criteria.

Minimum duration of intervention

We included studies with a minimum duration of 24 weeks. In the case of a cross‐over RCT, each intervention period had to be at least 24 weeks.

Minimum duration of follow‐up

Minimum duration of follow‐up was 24 weeks. In the case of a cross‐over RCT, duration of follow‐up for each intervention period had to be at least 24 weeks.

We defined any follow‐up period going beyond the original time frame for the primary outcome measure as specified in the power calculation of the study's protocol as an extended follow‐up period (also called 'open‐label extension study') (Buch 2011; Megan 2012).

Types of outcome measures

We did not exclude a study if it failed to report one or several of our primary or secondary outcome measures. If none of our primary or secondary outcomes was reported in the study, we did not include the study but provided some basic information in the 'Characteristics of studies awaiting classification' table.

We investigated the following outcomes using the methods and time points specified below.

Primary outcomes

All‐cause mortality.
Health‐related quality of life.
Severe hypoglycaemia.

Secondary outcomes

Cardiovascular mortality.
Non‐fatal myocardial infarction.
Non‐fatal stroke.
End‐stage renal disease.
Blindness.
Serious adverse events.
Diabetic ketoacidosis.
Non‐serious adverse events.
Nocturnal hypoglycaemia.
Mild/moderate hypoglycaemia.
Socioeconomic effects.
HbA1c levels.
Combined HbA1c levels and severe hypoglycaemia.

Method of outcome measurement

All‐cause mortality: defined as death from any cause.
Health‐related quality of life: defined as mental and physical health‐related quality of life and evaluated by a validated instrument such as Short‐Form‐36 (SF‐36). Scales focusing on treatment satisfaction and not health‐related quality of life as main outcome were not included.
Severe hypoglycaemia: requiring assistance from another person (was planned to be further categorised into 'assistance from other persons', assistance from medical staff, intravenous glucose administration, subcutaneous glucagon administration, hospitalisation, intensive‐care unit stay, coma).
Cardiovascular mortality, non‐fatal myocardial infarction, non‐fatal stroke, blindness: defined as reported in studies.
End‐stage renal disease: defined as need for dialysis and renal transplantation.
Serious adverse events (SAE): defined according to the International Conference on Harmonization (ICH) guidelines as, "any event that leads to death, that is life‐threatening, required in‐patient hospitalisation or prolongation of existing hospitalisation, resulted in persistent or significant disability, and any important medical event which may have had jeopardised the patient or required intervention to prevent it" (ICH 1997) or as reported in studies.
Diabetic ketoacidosis: potentially life‐threatening condition with high levels of ketones in the body which when building up in the blood make the blood more acidic.
Non‐serious adverse events: all adverse events, not classified as SAEs.
Nocturnal hypoglycaemia: hypoglycaemia during night‐time and defined as reported in studies.
Mild/moderate hypoglycaemia: hypoglycaemic episodes not requiring assistance from another person.
Socioeconomic effects: such as direct costs defined as admission or readmission rates; average length of stay; visits to general practitioner; accident or emergency visits; medication consumption; indirect costs defined as resources lost due to illness by the participant or their family member.
HbA1c levels: expressed as percentage or mmol/mol.
Combined HbA1c levels and severe hypoglycaemia: joint examination of the effects of HbA1c reduction and hypoglycaemia risk.

Timing of outcome measurement

For all outcome measures, we defined short‐term follow‐up as 24 weeks to ≤ 52 weeks, medium‐term follow‐up as > 1 year to ≤ 2 years and long‐term follow‐up as > 2 years.

Search methods for identification of studies

Electronic searches

We searched the following sources from the inception of each database to the date of search and did not place restrictions on the language of publication:

Cochrane Central Register of Controlled Trials (CENTRAL) via the Cochrane Register of Studies Online (CRSO) (searched 24 August 2020);
MEDLINE (Ovid MEDLINE ALL 1946 to Daily Update) (searched 24 August 2020);
ClinicalTrials.gov (www.clinicaltrials.gov) (searched 24 August 2020);
World Health Organization International Clinical Trials Registry Platform (ICTRP) (www.who.int/trialsearch) (searched 24 August 2020);
HTA database (https://database.inahta.org/) (searched 24 August 2020).

We did not include Embase in our search, as RCTs indexed in Embase are now prospectively added to CENTRAL via a highly sensitive screening process (Cochrane 2020).

For detailed search strategies, see Appendix 6.

Searching other resources

We identified other potentially eligible studies or ancillary publications by searching the reference lists of included studies, systematic reviews, meta‐analyses, and health technology assessment reports. In addition, we contacted the investigators of included studies to obtain additional information on the retrieved studies and establish whether we may have missed further studies.

We searched the grey literature, which we defined as searching the HTA database, as well as databases from regulatory agencies (European Medicines Agency (EMA) and Food and Drug Administration (FDA) ‐ Hart 2012;Schroll 2015). We searched for CSRs and clinical study synopses as provided on manufacturers' web sites (e.g. Novo Nordisk Trials) and via contact with manufacturers (Appendix 7).

We did not use abstracts or conference proceedings for data extraction unless full data were available from study authors because this information source does not fulfil the CONSORT requirements which consist of "an evidence‐based, minimum set of recommendations for reporting randomized trials" (CONSORT 2018; Scherer 2018). We presented information on abstracts or conference proceedings in the 'Characteristics of studies awaiting classification' table (Characteristics of studies awaiting classification).

Data collection and analysis

Selection of studies

We used Cochrane’s Screen4Me workflow to help assess the search results. Screen4Me comprises three components: known assessments – a service that matches records in the search results to records that have already been screened in Cochrane Crowd and been labelled as an 'RCT' or as 'Not an RCT'; the RCT classifier – a machine learning model that distinguishes RCTs from non‐RCTs, and, if appropriate, Cochrane Crowd – Cochrane’s citizen science platform where the Crowd helped to identify and describe health evidence. Detailed information regarding evaluations of the Screen4Me components can be found in the following publications: Marshall 2018; McDonald 2017; Noel‐Storr 2018; Thomas 2017.

Two review authors (BH, BR) independently screened the abstract, title, or both, of all records remaining after the Screen4Me workflow, to determine which studies we should assess further. We obtained the full text of all potentially relevant records. We would have resolved disagreements through consensus or by recourse to a third review author (MIM), if these had occurred. In case we were unable to resolve a disagreement, we planned to categorise the study as a 'Study awaiting classification' and would have contacted the study authors for clarification. We presented an adapted PRISMA flow diagram to show the process of study selection (Liberati 2009). We listed all articles excluded after full‐text assessment in a 'Characteristics of excluded studies' table and provided the reasons for exclusion (Characteristics of excluded studies).

Data extraction and management

For studies that fulfilled our inclusion criteria, two review authors (BH, BR) independently extracted key participant and intervention characteristics. We described interventions according to an adapted version of the 'template for intervention description and replication' (TIDieR) checklist (Hoffmann 2014; Hoffmann 2017).

We reported data on efficacy outcomes and adverse events using standardised data extraction sheets from the CMED Group. We resolved disagreements by discussion or, if required, by consultation with a third review author (MIM).

We provided information including trial identifier for potentially relevant ongoing trials in the 'Characteristics of ongoing studies' table and in a joint appendix 'Matrix of study endpoints (publications and trial documents)'. We tried to find the protocol and CSR for each included study.

We planned to email all authors of included studies, ongoing trials and studies awaiting classification to enquire whether they would be willing to answer questions regarding their studies. We presented the results of this survey in an appendix. We thereafter sought relevant missing information on the study from the primary study author(s), if required.

Dealing with duplicate and companion publications

In the event of duplicate publications, companion documents, or multiple reports of a primary study, we maximised the information yielded by collating all available data, and we used the most complete data set aggregated across all known publications and records. We listed duplicate publications, companion documents, multiple reports of a primary study, and trial documents of included studies (such as trial registry information and CSRs) as secondary references under the study ID of the included study. Furthermore, we listed duplicate publications, companion documents, multiple reports of a study, and trial documents of excluded studies (such as trial registry information) as secondary references under the study ID of the excluded study.

Data from clinical trials registers and CSR

If data from included studies were available as study results in clinical trials registers, such as ClinicalTrials.gov or as CSR, we made full use of this information and extracted the data. If there also was a full publication of the study, we collated and critically appraised all available data. If an included study was marked as a completed trial in a clinical trials register but no additional information (study results, publication, or both) was available, we added this study to the 'Characteristics of studies awaiting classification' table.

Assessment of risk of bias in included studies

Two review authors (BH, BR) independently assessed the risk of bias for each included study. We would have resolved disagreements by consensus or by consulting a third review author (MIM), if such occurred. If adequate information was unavailable from the publications, trial protocols, CSRs or other sources, we contacted the study authors for more details to request missing data on 'Risk of bias' items.

We undertook ‘Risk of bias’ assessment according to Chapter 7 and Chapter 8 of the CochraneHandbook for Systematic Reviews of Interventions (Boutron 2020; Higgins 2020). We used the Cochrane 'Risk of bias 2' (RoB 2) tool (version 22, August 2019) ‐ (Higgins 2017; Sterne 2019).

We focused on the assessment of the effect of assignment to the interventions at baseline. The effect was analysed as the result of a comparison between interventions on a certain outcome at a specific time point. The RoB 2 tool evaluates the following domains.

Bias arising from the randomisation process.
Bias due to deviations from the intended interventions.
Bias due to missing outcome data.
Bias in measurement of the outcome.
Bias in selection of the reported results.

Within each domain, signalling questions provided information about features of the study that were relevant to risk of bias. Possible answers to the signalling questions were 'Yes', 'Probably yes', 'Probably no', 'No' and 'No information'. After answering the signalling questions, we made a 'Risk of bias' judgement, assigning one of three levels ('low risk of bias', 'some concerns', 'high risk of bias') to each domain.
For each specific outcome, we established an overall 'Risk of bias' judgement using the following criteria.

Low risk of bias: the study was judged to be at low risk of bias for all domains for this result.
Some concerns: the study was judged to raise some concern in at least one domain for this result, but not to be at high risk of bias for any domain.
High risk of bias: the study was either judged to be at high risk of bias in at least one domain for this result, or the study was judged to have some concerns for multiple domains in a way that substantially lowers confidence in the result.

We distinguished between participant‐reported outcomes, observer‐reported outcomes not involving judgement, observer‐reported outcomes involving some judgement, outcomes reflecting decisions made by interventions providers and composite outcomes.

Participant‐reported outcomes: health‐related quality of life; mild/moderate and non‐severe nocturnal hypoglycaemia; non‐serious adverse events; socioeconomic effects.
Observer‐reported outcomes not involving judgement: all‐cause mortality, end‐stage renal disease, blindness, HbA1c levels.
Observer‐reported outcomes involving some judgement: cardiovascular mortality, non‐fatal myocardial infarction, non‐fatal stroke, socioeconomic effects.
Outcomes reflecting decisions made by interventions providers: SAEs, severe hypoglycaemia, severe nocturnal hypoglycaemia.
Composite outcomes: combined HbA1c levels and severe hypoglycaemia.

Measures of treatment effect

When at least two included studies were available for a comparison of a given outcome, we expressed dichotomous data as a risk ratio (RR) or an odds ratio (OR) with 95% confidence intervals (CI). For continuous outcomes measured on the same scale (e.g. HbA1c in %), we estimated the intervention effect using the mean difference (MD) with 95% CIs. For continuous outcomes that measured the same underlying concept (e.g. health‐related quality of life) but used different measurement scales, we would have calculated the standardised mean difference (SMD). We would have expressed time‐to‐event data as a hazard ratio (HR) with 95% CIs.

Unit of analysis issues

We took into account the level at which randomisation occurred, such as cross‐over studies, cluster‐randomised studies, and multiple observations for the same outcome. If more than one comparison from the same study had been eligible for inclusion in the same meta‐analysis, we would either have combined groups to create a single pair wise comparison, or we would appropriately reduce the sample size so that the same participants had not contributed data to the meta‐analysis more than once (splitting the 'shared' group into two or more groups). Although the latter approach offers some solution for adjusting the precision of the comparison, it does not account for correlation arising from inclusion of the same set of participants in multiple comparisons (Higgins 2011).

We would have re‐analysed cluster‐RCTs that had not appropriately adjusted for potential clustering of participants within clusters in their analyses. Variance of the intervention effects would have been inflated by a design effect. Calculation of a design effect involves estimation of an intracluster correlation coefficient (ICC). We would have obtained estimates of ICCs by contacting study authors, or by imputing ICC values using either estimates from other included studies that reported ICCs or external estimates from empirical research (e.g. Bell 2013). We would have examined the impact of clustering by performing sensitivity analyses.

Dealing with missing data

If possible, we obtained missing data from the authors of included studies. We carefully evaluated important numerical data such as screened, randomly assigned participants, as well as intention‐to‐treat and as‐treated and per‐protocol populations. We investigated attrition rates (e.g. dropouts, losses to follow‐up, withdrawals), and we critically appraised issues concerning missing data and use of imputation methods (e.g. last‐observation‐carried‐forward).

If studies were identified in which the standard deviation (SD) of the outcome was not available at follow‐up or we could not recreate it, we would have standardised by the mean of the pooled baseline SD from studies that reported this information.

If we had identified included studies not reporting means and SDs for outcomes, and we could not receive the requested information from study authors, we would have imputed these values by estimating the mean and the variance from the median, the range and the size of the sample (Hozo 2005).

We would have investigated the impact of imputation on meta‐analyses by performing sensitivity analyses, and we would have reported for every outcome which studies had imputed SDs.

Assessment of heterogeneity

In the event of clinical or methodological heterogeneity, we planned not to report study results as the pooled effect estimate in a meta‐analysis.

We identified heterogeneity (inconsistency) by visually inspecting the forest plots and by using a standard Chi² test with a significance level of α = 0.1 (Deeks 2017). In view of the low power of this test, we also considered the I² statistic, which quantifies inconsistency across studies, to assess the impact of heterogeneity on the meta‐analysis (Higgins 2002; Higgins 2003).

When we found heterogeneity, we planned to determine possible reasons for this by examining individual study and subgroup characteristics. If possible, we calculated prediction intervals to elucidate the clinical implication of the observed heterogeneity (for details see Data synthesis).

Assessment of reporting biases

If we had included 10 or more studies that investigated a particular outcome, we would have used funnel plots to assess small‐study effects. Several explanations may account for funnel plot asymmetry, including true heterogeneity of effect with respect to study size, poor methodological design (and hence bias of small studies), and publication bias (Sterne 2017). Therefore, we would have interpreted the results carefully (Sterne 2011).

Data synthesis

We undertook (or displayed) a meta‐analysis only if we judged participants, interventions, comparisons, and outcomes to be sufficiently similar to ensure an answer that was clinically meaningful. Unless good evidence showed homogeneous effects across studies of different methodological quality, we would have primarily summarised low risk of bias data using a random‐effects model (Wood 2008). We interpreted random‐effects meta‐analyses with due consideration for the whole distribution of effects and presented a prediction interval (Borenstein 2017a; Borenstein 2017b; Higgins 2009). A prediction interval requires at least three studies to be calculated and specifies a predicted range for the true treatment effect in an individual study (Riley 2011). For rare events such as event rates below 1%, we used the Peto odds ratio method, provided there was no substantial imbalance between intervention and comparator group sizes, and intervention effects were not exceptionally large. In addition, we performed statistical analyses according to the statistical guidelines presented in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2017).

Subgroup analysis and investigation of heterogeneity

We expected the following characteristics to introduce clinical heterogeneity, and we planned to carry out the following subgroup analyses including investigation of interactions (Altman 2003).

Head‐to‐head comparisons of insulin analogues.
Studies designed to blind participants and investigators versus open‐label studies.
NPH once daily versus NPH two‐ or three‐times daily.
Studies of long duration (more than two years) versus studies of short to medium duration (two years or less).
Studies performed in high‐income countries versus middle‐income countries versus low‐income countries.
According to healthcare setting.

Sensitivity analysis

We planned to perform sensitivity analyses to explore the influence of the following factors (when applicable) on effect sizes by restricting analysis to the following.

Published studies.
Effect of risk of bias, as specified in the Assessment of risk of bias in included studies section.
Very long or large studies to establish the extent to which they dominated the results.
Use of the following filters: diagnostic criteria, imputation, language of publication, source of funding (industry versus other), or country.

We tested the robustness of results by repeating analyses using different measures of effect size (i.e. RR, OR, etc.) and different statistical models (fixed‐effect and random‐effects models).

Summary of findings and assessment of the certainty of the evidence

Certainty of the evidence

We presented the overall certainty of the evidence for each outcome specified below, according to the GRADE approach, which takes into account issues related not only to internal validity (risk of bias, inconsistency, imprecision, publication bias), but also to external validity, such as directness of results. Two review authors (BH, BR) independently rated the certainty of the evidence for each outcome. If differences in assessment had occurred, they would have been solved by discussion or by consultation with a third review author (MIM).

We included an appendix entitled 'Checklist to aid consistency and reproducibility of GRADE assessments', to help with standardisation of the 'Summary of findings' tables (Meader 2014). Alternatively, we would have used the GRADEpro Guideline Development Tool (GDT) software and presented evidence profile tables as an appendix (GRADEproGDT 2015). We presented results for outcomes as described in the Types of outcome measures section. If meta‐analysis was not possible, we presented the results in a narrative format in the 'Summary of findings' table. We justified all decisions to downgrade the certainty of the evidence by using footnotes, and we made comments to aid the reader's understanding of the Cochrane Review when necessary.

'Summary of findings' table

We presented a summary of the evidence in a 'Summary of findings' table. This provided key information about the best estimate of the magnitude of effect, in relative terms and as absolute differences for each relevant comparison of alternative management strategies, numbers of participants and studies addressing each important outcome, and a rating of overall confidence in effect estimates for each outcome.

In the 'Summary of findings' table, we reported on the 'intervention' (ultra‐)long‐acting insulin analogue or its biosimilar insulin versus the 'comparator' human NPH insulin or another (ultra‐)long‐acting insulin analogue.

We created the 'Summary of findings' table using the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2017), along with Review Manager (RevMan 5.3) table editor (RevMan 2014). We reported the following outcomes, listed according to priority.

All‐cause mortality.
Health‐related quality of life.
Severe hypoglycaemia.
Non‐fatal myocardial infarction/stroke.
Severe nocturnal hypoglycaemia.
SAEs.
HbA1c levels.

Results

Description of studies

For a detailed description of studies, see Table 1, Characteristics of included studies, Characteristics of excluded studies and Characteristics of studies awaiting classification tables.

Results of the search

The initial search identified a total of 7747 records. In assessing the studies, we used Cochrane’s Screen4Me workflow to help identify potential reports of randomised studies. The results of the Screen4Me assessment process can be seen in Figure 1. Subsequently, we assessed the remaining 3265 records, as well as the 570 records retrieved by the update search prior to publication. We excluded most of the references on the basis of their titles and abstracts because they clearly did not meet the inclusion criteria. We evaluated a further 47 records identified as CSRs, clinical study synopses, a study protocol and one additional record identified through handsearching of reference lists of included studies (Figure 2).

Figure 1

Screen4Me: Cochrane´s screening service.

Figure 2

Study flow diagram

CSR: clinical study report; EMA: European Medicines Agency; HTA: health technology assessment; Screen4Me: Cochrane's screening service.

Searching the web pages of Novo Nordisk and Sanofi, we identified 23 CSRs, clinical study synopses or both. On request, we received 10 CSRs from Sanofi and six CSRs, sections of two CSRs and one study protocol from Novo Nordisk, respectively. The two studies with sections of CSRs only were Japanese studies (Kobayashi 2007; NCT00605137). For both studies, clinical study synopses were available and we could not get full access to the Japanese versions of the CSRs. For one study, a trial protocol was provided by Novo Nordisk (NCT00605137). One study had a clinical study synopsis only (NCT00595374). The total number of additional references from web pages and contact with manufacturers was 22 CSRs, 23 clinical study synopses and one study protocol.

We identified applications/documents through searching FDA and EMA web sites (EMA 2014; EMA 2015; EMA 2015a; EMA 2015b; FDA 2000; FDA 2002; FDA 2005; FDA 2015). These references did not provide information about additional studies.

In summary, after screening the full texts from the electronic search and additional sources, we identified 26 RCTs published in 202 records that met our inclusion criteria. Two studies were unpublished, but clinical study synopses and parts of the CSRs were obtained and provided data for inclusion (NCT00595374; NCT00605137). The remaining included studies were published. For all studies, except two, it was possible to retrieve additional information from clinical trials registers, documents from regulatory agencies, CSRs, clinical study synopses and investigators (Bolli 2009; Porcellati 2004). The number of records per included studies varied from 1 to 21. Thirteen studies are awaiting assessment.

Included studies

A detailed description of the characteristics of included studies is presented elsewhere (see Characteristics of included studies and Appendix 7; Appendix 8; Appendix 9; Appendix 10; Appendix 11; Appendix 12; Appendix 13; Appendix 14; Appendix 15; Appendix 16; Appendix 17; Appendix 18; Appendix 19; Appendix 20; Appendix 21; Appendix 1; Appendix 2; Appendix 3; Appendix 4; Appendix 5. The following is a succinct overview.

Overview of study populations

Twenty‐five studies reported the total number of participants screened (Bartley 2008; BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; BEGIN Young; Bolli 2009; Chase 2008; Davies 2014; Fulcher 2005; Heller 2009; Home 2005; Kobayashi 2007; Liu 2016; NCT00595374; NCT00605137; Pieber 2007; Porcellati 2004; PRESCHOOL; Ratner 2000; Robertson 2007; Russell‐Jones 2004; Schober 2002; Standl 2004; SWITCH 1; Thalange 2013; Vague 2003).

A total of 8784 participants were randomised: 2428 participants were randomised to NPH insulin, 2889 participants to insulin detemir, 2095 participants to insulin glargine and 1372 participants to insulin degludec (see Table 1). Eight of the studies included children and randomised 1835 participants, i.e. 21% of all participants (BEGIN Young; Chase 2008; Liu 2016; NCT00605137; Robertson 2007; Schober 2002; Thalange 2013; Urakami 2017). The remaining studies included adults.

See: Summary of findings 1 Summary of findings: insulin detemir versus NPH insulin; Summary of findings 2 Summary of findings: insulin glargine versus NPH insulin; Summary of findings 3 Summary of findings: insulin detemir versus insulin glargine; Summary of findings 4 Summary of findings: insulin degludec versus insulin detemir; Summary of findings 5 Summary of findings: insulin degludec versus insulin glargine

Table 1. Overview of study populations

Study ID (study design)	Intervention(s) and comparator(s)	Description of power and sample size calculation	Screened/eligible (n)	Randomised (n)	Analysed primary outcome) (n)	Finishing study (n)	Randomised finishing study (%)	Follow‐up (extended follow‐up)^a
Bartley 2008 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "A total of 489 patients were needed to obtain 245 evaluable patients on detemir and 123 on NPH to detect a clinically relevant difference of 0.4% in HbA1c with a power of 85%, assuming a standard deviation (SD) for HbA1c of 1.2 and an expected drop‐out rate of 25%"	557	331	320	278	84.3	24 months
	C: NPH insulin		557	166	159	144	86.7
	total:			497	479	422	85.0
BEGIN Basal‐Bolus Type 1^b (parallel‐group non‐inferiority RCT)	I: insulin degludec	Quote: "Sample size was determined by the primary objective with the assumption of a one sided t test at a significance level of 2·5%, a zero mean treatment difference, and an SD of 1·1% for HbA1c. A total of 624 participants were needed for at least 95% power after adjustment for a 15% dropout rate"	722	472	472	404	85.6	52 weeks (104 weeks)
	C: insulin glargine		722	157	157	137	87.0
	total:			629	629	541	86.0
BEGIN Flex T1^c (parallel‐group non‐inferiority RCT)	I: insulin degludec	Quote: "Sample size was determined on the basis of the primary objective under the assumption of a 1‐sided t test of size 2.5%, a zero mean treatment difference, and standard deviation of 1.1% for HbA1c"	549	165	165	139	84.2	26 weeks (52 weeks)
	C: insulin glargine		549	164	164	152	92.7
	total:			329	329	291	88.4
BEGIN Young^d (parallel‐group non‐inferiority RCT)	I: insulin degludec	Quote: "The sample size was determined using a t‐statistic under the assumption of a one‐sided test of size 2.5%, a zero mean treatment difference and standard deviation (SD) of 1.25% for HbA1c. A total of 346 participants had to be randomized to achieve at least 80% or greater power in the evaluation of the per protocol (PP) analysis set, after adjustment for a 10% dropout rate"	363	174	174	170	97.7	26 weeks (52 weeks)
	C: insulin detemir		363	176	176	163	93.7
	total:			350	350	333	95.1
Bolli 2009 (parallel‐group superiority RCT)	I: insulin glargine	Quote: "The expected FBG difference in the two groups at the end of the study treatment was estimated to be 30+/‐60 mg/dL. Using a two‐sided test with ɑ = 0.01 and ß = 0.1 (i.e., power: 1‐ß = 0.9), 240 evaluable patients were to be included. Due to an expected dropout rate of 20% and to the randomization schedule, which was restricted and stratified by centre (26 centres), 312 patients were planned to be enrolled"	213	85	85	78	91.8	24 weeks (30 weeks)
	C: NPH insulin		213	90	90	74	82.2
	total:			175	175	152	86.7
Chase 2008 (parallel‐group non‐inferiority RCT)	I: insulin glargine	Quote: "The primary clinical outcome (the mean change in A1C from baseline [week 0] to endpoint [week 24 or last post randomization assessment]) was compared in the 2 treatment groups using analysis of covariance (ANCOVA), with treatment group, study centre (pooled), CGMS values, sex, and baseline value as covariates (α = 0.05; 2‐sided test). The 95% confidence intervals (CIs) were computed for the adjusted mean difference between treatment groups from the ANCOVA to test for noninferiority (defined as an upper bound of the 95% CI for the mean difference in A1C of ≤ 0.4%)"	235	85	84	81	95.3	24 weeks (25 weeks)
	C: NPH insulin/Lente		235	90	84	76	84.4
	total:			175	168	157	89.7
Davies 2014^e (parallel‐group non‐inferiority RCT)	I: insulin degludec	Quote: "Assuming a standard deviation (SD) of 1.1% for the primary endpoint, the trial had 90% power with 360 participants randomized 2:1"	512	303	302	283	93.4	26 weeks (52 weeks)
	C: insulin detemir		512	153	153	138	90.2
	total:			456	455	421	92.5
Fulcher 2005 (parallel‐group non‐inferiority RCT)	I: insulin glargine	Quote: "The sample size was calculated assuming a 20% dropout rate, so that 118 patients (59 in each group) were enrolled in order to have 96 patients (48 in each group) available for evaluation at end‐point. Assuming a SD of 1.2 for HbA1c (based on previous Phase IIIa studies), the study had 80% power to detect a 0.7% difference in HbA1c"	173^f	62	62	58	94	30 weeks
	C: NPH insulin		173^f	63	62	49	78
	total:			125	124^f	107	85.6
Heller 2009 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The sample size was determined for 2:1 (detemir:glargine) randomization and based on a 1‐sided t test at a 2.5% significance level. Assuming an SD of 1.0% for HbA1c and a dropout rate of 15%, a sample size of 435 patients gave 95% power to demonstrate noninferiority"	515	300	299	263	87.7	52 weeks
	C: insulin glargine		515	147	144	122	83.0
	total:			447	443	385	86.1
Home 2005 (parallel‐group superiority RCT)	I: insulin glargine	Quote from CSR: "It was planned to treat 520 subjects, 260 subjects in each group. Each investigation site was to randomise 10‐20 subjects.The primary efficacy variable for the comparison between HOE 901 and NPH insulin was the change from baseline in GHb at the study endpoint for the individual subject ... The standard deviation for change from baseline in GHb at endpoint was estimated to be 1.6%. Based on 1:1 randomization and using a t‐test, a total number of 440 subjects (220 subjects for each group) was required to detect a mean difference of 0.5% GHb between HOE 901 and NPH with a type I error of α = 5% and a statistical power of 90%. With an expected drop‐out rate of 15% during the course of the study, a total number of 520 subjects (260 subjects in each group) were to be enrolled in order to have 440 subjects (220 subjects in each group) evaluable at week 28"	655	298	292	276	94.5	28 weeks
	C: NPH insulin		655	305	293	272	92.8
	total:			603^g	585^h	548	93.6
Kobayashi 2007 (parallel‐group non‐inferiority RCT)	I: insulin detemir	—	454ⁱ	197	195	183	93.4	48 weeks
	C: NPH insulin	—	454ⁱ	99	98	91	92.9
	total:			296	293	274	92.6
Liu 2016 (parallel‐group non‐inferiority RCT)	I: insulin glargine	Quote from CSR: "The planned sample size was reduced from 366 to 150 patients in view of extremely difficult recruitment progress over the 2 years since first patient’s enrolment..."	196	107	108	106	99.1	24 weeks (25 weeks)
	C: NPH insulin		196	55	54	50	90.9
	total:			162	161	156	96.3
NCT00595374^f (parallel‐group non‐inferiority RCT)	I: insulin detemir	—	124	75	—	70	93.3	26 weeks
	C: NPH insulin	—	124	38	—	34	92.1
	total:			113	—	104	92.0
NCT00605137^f (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote from trial protocol: "This power calculation is based on a two‐sample poisson test at a significance level of 5% for the comparison of the mean rate of nocturnal episodes per four weeks although nocturnal episodes will be analysed as recurrent events using gamma frailty model in the trial analysis"	88	57	55	55	96.5	24 weeks
	C: NPH insulin		88	29	27	27	93.1
	total:			86	82	82	95.3
Pieber 2007 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The sample size was determined in order to test non‐inferiority in a 1:1 randomization. Assuming a standard deviation for HbA1c of 1.2% and a clinically relevant, absolute difference in HbA1c of 0.4%, a total of 286 randomized participants were needed to achieve a power of 80%. Assuming a 10% drop‐out rate, 159 randomized participants were needed in each group"	415	161	161	147	91.3	26 weeks
	C: insulin glargine		415	161	159	146	90.7
	total:			322	319	293	91.0
Porcellati 2004 (parallel‐group superiority RCT)	I: insulin glargine	Quote: "In this design, a total of 120 participants were required to achieve 90% power to detect a difference of 0.3% among the means with group standard deviations of 0.4 at the significance level (alpha) of 5%"	130	61	61	61	100	1 year
	C: NPH insulin		130	60	60	60	100
	total:			121	121	121	100
PRESCHOOL (parallel‐group non‐inferiority RCT)	I: insulin glargine	Quote: "Sample size calculation was based on an expected composite hypoglycemia rate of 0.8 events/100 patient‐yr of exposure to insulin glargine or to NPH insulin. The sample size and novel composite outcome was planned to ensure sufficient power so that the upper bound of the two‐sided 95% confidence interval (CI) for the insulin glargine:NPH ratio of the mean composite hypoglycemia rates for the comparison of treatment groups would not exceed 1.15. A sample size of 35 completed patients per treatment group was to provide 96% power to demonstrate noninferiority of insulin glargine vs. NPH"	165	61	61	57	93.4	24 weeks (26 weeks)
	C: NPH insulin		165	64	64	54	84.4
	total:			125^j	125	111	88.8
Ratner 2000 (parallel‐group superiority RCT)	I: insulin glargine	Quote: "An estimated 440 participants (220 in each treatment group) were required to detect a mean difference of 0.5% in GHb levels between treatment with a type 1 error of α = 5% and a statistical power of 90%"	677^f	266	256	233	88.3	28 weeks
	C: NPH insulin		677^f	274	262	248	91.9
	total:			540^k	518	481^f	90.1
Robertson 2007 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The sample size was determined for a 2: 1 randomization based on a two‐sided t ‐test on a 5% significance level. Assuming a SD for HbA 1c of 1.1% and a clinically relevant difference in HbA 1c of 0.4% (absolute), 270 children were needed to achieve a power of 80%. With an expected drop‐out rate of 20%, 338 children were to be allocated to study treatment"	363^f	232	232	226	97.4	26 weeks
	C: NPH insulin		363^f	115	114	109	94.8
	total:			347	347	335^f	96.5
Russell‐Jones 2004 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "Sample size was based on an SD for HbA1c of 1.4% and the assumption that a 0.4% absolute difference in HbA1c represents a clinically relevant difference" and " All comparisons were 2‐tailed tests with a 5% level of significance"	838^f	492	491	465	94.7	6 months
	C: NPH insulin		838^f	257	256	235	91.8
	total:			749^f	747	700	93.5
Schober 2002 (parallel‐group superiority RCT)	I: insulin glargine	Quote: "The sample size was calculated to detect a mean difference in HbA1C from baseline to endpoint of 0.5% with a statistical power of 90%. Assuming a 20% dropout rate, the minimum sample size required was 360 patients"	385	180	155	169	93.9	28 weeks
	C: NPH insulin		385	181	156	168	92.8
	total:			361^l	311	337^f	93.4
Standl 2004^m (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote from CSR: "A total of 440 type 1 participants were planned for randomisation in order to obtain 400 evaluable participants, assuming a dropout rate of approximately 10%"	505^f	237	210	212	89.5	6 months (12 months)
	C: NPH insulin		505^f	224	206	209	93.3
	total:			461^f	416^f	421	91.3
SWITCH 1ⁿ (cross‐over non‐inferiority RCT)	I: insulin degludec	Quote: "The trial was powered to show noninferiority of the primary end point. Based on the assumption that up to 10% of the randomised patients may not contribute to the analysis, 400 patients needed to contribute to the analysis if 446 patients were randomised to ensure a power of 94%, to demonstrate noninferiority with an expected rate of overall symptomatic hypoglycemia of 5.0 episodes per patient‐years’ exposure (PYE)"	634	249	249	209	83.9	32 weeks
	C: insulin glargine		634	252	251	205	81.3
	total:			501	414	414	82.6
Thalange 2013 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The power calculation was analysed on this basis: using a two‐sided t‐test with a one‐sided significance level of 2.5%, assuming SD of 1.1, a non‐inferiority criterion of 0.4%, a power of 85% and an expected dropout rate of 20%, a total of 344 children were to be randomized"	381	177	171	164	92.7	52 weeks (104 weeks)
	C: NPH insulin		381	171	168	161	94.2
	total:			348	339	325	93.4
Urakami 2017^p (cross‐over superiority RCT)	I: insulin degludec	—	—	9	9	9	100	24 weeks
	C: insulin glargine	—	—	9	9	9	100
	total:			18	18	18	100
Vague 2003 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The initial cohort size was calculated to achieve a power of 85% on the basis of non‐inferiority testing at the 5% significance level and a 2:1 randomization"	471^f	301	280	284	94.4	6 months (12 months)
	C: NPH insulin		471^f	147	139	141	96.6
	total:			448	419	425	95.1
Overall total	All insulin detemir			2889		2648
	All insulin degludec			1372		1214
	All insulin glargine			2095		1890
	All NPH insulin			2428		2202
	All interventions and comparators			8784		7954

— denotes not reported

^aFollow‐up under randomised conditions until end of study (= duration of intervention + follow‐up post‐intervention or identical to duration of intervention); extended follow‐up refers to follow‐up of participants once the original study was terminated as specified in the power calculation.
^bData in the table are for the main period.After 52 weeks, the participants of the initial study were invited to an extension study. 74% in the degludec and 75% in the glargine participated. Of the one included in the extension period, 94% (330/351) participants completed in the degludec group and 96% (113/118) participants in the glargine group.
^cAn additional study arm existed, which was not included in this review.
^dData in the table are for the main period. In the insulin degludec group, 152 participants entered the extension study and 151 participants completed; in the insulin detemir group, 128 participants entered the extension study and 122 participants completed.
^eData in the table are for the main period. In the insulin degludec group, 248 participants entered the extension study and 242 participants completed (79.9% of those initially randomised); in the insulin detemir group, 122 participants entered the extension study and 115 participants completed (75.2% of those initially randomised).
^fData from clinical study report/synopsis.
^gIn the publication, it was only mentioned that 602 participants were randomised, but not explained how these were divided between the intervention groups. This was reported in the clinical study report. In the publication, there was only information about the allocation of the 585 participants who received the intervention.
^hIn the main publication, the number of participants analysed was not clearly described; this number was provided by the clinical study report.
ⁱBoth people with type 1 diabetes mellitus and type 2 diabetes mellitus were screened.
^jOne participant randomised to NPH insulin was actually treated with insulin glargine, thus the safety population comprised 62 participants for insulin glargine and 63 participants for NPH insulin.
^kIn the main publication, it was stated that 534 participants were randomised (264 participants allocated to insulin glargine; 270 participants allocated to NPH insulin). In the clinical study report, it was stated that a total of 540 participants were randomised, but six were never treated (2 participants in the insulin glargine group; 4 participants in the NPH insulin group).
^lOf the 361 participants randomised, 12 withdrew their consent before being treated, therefore a total 349 participants were treated: 174 participants in the glargine group compared with 175 participants in the NPH group.
^mData in the table are for the main period. In the insulin detemir group, 154 participants entered the extension study and 118 participants completed (49.8% of those initially randomised); in the NPH insulin group, 135 participants entered the extension study and 134 participants completed (59.8% of those initially randomised).
ⁿData from first treatment period before cross‐over (32 weeks).
^oExtension only performed for the detemir group.
^pNot reported if any participant dropped out during the study. All randomised participants were included in all analyses.

A1c: glycosylated haemoglobin A1c
ANCOVA: analysis of covariance
C: comparator
CGMS: continuous glucose monitoring system
CI: confidence interval
CSR: clinical study report
FBG: fasting blood glucose
GHb: glycated haemoglobin
HbA1c: glycosylated haemoglobin A1c
HOE 901: insulin glargine
I: intervention
NPH: neutral protamine Hagedorn
PYE: patient‐years’ exposure
RCT: randomised controlled trial
SD: standard deviation
vs: versus

The proportion of participants finishing the studies varied from 78% to 100% (Fulcher 2005; Porcellati 2004).

Study design

Two studies had a cross‐over design (SWITCH 1; Urakami 2017). The remaining studies were parallel‐group RCTs. All studies had an open‐label design, except for one which was double‐blinded (SWITCH 1). The duration of the intervention ranged from 24 weeks to 24 months. Seven studies had an additional extension period (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; BEGIN Young; Davies 2014; Standl 2004; Thalange 2013; Vague 2003).

All studies except two were multicentre studies (Porcellati 2004; Urakami 2017). The number of study centres ranged from 1 to 90. Sixteen studies were multinational (Bartley 2008; BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; BEGIN Young; Davies 2014; Heller 2009; Home 2005; Pieber 2007; PRESCHOOL; Robertson 2007; Russell‐Jones 2004; Schober 2002; Standl 2004; SWITCH 1; Thalange 2013; Vague 2003). None of the studies was performed in low‐ or middle‐income countries. None of the studies was terminated early.

Participants

Twenty‐three studies reported the ethnicity of the participants: 19 studies included mainly white people (Bartley 2008; BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; BEGIN Young; Chase 2008; Fulcher 2005; Heller 2009; Home 2005¸ NCT00595374; Pieber 2007; PRESCHOOL; Ratner 2000; Robertson 2007; Russell‐Jones 2004; Schober 2002; Standl 2004; SWITCH 1; Thalange 2013; Vague 2003), one study mainly Asian people (Davies 2014) and three studies included Asian people only (Kobayashi 2007; Liu 2016; NCT00605137) (Appendix 9).

All studies included both genders. The age of the participants varied from 4.2 to 44 years. The duration of T1DM varied from 2.1 to 23.2 years (Appendix 10).

Interventions

Nine studies compared insulin detemir with NPH insulin (Bartley 2008; Kobayashi 2007; NCT00595374; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003). Nine studies compared insulin glargine with NPH insulin (Bolli 2009; Chase 2008; Fulcher 2005; Home 2005; Liu 2016; Porcellati 2004; PRESCHOOL; Ratner 2000; Schober 2002). Two studies compared insulin detemir with insulin glargine (Heller 2009; Pieber 2007) and two studies compared insulin degludec with insulin detemir (BEGIN Young; Davies 2014). Finally, four studies compared insulin degludec with insulin glargine (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; SWITCH 1; Urakami 2017).

All studies except one applied NPH insulin once or two times daily. Porcellati 2004 applied NPH insulin four times a day.

Studies started insulin administration in different ways: four studies comparing insulin detemir with NPH insulin started with lower doses of insulin detemir compared with NPH insulin (Kobayashi 2007; NCT00605137; Russell‐Jones 2004; Standl 2004). One study comparing insulin degludec with insulin glargine stated that if prior basal insulin was taken more than once daily, then the dose of glargine had to be reduced by 20% to 30% and insulin degludec dose was reduced based on the investigators' decision (BEGIN Flex T1). Another study comparing insulin degludec with insulin glargine stated that if more than one daily dose had been taken prior to the study, then the total daily basal dose was calculated and replaced with insulin degludec in a 1:1 ratio and the insulin glargine dose was recommended to be reduced by 20% to 30% (BEGIN Basal‐Bolus Type 1). One study comparing insulin detemir with insulin glargine stated that the insulin detemir dose was reduced by 30% in both the morning and evening doses from the previous regimen and insulin glargine was started with a dose of 20% to 30% less than the previous regimen (Pieber 2007).

Eleven studies applied insulin aspart as fast‐acting insulin (Bartley 2008; BEGIN Young; Davies 2014; Heller 2009; Kobayashi 2007; Liu 2016; NCT00595374; Pieber 2007; Robertson 2007; Thalange 2013; Vague 2003); five studies applied insulin lispro (Bolli 2009; Chase 2008; Fulcher 2005; Porcellati 2004; PRESCHOOL); five studies applied human insulin (Home 2005; Ratner 2000; Schober 2002; Russell‐Jones 2004; Standl 2004) and one study did not specify the type of fast‐acting insulin applied (NCT00605137).

Outcomes

We could retrieve detailed study information for 23 studies (Bartley 2008; BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; BEGIN Young; Chase 2008; Davies 2014; Fulcher 2005; Heller 2009; Home 2005; Kobayashi 2007; Liu 2016; NCT00595374; NCT00605137; Pieber 2007; PRESCHOOL; Ratner 2000; Robertson 2007; Russell‐Jones 2004; Schober 2002; Standl 2004; SWITCH 1; Thalange 2013; Vague 2003). For six of the studies, trial protocols were available through the CSRs (Fulcher 2005; Home 2005; Ratner 2000; Schober 2002; Standl 2004; Vague 2003). For the remaining studies with a trial registration, information could be retrieved from the clinical trials register (see Appendix 12). Three studies provided data through publications only (Bolli 2009; Porcellati 2004; Urakami 2017) and one study author sent additional data (Urakami 2017).

All studies except three had predefined HbA1c as the primary outcome (NCT00605137; PRESCHOOL; SWITCH 1). All studies reported one or more outcome measures of relevance for this review.

Source of data

We contacted all study authors or investigators through email (see Appendix 14). When important information was lacking on ongoing trials and excluded studies, we contacted investigators for clarification (see Appendix 14).

Excluded studies

We excluded 22 studies after full‐text evaluation: eight studies had a wrong study design (not an RCT), six studies applied the wrong intervention, three studies included the wrong population, four studies had a short study duration and one reference was an irrelevant congress report. We evaluated four systematic reviews for identification of studies (Laranjeira 2018; Monami 2009; Tricco 2014; Tricco 2018). For further details see Characteristics of excluded studies.

Risk of bias in included studies

For the Cochrane RoB 2 assessment, we obtained CSRs, clinical study reports or both for 23 studies (Bartley 2008; BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; BEGIN Young; Chase 2008; Davies 2014; Fulcher 2005; Heller 2009; Home 2005; Kobayashi 2007; Liu 2016; NCT00595374; NCT00605137; Pieber 2007; PRESCHOOL; Ratner 2000; Robertson 2007; Russell‐Jones 2004; Schober 2002; Standl 2004; SWITCH 1; Thalange 2013; Vague 2003). We primarily used data from CSRs to evaluate risk of bias because the CSRs provided detailed information on all risk of bias domains for the RoB 2 tool. For two studies, we could obtain only parts of the original CSRs because the original documentation was written in Japanese and we did not get access to the full CSR (Kobayashi 2007; NCT00605137). For two studies, the clinical study synopses and a study protocol were the only source for data extraction (NCT00595374; NCT00605137).

For each specific outcome, we established an overall 'Risk of bias' judgement, as well as judgements per 'Risk of bias' domain (bias arising from the randomisation process, bias due to deviations from the intended interventions, bias due to missing outcome data, bias in measurement of the outcome, bias in selection of the reported results).

All‐cause mortality

All studies reporting deaths except two had a low overall risk of bias. Porcellati 2004 and Urakami 2017 had 'some concerns' because in these open‐label studies there was scarce information on methodological aspects of the studies.

Health‐related quality of life

All studies reporting health‐related quality of life except one had 'some concerns' for overall risk of bias because in these open‐label studies this outcome measure was primarily participant‐reported. SWITCH 1 had a low overall risk of bias for this outcome measure.

Severe hypoglycaemia

All studies reporting severe hypoglycaemia except three had a low overall risk of bias. Bolli 2009, Porcellati 2004 and Urakami 2017 had 'some concerns' because in these open‐label studies there was scarce information on methodological aspects of the studies.

Cardiovascular mortality

Non‐fatal myocardial infarction/stroke

All studies reporting non‐fatal myocardial infarction, non‐fatal stroke or both except one had a low overall risk of bias. Urakami 2017 had 'some concerns' because in this open‐label study there was scarce information on methodological aspects of the study.

End‐stage renal disease/blindness

The single study reporting end‐stage renal disease and blindness had a low overall risk of bias.

Serious adverse events

All studies reporting SAEs except two had a low overall risk of bias. Bolli 2009 and Urakami 2017 had 'some concerns' because in these open‐label studies there was scarce information on methodological aspects of the studies.

Diabetic ketoacidosis

All studies reporting diabetic ketoacidosis except one had a low overall risk of bias. Urakami 2017 had 'some concerns' because in this open‐label study there was scarce information on methodological aspects of the study.

Non‐serious adverse events

All studies reporting non‐serious adverse events had 'some concerns' for overall risk of bias because in these open‐label studies this outcome measurement was primarily participant‐reported.

Severe nocturnal hypoglycaemia

All studies reporting severe nocturnal hypoglycaemia except one had a low overall risk of bias. Urakami 2017 had 'some concerns' because in this open‐label study there was scarce information on methodological aspects of the study.

Mild/moderate hypoglycaemia

All studies reporting mild/moderate hypoglycaemia had some concerns for overall risk of bias because in these open‐label studies this outcome measurement was primarily participant‐reported.

Socioeconomic effects

No studies reported the costs of the intervention during the study period.

HbA1c levels

All studies reporting HbA1c except three had a low overall risk of bias. Bolli 2009, Porcellati 2004 and Urakami 2017 had 'some concerns' because in these open‐label studies there was scarce information on methodological aspects of the studies.

Combined HbA1c and severe hypoglycaemia

The studies providing some data on combined HbA1c and severe hypoglycaemia had a low overall risk of bias.

In general, referring to detailed information from the CSRs, the risk of bias evaluation was much more exhaustive compared to details reported in the publications. Most of our outcomes represented hard clinical (semi)objective outcomes with overall low risk of bias. However, for some outcomes, due to their subjective, participant‐reported nature, we attributed 'some concerns' to overall risk of bias for the outcomes health‐related quality of life, non‐serious adverse events, most measures of nocturnal hypoglycaemia and mild/moderate hypoglycaemia.

Risk of bias assessments for each outcome are located in the risk of bias table section after the characteristics of studies awaiting assessment and at the side of forest plots. For further details on the Excel file of risk of bias evaluation stored online in an open repository (Zenodo), please use the following link: https://zenodo.org/record/4549440.

Effects of interventions

Baseline characteristics

For details of baseline characteristics, see Appendix 9; Appendix 10.

Insulin degludec compared with NPH insulin

We identified no studies comparing insulin degludec with NPH insulin.

Insulin detemir compared with NPH insulin

For an overview of main results for this comparison see summary of findings Table 1.

Nine studies compared insulin detemir with NPH insulin (Bartley 2008; Kobayashi 2007; NCT00595374; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003). A total of 3345 participants were randomised, 2099 participants to insulin detemir and 1246 participants to NPH insulin (see Table 1). Three studies included children and randomised 781 children, 466 children to insulin detemir and 315 children to NPH insulin (NCT00605137; Robertson 2007; Thalange 2013). The mean age of the children varied from 8.4 to 9.9 years. Two of the studies did not have full‐text publications (NCT00595374; NCT00605137). We retrieved unpublished information on baseline variables or outcomes for all studies for this comparison.

Two studies randomised the participants to insulin detemir and NPH insulin once daily (Bartley 2008; Russell‐Jones 2004). However, a second dose of insulin detemir and NPH insulin could be added if necessary. For one of the studies, it was reported that 37% of the participants in the insulin detemir group and 45% of the participants in the NPH insulin group completed the study on a once‐daily regimen (Bartley 2008). Four studies randomised participants to NPH insulin once or twice daily (Kobayashi 2007; NCT00595374; NCT00605137; Robertson 2007). One study applied insulin detemir and NPH insulin once or twice daily according to a pre‐study regimen (Thalange 2013). One study randomised participants to insulin detemir and NPH insulin twice daily (Vague 2003). Six studies applied insulin aspart as fast‐acting insulin at meals (Bartley 2008; Kobayashi 2007; NCT00595374; Robertson 2007; Thalange 2013; Vague 2003). Two studies applied human insulin as fast‐acting insulin (Russell‐Jones 2004; Standl 2004). One study did not specify the type of fast‐acting insulin applied (NCT00605137).

The duration of the intervention varied from 24 weeks to 104 weeks (see Table 1).

Primary outcomes

All‐cause mortality

We could retrieve data on all‐cause mortality from all nine studies. However, only two studies reported mortality in their full‐text publication (Bartley 2008; Thalange 2013). We retrieved the remaining data from CSRs/clinical study synopses and medical reviews from regulatory agencies (Kobayashi 2007; NCT00595374; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Vague 2003).

A total of 5/2095 participants allocated to insulin detemir died compared with 0/1239 participants allocated to NPH insulin (Peto OR 4.97, 95% CI 0.79 to 31.38; P = 0.09; 9 studies, 3334 participants; moderate‐certainty evidence; Analysis 1.1). We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analyses

Analysing unpublished data only, 1/1587 participants in the insulin detemir group died compared with 0/905 participants in the NPH insulin group (2492 participants; 7 studies; Analysis 1.2). Analysing published data only 4/508 participants in the insulin detemir group compared with 0/334 in the NPH insulin group died (2 studies, 842 participants; Analysis 1.2). All five deaths occurred in studies including adults. The test for subgroup differences did not indicate interaction (P = 0.84). We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Health‐related quality of life

We rated the certainty of the evidence of the three studies with 870 participants providing information on health‐related quality of life as low. We judged the overall risk of bias for this outcome as 'some concerns'.

No study reported health‐related quality of life in a format making the data suitable for meta‐analysis. Kobayashi 2007 applied the Insulin Therapy Related Quality of Life at Night questionnaire (ITR‐QOLN); data were reported in the clinical study synopsis. The evaluation of ITR‐QOLN after 48 weeks showed higher scores in the insulin detemir group compared with the NPH insulin group (Kobayashi 2007). Standl 2004 applied the Diabetes Health Profile scale (only one of the three dimensions of the scale 'Barriers to activity'); data were reported in the CSR. After 26 weeks, the 'Barriers to activity' in the Diabetes Health profile was 0.71 (SD 0.75) in 210 participants in the insulin detemir group compared with 0.20 (SD 0.78) in 208 participants in the NPH insulin group. The P value was 0.52 (Standl 2004). Diabetes treatment satisfaction was also reported in the CSR (Standl 2004). Another unpublished trial reported in the clinical study synopsis that health‐related quality of life did not show any statistically significant differences between the interventions after 26 weeks but did not provide numerical data (NCT00595374).

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Severe hypoglycaemia

Eight studies reported data on severe hypoglycaemia (Bartley 2008; Kobayashi 2007; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003).

Analysing all available data showed that 171/2019 participants (8.5%) in the insulin detemir group compared with 138/1200 participants (11.5%) in the NPH insulin group experienced severe hypoglycaemia. There was a reduction in severe hypoglycaemia in favour of insulin detemir (RR 0.69, 95% CI 0.52 to 0.92; P = 0.01; 8 studies, 3219 participants; moderate‐certainty evidence; Analysis 1.3; Figure 3). The 95% prediction interval ranged between 0.34 and 1.39. We judged the overall risk of bias for this outcome as 'low'.

Figure 3

Severe hypoglycaemia

One study had an extension period (Standl 2004). We used data from the core period (six months) in the meta‐analysis. From the publication, only data after the end of the extension period (12 months) were available. However, we could retrieve additional data from the FDA medical review and the CSR (FDA 2002; Standl 2004). One study was unpublished, but data were available from a clinical study synopsis (NCT00605137). Another unpublished study reported no statistically significant differences for severe hypoglycaemia between the intervention groups but did not provide numerical data (NCT00595374). Five studies reported severe hypoglycaemia as requiring third party assistance (Bartley 2008; Kobayashi 2007; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004). One study added to this definition that blood glucose < 2.8 mmol/L should be recorded or symptom reversal with food, glucose or glucagon (Vague 2003). One study defined severe hypoglycaemia as episodes where the children were semi‐conscious, unconscious or in a coma, with or without convulsions (Thalange 2013). Bartley 2008 reported most events: data from the CSR of this study showed that 5/331 participants (1.5%) in the insulin detemir group compared with 6/164 participants (3.7%) in the NPH insulin group experienced a hypoglycaemic coma; 2/331 participants (0.6%) in the insulin detemir group compared with 0/164 participants (0%) in the NPH insulin group experienced hypoglycaemic convulsions; 0/331 participants (0%) in the insulin detemir group compared with 1/164 participants (0.6%) in the NPH insulin group experienced loss of consciousness due to hypoglycaemia. Robertson 2007 reported most events in children: 3/232 children (1.3%) in the insulin detemir group compared with 3/115 children (2.6%) in the NPH insulin group were admitted to hospital due to hypoglycaemia; 4/232 children (1.7%) in the insulin detemir group compared with 4/115 children (3.4%) in the NPH insulin group were unconscious due to hypoglycaemia; 2/332 children (0.6%) in the insulin detemir group compared with 4/115 children (3.4%) in the NPH insulin group experienced hypoglycaemia with convulsions; 4/332 children (1.2%) in the insulin detemir group compared with 2/115 children (1.7%) in the NPH insulin group received glucagon treatment. One study stipulated, that the risk of experiencing hypoglycaemia could have been influenced by lack of blinding: "Investigators and patients in this trial may have been reluctant to aggressively increase the dose of a new basal insulin preparation such as insulin detemir because of the fear of hypoglycemia, especially during the night" (Vague 2003).

Subgroup and sensitivity analyses

Analysing studies including adults only indicated an RR of 0.71, 95% CI 0.49 to 1.03; 5 studies, 2443 participants; Analysis 1.3. Analysing studies including children only indicated an RR of 0.61, 95% CI 0.30 to 1.23; 3 studies, 776 children; Analysis 1.3. The test for subgroup differences did not indicate interaction (P = 0.72).

Restricting the analysis to published data only indicated an RR of 0.62, 95% CI 0.50 to 0.78; 6 studies, 2677 participants; Analysis 1.4; favouring insulin detemir. Restricting the analyses to unpublished data only indicated an RR of 1.42, 95% CI 0.77 to 2.62; 2 studies, 498 participants; Analysis 1.4 . The test for subgroup differences indicated interaction (P = 0.01). This has to be interpreted with caution because the subgroup of studies with unpublished data consisted of two studies only and the CIs slightly overlapped.

A sensitivity analysis excluding the largest study (Russell‐Jones 2004) indicated an RR of 0.68, 95% CI 0.48 to 0.97. A sensitivity analysis excluding the longest study (Bartley 2008) indicated an RR of 0.72, 95% CI 0.51 to 1.04.

A sensitivity analysis with data from studies published in English only (Bartley 2008; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003) indicated an RR of 0.70, 95% CI 0.52 to 0.95.

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Hypoglycaemia reported as a serious adverse event

A total of 30/2019 participants (1.5%) in the insulin detemir group compared with 19/1200 participants (1.6%) in the NPH insulin group had a SAE due to hypoglycaemia. There was no evidence of a difference in hypoglycaemia reported as a SAE (RR 0.93, 95% CI 0.51 to 1.71; P = 0.82; 8 studies, 3219 participants; Analysis 1.5). The 95% prediction interval ranged between 0.44 and 1.99. We judged the overall risk of bias for this outcome as 'low' (data not shown).

The test for subgroup differences comparing adults with children did not indicate interaction (P = 1.00). We judged the overall risk of bias for this outcome as 'low' (data not shown).

Secondary outcomes

Cardiovascular mortality

We could retrieve data on cardiovascular mortality from all studies. Only two studies reported cardiovascular mortality in their full‐text publication (Bartley 2008; Thalange 2013). We retrieved the remaining data from CSRs/clinical study synopses/medical reviews from regulatory agencies (Kobayashi 2007; NCT00595374; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Vague 2003).

Only one adult participant died due to cardiovascular disease (Analysis 1.6). This participant belonged to the insulin detemir group (1/2069 participants). No participant died in the NPH insulin group (0/1221 participants). We judged the overall risk of bias for this outcome as 'low'.

Subgroup analysis and sensitivity analysis

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐fatal myocardial infarction

None of the included studies reported non‐fatal myocardial infarction in the publications. One study had data on non‐fatal myocardial infarction from the CSR (Bartley 2008). In this study, 1/331 participants in the insulin detemir group compared with 0/164 participants in the NPH insulin group experienced a non‐fatal myocardial infarction (low‐certainty evidence; Analysis 1.7). One study reported data at the end of the extension period (duration of intervention was six months with an additional six months extension period) with 1/154 participants in the insulin detemir group and 0/135 participants in the NPH insulin group experiencing a myocardial infarction (Standl 2004). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐fatal stroke

No study reported on non‐fatal stroke.

End‐stage renal disease

No study reported on end‐stage renal disease.

Blindness

No study reported on blindness.

Serious adverse events

We could retrieve data on SAEs from all studies.

In the insulin detemir group, 165/2094 participants (7.9%) reported a SAE compared with 102/1238 participants (8.2%) in the NPH insulin group. There was no evidence of a difference in SAEs (RR 0.95, 95% CI 0.75 to 1.21; P = 0.67; 9 studies, 3332 participants; moderate‐certainty evidence; Analysis 1.8). The 95% prediction interval ranged between 0.71 and 1.27. We judged the overall risk of bias for this outcome as 'low'.

Three studies reported SAEs in the main publications in a format making data unsuitable for meta‐analysis: one study reported that the frequency and type of adverse events observed during the study were similar with insulin detemir and NPH insulin (Russell‐Jones 2004); one study reported that fewer than 5% in each intervention group reported SAEs (Vague 2003) and one study reported that about 10% of participants in both intervention groups experienced SAEs (Standl 2004). However, in the CSRs of these studies, data were reported in a way making them suitable for meta‐analysis.

Subgroup and sensitivity analyses

Six studies had data on SAEs for adults: 124/1630 participants (7.6%) in the insulin detemir group compared with 71/926 participants (7.7%) in the NPH insulin group experienced SAEs. The RR was 0.97, 95% CI 0.73 to 1.28; 6 studies, 2556 participants; Analysis 1.8. Three studies had data on SAEs for children: 41/464 children (8.8%) in the insulin detemir group compared with 31/312 children (9.9%) in the NPH insulin group experienced SAEs. The RR was 0.89, 95% CI 0.69 to 1.27; 3 studies, 776 children; Analysis 1.8. The test for subgroup differences did not indicate interaction (P = 0.77).

Restricting the analyses to published data only for SAEs indicated an RR of 0.66, 95% CI 0.40 to 1.09; 2 studies, 641 participants; Analysis 1.9. Restricting analysis to unpublished data only indicated an RR of 1.06, 95% CI 0.80 to 1.39; 6 studies, 2691 participants; Analysis 1.9. The test for subgroup differences did not indicate interaction (P = 0.11).

Sensitivity analysis excluding the largest study (Vague 2003) indicated an RR of 0.93, 95% CI 0.70 to 1.25. Sensitivity analysis excluding the longest study (Bartley 2008) indicated an RR of 0.96, 95% CI 0.72 to 1.29.

A sensitivity analysis with data from studies published in English only indicated an RR of 0.89, 95% CI 0.56 to 1.43 (Bartley 2008; NCT00595374; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003).

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Diabetic ketoacidosis

We could retrieve data on diabetic ketoacidosis from six studies (Bartley 2008; Kobayashi 2007; NCT00605137; Robertson 2007; Thalange 2013; Vague 2003). Two studies reported ketoacidosis in their full‐text publications (Robertson 2007; Thalange 2013). One study was unpublished, but we retrieved data from the clinical study synopsis (NCT00605137). Three studies reported diabetic ketoacidosis in CSRs (Bartley 2008; Kobayashi 2007; Vague 2003). It appeared likely that all studies had evaluated this outcome but some did not report this outcome measure (NCT00595374; Russell‐Jones 2004; Standl 2004).

A total of 14/1292 participants (1.1%) experienced diabetic ketoacidosis in the insulin detemir group compared with 10/720 participants (1.4%) in the NPH insulin group. There was no evidence of a difference in diabetic ketoacidosis (RR 0.80, 95% CI 0.36 to 1.76; P = 0.58; 6 studies, 2012 participants; Analysis 1.10). We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analyses

Three studies reported diabetic ketoacidosis in adults; the RR was 0.84, 95% CI 0.24 to 2.92; 3 studies, 1236 participants; Analysis 1.10. Three studies reported diabetic ketoacidosis in children; the RR was 0.77, 95% CI 0.27 to 2.15; 3 studies, 776 children; Analysis 1.10. The test for subgroup differences did not indicate interaction (P = 0.91).

Restricting the analyses to only published data for diabetic ketoacidosis indicated an RR of 0.83, 95% CI 0.27 to 2.52; 2 studies, 694 participants; Analysis 1.11. Restricting the analyses to only unpublished data for diabetic ketoacidosis indicated an RR of 0.77, 95% CI 0.25 to 2.38; 4 studies, 1318 participants; Analysis 1.11.

Sensitivity analysis excluding the largest study and the longest study indicated an RR of 0.86, 95% CI 0.34 to 2.20 (Bartley 2008).

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐serious adverse events

We could retrieve data on non‐serious adverse events from all studies. Only four studies reported non‐serious adverse events in a format suitable for meta‐analysis in their full‐text publications (Robertson 2007; Standl 2004; Thalange 2013; Vague 2003). For the remaining studies, we retrieved data from CSRs/clinical study synopses (Bartley 2008; Kobayashi 2007; NCT00595374; NCT00605137; Russell‐Jones 2004).

A total of 1622/2094 participants (77.5%) in the insulin detemir group compared with 968/1238 participants (78.2%) in the NPH insulin group experienced a non‐serious adverse event. There was no evidence of a difference in non‐serious adverse events (RR 0.98, 95% CI 0.94 to 1.01; P = 0.22; 9 studies, 3332 participants; Analysis 1.12). The 95% prediction interval ranged between 0.95 and 1.02. We judged the overall risk of bias for this outcome as 'some concerns'.

Subgroup analysis and sensitivity analysis

Five studies reported non‐serious adverse events in adults. A total of 1242/1630 participants (76.2%) in the insulin detemir group compared with 706/926 participants (76.2%) in the NPH insulin group experienced a non‐serious adverse event. The RR was 0.99, 95% CI 0.95 to 1.03; Analysis 1.12. Three studies including children reported 380/464 children (81.9%) in the insulin detemir group compared with 262/312 (84.0%) children in the NPH insulin group experienced a non‐serious adverse event. The RR was 0.96, 95% CI 0.90 to 1.02; Analysis 1.12. The test for subgroup differences did not indicate interaction (P = 0.40).

Restricting the analyses to only published data indicated 553/710 participants (77.9%) in the insulin detemir group compared with 351/431 participants (81.4%) in the NPH insulin group experienced a non‐serious adverse event. The RR was 0.95, 95% CI 0.90 to 1.01; Analysis 1.13. Restricting the analyses to only unpublished data indicated 1069/1384 participants (77.2%) in the insulin detemir group compared with 617/807 participants (76.5%) in the NPH insulin group experienced a non‐serious adverse event. The RR was 1.00, 95% CI 0.95 to 1.04; Analysis 1.13. The test for subgroup differences did not indicate interaction (P = 0.25).

Sensitivity analysis excluding the largest study (Russell‐Jones 2004) indicated an RR of 0.97, 95% CI 0.93 to 1.01 and excluding the longest study (Bartley 2008) indicated an RR of 0.98, 95% CI 0.94 to 1.02.

A sensitivity analysis with data from studies published in English only indicated an RR of 0.98, 95% CI 0.94 to 1.01 (Bartley 2008; NCT00595374; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003).

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Withdrawals due to adverse events

A total of 30/2020 participants (1.5%) in the insulin detemir group compared with 6/1202 participants (0.5%) in the NPH insulin group withdrew because of adverse events. There was no evidence of a difference in withdrawals due to adverse events (RR 2.23, 95% CI 0.98 to 5.05; P = 0.05; 8 studies, 3222 participants; Analysis 1.14). The 95% prediction interval ranged between 0.80 and 6.19. We judged the overall risk of bias for this outcome as 'low' (data not shown).

Nocturnal hypoglycaemia

We could retrieve data on nocturnal hypoglycaemia from eight studies (Bartley 2008; Kobayashi 2007; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003).

Seven studies reported severe nocturnal hypoglycaemia. A total of 70/1823 participants (3.8%) in the insulin detemir group compared with 60/1102 participants (5.4%) in the NPH insulin group experienced a severe nocturnal hypoglycaemic event. There was no evidence of a difference in severe nocturnal hypoglycaemia (RR 0.67, 95% CI 0.39 to 1.17, P = 0.16; 7 studies, 2925 participants; moderate‐certainty evidence; Analysis 1.18). We judged the overall risk of bias for this outcome as 'low'.

The studies applied different ways of reporting nocturnal hypoglycaemia. In the trial synopsis of one study, authors wanted to investigate major nocturnal hypoglycaemia, minor nocturnal hypoglycaemia, nocturnal hypoglycaemia with symptoms only and biochemical nocturnal hypoglycaemia (defined as asymptomatic plasma glucose value). However, only the outcome of any nocturnal hypoglycaemic events was reported (Kobayashi 2007). In the CSR of this study, data for subtypes of hypoglycaemia were provided in a format making them unsuitable for meta‐analysis: minor nocturnal hypoglycaemia had an RR of 0.67, 95% CI 0.42 to 1.06; symptoms only nocturnal hypoglycaemia had an RR of 0.58, 95% CI 0.31 to 1.09 and biochemical nocturnal hypoglycaemia had an RR of 0.77, 95% CI 0.45 to 1.33 (Kobayashi 2007). One unpublished study reported nocturnal hypoglycaemia in a format suitable for meta‐analysis in the CSR (NCT00605137). Data for another unpublished study (NCT00595374) were reported in a format making them unsuitable for meta‐analysis ('no significant differences between the intervention groups').

The data for the analysis of any type of nocturnal hypoglycaemia were available in the full‐text articles of six studies (Bartley 2008; Kobayashi 2007; Robertson 2007; Russell‐Jones 2004; Thalange 2013; Vague 2003). Two studies provided data in the CSRs (NCT00605137; Standl 2004). Data for mild nocturnal hypoglycaemia and symptomatic nocturnal hypoglycaemia (without confirmed blood glucose values) could be retrieved from seven studies: four studies reported the outcome in the publication (Bartley 2008; Robertson 2007; Russell‐Jones 2004; Thalange 2013) and three studies provided the data from unpublished sources (NCT00605137; Standl 2004; Vague 2003). One study reported data on asymptomatic hypoglycaemia (Thalange 2013).

A total of 1041/1555 participants (66.9%) in the insulin detemir group compared with 877/1200 participants (73.1%) in the NPH insulin group experienced any type of nocturnal hypoglycaemic event. There was a reduction in any type of nocturnal hypoglycaemia in favour of insulin detemir (RR 0.91, 95% CI 0.87 to 0.95; P < 0.001; 8 studies, 3219 participants; Analysis 1.15). The 95% prediction interval ranged between 0.86 and 0.96. There was a reduction in mild nocturnal hypoglycaemia in favour of insulin detemir (RR of 0.90, 95% CI 0.85 to 0.96; P = 0.002; 7 studies, 3073 participants; Analysis 1.16). There was a reduction in nocturnal hypoglycaemia with symptoms in favour of insulin detemir (RR 0.88, 95% CI 0.79 to 0.98; P = 0.02; 6 studies, 2578 participants; Analysis 1.17). One study reported asymptomatic nocturnal hypoglycaemia in 83/177 participants (46.9%) in the insulin detemir group compared with 85/170 participants (50%) in the NPH insulin group (Thalange 2013). We judged the overall risk of bias for all these outcomes except for severe nocturnal hypoglycaemia as 'some concerns' (data not shown).

Subgroup and sensitivity analyses

Five studies reported any type of nocturnal hypoglycaemia in adults. A total of 1041/1555 participants (66.9%) in the insulin detemir group compared with 629/888 participants (70.8%) in the NPH insulin group experienced any type of nocturnal hypoglycaemia. The RR was 0.93, 95% CI 0.88 to 0.98; Analysis 1.15; favouring insulin detemir. Three studies including children reported that 337/464 children (72.6%) in the insulin detemir group compared with 258/312 children (82.7%) in the NPH insulin group experienced any type of nocturnal hypoglycaemia. The RR was 0.87, 95% CI 0.81 to 0.94; Analysis 1.15; favouring insulin detemir. The test for subgroup differences did not indicate interaction (P = 0.23).

Four studies reported mild nocturnal hypoglycaemia in adults. The RR was 0.91, 95% CI 0.83 to 1.00; Analysis 1.16; favouring insulin detemir. Three studies reported mild nocturnal hypoglycaemia in children. The RR was 0.88, 95% CI 0.78 to 1.00; Analysis 1.16; favouring insulin detemir. The test for subgroup differences did not indicate interaction (P = 0.66).

Four studies reported nocturnal hypoglycaemia with symptoms in adults. The RR was 0.91, 95% CI 0.82 to 1.01; Analysis 1.17. Two studies reported nocturnal hypoglycaemia with symptoms in children. The RR was 0.55, 95% CI 0.19 to 1.61; Analysis 1.17. The test for subgroup differences did not indicate interaction (P = 0.36).

Four studies reported severe nocturnal hypoglycaemia in adults. The RR was 0.57, 95% CI 0.35 to 0.93; Analysis 1.8; favouring insulin detemir. Three studies including children reported severe nocturnal hypoglycaemia. The RR was 0.64, 95% CI 0.13 to 3.17; Analysis 1.18. The test for subgroup differences did not indicate interaction (P = 0.88).

Six studies had published information on any type of nocturnal hypoglycaemia. The RR was 0.90, 95% CI 0.86 to 0.95; Analysis 1.19; favouring insulin detemir. Two studies had unpublished data on any type of nocturnal hypoglycaemia. The RR was 0.91, 95% CI 0.80 to 1.04; Analysis 1.19. The test for subgroup differences did not indicate interaction (P = 0.90).

Four studies had published information on mild nocturnal hypoglycaemia. The RR was 0.91, 95% CI 0.85 to 0.98; Analysis 1.20; favouring insulin detemir. Three studies had unpublished information on mild nocturnal hypoglycaemia. The RR was 0.89, 95% CI 0.75 to 1.07; Analysis 1.20. The test for subgroup differences did not indicate interaction (P = 0.83).

Three studies had published information on nocturnal hypoglycaemia with symptoms. The RR was 0.90, 95% CI 0.81 to 0.99; Analysis 1.21; favouring insulin detemir. Three studies had unpublished information on nocturnal hypoglycaemia with symptoms. The RR was 0.79, 95% CI 0.57 to 1.08; Analysis 1.21. The test for subgroup differences did not indicate interaction (P = 0.44).

Five studies had published information on severe nocturnal hypoglycaemia. The RR was 0.63, 95% CI 0.32 to 1.25; Analysis 1.22. Two studies had unpublished information on severe nocturnal hypoglycaemia. The RR was 0.90, 95% CI 0.33 to 2.45; Analysis 1.22. The test for subgroup differences did not indicate interaction (P = 0.56).

Sensitivity analysis excluding the longest study (Bartley 2008) for any type of nocturnal hypoglycaemia indicated an RR of 0.90, 95% CI 0.86 to 0.94 favouring insulin detemir.

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Mild/moderate hypoglycaemia

We could retrieve data on mild/moderate hypoglycaemia from eight studies (Bartley 2008; Kobayashi 2007; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003). One unpublished study reported data for mild/moderate hypoglycaemia in a format making the data unsuitable for meta‐analysis (NCT00595374). One study did not specify mild hypoglycaemia; for this study we used data for any type of hypoglycaemia (NCT00605137). For the remaining studies, data for mild hypoglycaemia were available (Bartley 2008; Kobayashi 2007; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003).

A total of 1726/2019 participants (85.5%) in the insulin detemir compared with 1028/1200 participants (85.7%) in the NPH insulin group experienced mild/moderate hypoglycaemia. There was a reduction in mild/moderate hypoglycaemia in favour of insulin detemir (RR 0.97, 95% CI 0.94 to 0.99; P = 0.01, 8 studies, 3219 participants; Analysis 1.24). The 95% prediction interval ranged between 0.95 and 1.00. We judged the overall risk of bias for this outcome as 'some concerns'.

Subgroup and sensitivity analyses

Five studies reported mild/moderate hypoglycaemia in adults. A total of 1313/1555 participants (84.4%) in the insulin detemir group compared with 742/888 participants (83.4%) in the NPH insulin group experienced mild/moderate hypoglycaemia. The RR was 0.97, 95% CI 0.93 to 1.02; Analysis 1.24. Three studies including children reported 413/464 children (89.0%) in the insulin detemir group compared with 286/312 children (91.7%) in the NPH insulin group experienced mild/moderate hypoglycaemia. The RR was 0.97, 95% CI 0.93 to 1.01; Analysis 1.24. The test for subgroup differences did not indicate interaction (P = 0.82).

Six studies had published information on mild/moderate hypoglycaemia. The RR was 0.97, 95% CI 0.93 to 1.00; Analysis 1.25; favouring insulin detemir. Two studies had unpublished information on mild/moderate hypoglycaemia. The RR was 0.98, 95% CI 0.92 to 1.05; Analysis 1.25. The test for subgroup differences did not indicate interaction (P = 0.69).

Sensitivity analysis excluding the largest study indicated an RR of 0.96, 95% CI 0.93 to 0.98 (Russell‐Jones 2004) favouring insulin detemir. Sensitivity analysis excluding the longest study indicated an RR of 0.97, 95% CI 0.94 to 1.00 (Bartley 2008) favouring insulin detemir.

A sensitivity analysis with data from studies published in English only indicated an RR of 0.97, 95% CI 0.94 to 1.00 favouring insulin detemir.

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Socioeconomic effects

No studies reported direct or indirect costs of the intervention during the study period. One study reported economic predictions of the interventions based on simulation cohorts in Belgian, Canadian, French, German, Italian and Spanish, Swedish settings (Bartley 2008).

HbA1c

We could retrieve data on HbA1c levels from eight studies (Bartley 2008; Kobayashi 2007; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003). Six studies reported HbA1c levels in publications (Bartley 2008; Kobayashi 2007; Robertson 2007; Russell‐Jones 2004; Thalange 2013; Vague 2003). Standl 2004 only reported HbA1c after the end of the extension period in publications, but through FDA review and CSR, we could retrieve data at the end of the regular intervention period. One unpublished study reported HbA1c in the clinical study synopsis (NCT00605137).

There was no evidence of a difference in HbA1c (MD 0.01%, 95% CI ‐0.1 to 0.1; P = 0.11; 8 studies, 3122 participants; moderate‐certainty evidence; Analysis 1.26). The 95% prediction interval ranged between ‐0.1% and 0.1%. We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analyses

Five studies reported HbA1c levels in adults. The MD of HbA1c was ‐0.03%, 95% CI ‐0.1 to 0.1; Analysis 1.26. Three studies including children reported HbA1c levels. The MD of HbA1c was 0.1%, 95% CI ‐0.04 to 0.3; Analysis 1.26. The test for subgroup differences did not indicate interaction (P = 0.11).

Analysing only published data indicated a MD of HbA1c of ‐0.02%, 95% ‐0.1 to 0.1; Analysis 1.27. Analysing only unpublished data indicated a MD of HbA1c of 0.1%, 95% CI ‐0.1 to 0.3; Analysis 1.27. The test for subgroup differences did not indicate interaction (P = 0.28). One unpublished study reported data for HbA1c in a format making the data unsuitable for meta‐analysis (NCT00595374).

Sensitivity analysis excluding the largest study indicated a MD of HbA1c of 0.02%, 95% CI ‐0.1 to 0.1 (Russell‐Jones 2004). Sensitivity analysis excluding the longest study indicated a MD of HbA1c of 0.04%, 95% CI ‐0.1 to 0.1 (Bartley 2008).

Sensitivity analysis exclusively analysing data from studies published in English indicated a MD of HbA1c of ‐0.01%, 95% CI ‐0.1 to 0.1 (Bartley 2008; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003).

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Combined HbA1c and severe hypoglycaemia

No study reported on combined HbA1c and severe hypoglycaemia.

One study provided data on the combined outcome HbA1c and hypoglycaemia (Bartley 2008). We extracted these data from the CSR. This specified the percentage of participants who reached HbA1c ≤ 7.0% at the end of the study without symptomatic hypoglycaemia with a plasma glucose < 4.0 mmol/L or any single plasma glucose value < 3.1 mmol/L during the last month of treatment. This number was 71/321 participants (22.2%) in the insulin detemir group compared with 21/159 participants (13.2%) in the NPH insulin group.

Two studies stated that similar results were seen for hypoglycaemia when adjusted for HbA1c (Robertson 2007; Vague 2003). One study reported in the CSR that the observed risk of hypoglycaemia was not explained by differences in HbA1c (Russell‐Jones 2004).

Insulin glargine compared with NPH insulin

For an overview of main results for this comparison see summary of findings Table 2.

Nine studies compared insulin glargine with NPH insulin (Bolli 2009; Chase 2008; Fulcher 2005; Home 2005; Liu 2016; Porcellati 2004; PRESCHOOL; Ratner 2000; Schober 2002). A total of 2387 participants were randomised, 1205 participants to insulin glargine and 1182 participants to NPH insulin (see Table 1). Four studies included children and randomised 823 children, 433 children to insulin glargine and 390 children to NPH insulin (Chase 2008; Liu 2016; PRESCHOOL; Schober 2002). The mean age of the children varied from 4.2 to 13.2 years.

All studies were published as full‐text articles in English. However, we retrieved unpublished information from most studies for this comparison (Chase 2008; Fulcher 2005; Home 2005; Liu 2016; PRESCHOOL; Ratner 2000; Schober 2002). Two studies had information solely based on full‐text publications (Bolli 2009; Porcellati 2004). We contacted investigators in order to achieve additional information, but did not receive a reply (see Appendix 19).

One study randomised participants to insulin glargine once daily and NPH insulin once daily (Fulcher 2005). One study randomised participants to insulin glargine once daily and NPH insulin or Lente insulin twice daily according to a pre‐study regimen. However, only three participants received Lente insulin (Chase 2008). Six studies randomised participants to insulin glargine once daily and NPH insulin (Bolli 2009; Home 2005; PRESCHOOL; Liu 2016; Ratner 2000; Schober 2002). One study randomised participants to insulin glargine once daily and NPH insulin four times a day (Porcellati 2004).

Five studies applied insulin lispro as fast‐acting insulin at meals (Bolli 2009; Chase 2008; Fulcher 2005; Porcellati 2004; PRESCHOOL). One study applied insulin aspart as fast‐acting insulin (Liu 2016). Three studies applied human insulin as fast‐acting insulin (Home 2005; Ratner 2000; Schober 2002).

The duration of the intervention varied from 24 weeks to 30 weeks.

Primary outcomes

All‐cause mortality

We could retrieve data on all‐cause mortality from eight studies (Chase 2008; Fulcher 2005; Home 2005; Liu 2016; Porcellati 2004; PRESCHOOL; Ratner 2000; Schober 2002). Only one of these studies reported all‐cause mortality in the full‐text publication (Porcellati 2004). We obtained the remaining data from unpublished sources.

A total of 0/1207 participants allocated to insulin glargine died compared with 1/1068 participants allocated to NPH insulin (Peto OR 0.14, 95% CI 0.00 to 6.98; P = 0.32; 8 studies, 2175 participants; moderate‐certainty evidence; Analysis 2.1). We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analyses

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Health‐related quality of life

Four studies reported health‐related quality of life (Bolli 2009; Chase 2008; Home 2005; Ratner 2000). We judged the certainty of the evidence for these studies with 1013 participants as low. We judged the overall risk of bias for this outcome as 'some concerns'.

One study applied the Well‐Being Enquiry for Diabetics (Bolli 2009), two studies applied the General Well‐being scale (Home 2005; Ratner 2000) and one study applied the Diabetes Quality of Life for Youth and Parents' Diabetes Quality of Life (Chase 2008). Bolli 2009 randomised 175 participants. After six months, data from 133 participants were evaluated for impact domain, 114 participants for level of satisfaction, 108 participants for general worries and 111 participants for diabetes‐related worries. It was not reported how many participants in each intervention arm were included in the analysis. The only domain showing a statistically significant difference after six months was diabetes‐related worries, which showed greater improvements in the insulin glargine group (P = 0.05). At six months, the impact domain score was 77 (quartiles 73 to 82) in the insulin glargine group and 80 (quartiles 73 to 85) in the NPH insulin group. Changes in percentage from baseline were ‐1.4 (quartiles ‐10 to 8) in the insulin glargine group and ‐4.4 (quartiles ‐14 to 7) in the NPH insulin group. At six months, the level of satisfaction score was 31 (quartiles 27 to 35) in the insulin glargine group and 32 (quartiles 27 to 38) in the NPH insulin group. Changes in percentage from baseline were 0.0 (quartiles ‐10 to 8) in the insulin glargine group and ‐3.0 (quartiles ‐7 to 3) in the NPH insulin group. At six months, the general worries score was 32 (quartiles 27 to 34) in the insulin glargine group and 32 (quartiles 26 to 35) in the NPH insulin group. Changes in percentage from baseline were ‐1.4 (quartiles ‐7 to 3) in the insulin glargine group and 0.0 (quartiles ‐11 to 4) in the NPH insulin group. At six months, the diabetes‐related worries score was 32 (quartiles 27 to 34) in the insulin glargine group and 31 (quartiles 25 to 34) in the NPH insulin group. Changes in percentage from baseline were ‐5.7 (quartiles ‐12 to 4) in the insulin glargine group and 0.0 (quartiles ‐8 to 8) in the NPH insulin group (P = 0.05) (Bolli 2009). Two studies applied the General Well‐being scale (Home 2005; Ratner 2000). One study reported health‐related quality of life through a CSR (Ratner 2000). Home 2005 reported in a co‐publication that the mean score for the General Well‐being scale showed an increase (i.e. better well‐being) of 1.44 points at week 28 in the insulin glargine group compared with 1.57 points in the NPH insulin group with all four subscales contributing to these improvements (Home 2005). In the CSR, health‐related quality of life with SDs at the end of intervention were reported (Home 2005). Combining data from the two studies applying the General Well‐being scale did not show evidence of a difference (MD 0.62 points, 95% CI ‐0.71 to 1.96; P = 0.36; 2 studies, 880 participants; Analysis 2.2). For both studies, the difference between the treatments was not statistically significant at the end of follow‐up for each separate item of the General Well‐being scale (depression, anxiety, energy, positive well‐being). One study evaluated health‐related quality of life in children (Chase 2008). Data were available from the clinical study synopsis. This study applied the Diabetes Quality of Life for Youth and Parents' Diabetes Quality of Life (Chase 2008). This study did not find evidence of a difference between the interventions. No information about scores or number of participants included in the analysis was reported.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Severe hypoglycaemia

Nine studies reported data on severe hypoglycaemia (Bolli 2009; Chase 2008; Fulcher 2005; Home 2005; Liu 2016; Porcellati 2004; PRESCHOOL; Ratner 2000; Schober 2002). All studies defined severe hypoglycaemia as hypoglycaemia requiring third party assistance. For two studies, we retrieved unpublished data from the CSRs (Fulcher 2005; Ratner 2000).

A total of 122/1191 participants (10.2%) in the insulin glargine group compared with 145/1159 participants (12.5%) in the NPH insulin group experienced severe hypoglycaemia. There was no evidence of a difference in severe hypoglycaemia (RR 0.84, 95% CI 0.67 to 1.04; P = 0.11; 9 studies, 2350 participants; moderate‐certainty evidence; Analysis 2.3; Figure 4). The 95% prediction interval ranged between 0.65 and 1.09. We judged the overall risk of bias for this outcome as 'low'.

Figure 4

Severe hypoglycaemia

One study in the main publication defined severe hypoglycaemia as requiring third party assistance in the methods section of the main publication (Ratner 2000). However, the definition of severe hypoglycaemia reported in the results section in the main publication was severe hypoglycaemic event with blood glucose levels < 2.0 mmol/L. In the CSR, severe hypoglycaemia with and without confirmed blood glucose < 2.0 mmol/L was reported. With the definition of severe hypoglycaemia according to the methods section, 23/264 participants (8.7%) in the insulin glargine group compared with 28/270 participants (10.4%) in the NPH insulin group experienced severe hypoglycaemia. This number was used for the meta‐analysis. Using severe hypoglycaemia applying the definition of blood glucose < 2.0 mmol/L showed that 7/264 participants (2.5%) in the insulin glargine group compared with 16/270 participants (5.9%) in the NPH insulin group experienced severe hypoglycaemia (Ratner 2000). From the CSR, it was also apparent, that during the screening phase no participants receiving insulin glargine during the study had an episode of severe hypoglycaemia compared with 6/270 participants (2.2%) receiving NPH insulin (Ratner 2000). One study stated in the FDA report that the participants receiving NPH insulin twice daily tended to have less hypoglycaemia than the participants receiving insulin glargine (FDA 2000; Home 2005). Schober 2002 reported the greatest number of events in children: in the CSR, 1/174 children (0.6%) in the insulin glargine group compared with 1/175 children (0.6%) in the NPH insulin group experienced coma due to hypoglycaemia; 4/174 children (2.3%) in the insulin glargine group compared with 3/175 children (1.7%) in the NPH insulin group experienced convulsions due to hypoglycaemia; 6/174 children (3.4%) in the insulin glargine group compared with 1/175 children (0.6%) in the NPH insulin group experienced syncope due to hypoglycaemia. Home 2005 reported the greatest number of events in adults: in the CSR, 7/292 participants (2.4%) in the insulin glargine group compared with 12/293 participants (4.1%) in the NPH insulin group experienced coma, convulsions or syncope reported as associated symptoms from severe hypoglycaemia (Home 2005).

Subgroup analysis and sensitivity analysis

Analysing studies including only adults indicated an RR of 0.78, 95% CI 0.58 to 1.05; Analysis 2.3. Analysing studies including only children indicated an RR of 1.14, CI 95% CI 0.59 to 2.21; Analysis 2.3. The test for subgroup differences did not indicate interaction (P = 0.31).

Restricting the analysis to only published data indicated an RR of 0.87, 95% CI 0.63 to 1.22; Analysis 2.4. Restricting the analysis to only unpublished data indicated an RR of 0.83. 95% CI 0.56 to 1.25; Analysis 2.4. The test for subgroup differences did not indicate interaction (P = 0.87).

Sensitivity analysis excluding the largest study and the longest study indicated an RR of 0.88, 95% CI 0.68 to 1.14 (Home 2005).

All studies except one had received funding from the pharmaceutical industry (Porcellati 2004). Porcellati 2004 applied NPH insulin four times a day. Excluding this study from the analysis indicated an RR of 0.83, 95% CI 0.67 to 1.04.

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Hypoglycaemia reported as a serious adverse event

A total of 52/1131 participants (4.6%) in the insulin glargine group compared with 56/1098 participants (5.1%) in the NPH insulin group had a SAE due to hypoglycaemia. There was no evidence of a difference in hypoglycaemia reported as a SAE (RR 0.94, 95% CI 0.64 to 1.39; P = 0.76; 8 studies, 2229 participants; Analysis 2.5). The 95% prediction interval ranged between 0.52 and 1.71. We judged the overall risk of bias for this outcome as 'low' (data not shown).

The test for subgroup differences comparing adults with children did not indicate interaction (P = 0.90).

Secondary outcomes

Cardiovascular mortality

We could retrieve data on cardiovascular mortality from eight studies (Chase 2008; Fulcher 2005; Home 2005; Liu 2016; Porcellati 2004; PRESCHOOL; Ratner 2000; Schober 2002). Only one of these studies reported cardiovascular mortality in the full‐text publication (Porcellati 2004). We retrieved the remaining data from unpublished sources.

Analysing all available data showed 0/1106 participants allocated to insulin glargine died compared with 1/1068 participants allocated to NPH insulin (Analysis 2.6). We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analyses

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐fatal myocardial infarction

None of the included studies reported non‐fatal myocardial infarction in the publications. One study in adults had data on non‐fatal myocardial infarction from the CSR (Home 2005). In this study, 0/292 participants in the insulin glargine group compared with 0/293 participants in the NPH insulin group experienced a non‐fatal myocardial infarction (low‐certainty evidence; Analysis 2.7). We judged the overall risk of bias for this outcome as 'low'.

Non‐fatal stroke

None of the included studies reported non‐fatal stroke in the publications. One study in adults had data on cerebral ischaemia from the CSR (Home 2005). In this study, 0/292 participants in the insulin glargine group compared with 1/293 participants in the NPH insulin group experienced cerebral ischaemia (low‐certainty evidence; Analysis 2.8). We judged the overall risk of bias for this outcome as 'low'.

End‐stage renal disease

None of the studies reported on end‐stage renal disease.

Blindness

None of the studies reported on blindness.

Serious adverse events

Eight studies reported data on SAEs (Bolli 2009; Chase 2008; Fulcher 2005; Home 2005; Liu 2016; PRESCHOOL; Ratner 2000; Schober 2002).

A total of 109/1131 participants (9.6%) in the insulin glargine group compared with 110/1098 participants (10.0%) in the NPH insulin group experienced SAEs. There was no evidence of a difference in SAEs (RR 1.08, 95% CI 0.63 to 1.84; P = 0.79; 8 studies, 2229 participants; moderate‐certainty evidence; Analysis 2.9). The 95% prediction interval ranged between 0.22 and 5.21. We judged the overall risk of bias for this outcome as 'low'.

One study did not mention SAEs in the publication (Porcellati 2004). One study reported SAEs other than hypoglycaemia in the main publication (Fulcher 2005): 5/62 participants (8.0%) in the insulin glargine group compared with 3/63 participants (4.7%) in the NPH insulin group experienced a SAE. From the CSR, the number of participants experiencing any SAE was reported and used in the meta‐analysis (Fulcher 2005). Three other studies contributed with data from additional sources (Liu 2016; PRESCHOOL; Ratner 2000). Three studies reported SAEs in the main publication (Bolli 2009; Home 2005; Schober 2002).

Subgroup analysis and sensitivity analysis

Analysing studies including only adults indicated an RR of 0.99, 95% CI 0.72 to 1.35; Analysis 2.9. Analysing studies including only children indicated an RR of 1.02, CI 95% CI 0.28 to 3.64; Analysis 2.9. The test for subgroup differences did not indicate interaction (P = 0.96).

Restricting the analysis to only published data indicated an RR of 1.11, 95% CI 0.11 to 2.70; Analysis 2.10. Restricting the analysis to only unpublished data indicated an RR of 1.10, 95% CI 0.46 to 2.60; Analysis 2.10. The test for subgroup differences did not indicate interaction (P = 0.99).

Sensitivity analysis excluding the largest study and the longest study indicated an RR of 1.15, 95% CI 0.58 to 2.30 (Home 2005).

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Diabetic ketoacidosis

We could retrieve data on diabetic ketoacidosis from seven studies (Chase 2008; Fulcher 2005; Home 2005; Liu 2016; PRESCHOOL; Ratner 2000; Schober 2002). Three studies reported ketoacidosis in their full‐text publications (Chase 2008; Liu 2016; Schober 2002).

A total of 6/1046 participants (0.6%) had ketoacidosis in the insulin glargine group compared with 8/1008 participants (0.1%) in the NPH insulin group. There was no evidence of a difference in diabetic ketoacidosis (RR 0.53, 95% CI 0.19 to 1.44; P = 0.21; 7 studies, 2054 participants; Analysis 2.11). We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analyses

Analysing diabetic ketoacidosis in only adults indicated an RR of 1.00, 95% CI 0.11 to 9.58; Analysis 2.11. Analysing diabetic ketoacidosis in only children indicated an RR of 0.45, 95% CI 0.15 to 1.39, Analysis 2.11. The test for subgroup differences did not indicate interaction (P = 0.53).

Analysing only published data indicated that 4/366 participants (1.1%) in the insulin glargine group compared with 8/319 participants (2.5%) in the NPH insulin group experienced diabetic ketoacidosis. The RR was 0.39, 95% CI 0.11 to 1.31; Analysis 2.12. Analysing only unpublished data indicated that 2/680 participants (0.3%) in the insulin glargine group compared with 3/689 participants (0.4%) in the NPH insulin group experienced diabetic ketoacidosis. The RR was 1.01, 95% CI 0.18 to 5.77; Analysis 2.12. The test for subgroup differences did not indicate interaction (P = 0.38).

Sensitivity analysis excluding the largest study and the longest study indicated an RR of 0.43, 95% CI 0.16 to 1.17 (Home 2005).

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐serious adverse events

Eight studies reported data on non‐serious adverse events (Bolli 2009; Chase 2008; Fulcher 2005; Home 2005; Liu 2016; PRESCHOOL; Ratner 2000; Schober 2002).

A total of 792/1131 participants (70.0%) in the insulin glargine group compared with 747/1098 (68.0%) participants in the NPH insulin group experienced a non‐serious adverse event. There was no evidence of a difference in non‐serious adverse events (RR 1.01, 95% CI 0.96 to 1.06; P = 0.72; 8 studies, 2229 participants; Analysis 2.13). The 95% prediction interval ranged between 0.95 and 1.07. We judged the overall risk of bias for this outcome as 'some concerns'.

One study did not mention adverse events in the publication (Porcellati 2004).

Subgroup analysis and sensitivity analysis

Analysing studies including only adults indicated an RR of 1.01, 95% CI 0.95 to 1.07; Analysis 2.13. Analysing studies including only children indicated an RR of 1.02, CI 95% CI 0.93 to 1.12; Analysis 2.13. The test for subgroup differences did not indicate interaction (P = 0.81).

Restricting the analysis to only published data indicated an RR of 1.00, 95% CI 0.94 to 1.05, Analysis 2.14. Restricting the analysis to only unpublished data indicated an RR of 1.03, 95% CI 0.94 to 1.14, Analysis 2.14. The test for subgroup differences did not indicate interaction (P = 0.53).

Sensitivity analysis excluding the largest study and the longest study indicated an RR of 1.00, 95% CI 0.95 to 1.06 (Home 2005).

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Withdrawals due to adverse events

A total of 11/1130 participants (1%) in the insulin glargine group compared with 9/1100 participants (0.8%) in the NPH insulin group withdrew because of adverse events. There was no evidence of a difference in withdrawals due to adverse events (RR 0.80, 95% CI 0.24 to 2.81; P = 0.76; 8 studies, 2130 participants; Analysis 2.15). The 95% prediction interval ranged between 0.07 and 10.27. We judged the overall risk of bias for this outcome as 'low' (data not shown).

Nocturnal hypoglycaemia

We could retrieve data on nocturnal hypoglycaemia from seven studies (Chase 2008; Fulcher 2005; Home 2005; Liu 2016; PRESCHOOL; Ratner 2000; Schober 2002).

Four studies reported severe nocturnal hypoglycaemia in the CSRs (Chase 2008; Fulcher 2005; Home 2005; Ratner 2000) and two studies reported severe nocturnal hypoglycaemia in the publications (PRESCHOOL; Schober 2002). A total of 69/938 participants (7.4%) in the insulin glargine group compared with 83/955 participants (8.7%) in the NPH insulin group experienced severe nocturnal hypoglycaemia. There was no evidence of a difference in severe nocturnal hypoglycaemia (RR 0.83, 95% CI 0.62 to 1.12; P = 0.23; 6 studies, 1893 participants; moderate‐certainty evidence; Analysis 2.19). We judged the overall risk of bias for this outcome as 'low'.

One study only reported frequency of nocturnal hypoglycaemia for the last month of treatment and not for the whole intervention period (12 months): there were 1.2 (SD 0.2) episodes/patient‐month in the insulin glargine group compared with 3.2 (SD 0.3) episodes/patient‐month in the NPH insulin group (Porcellati 2004). One study reported that there was no statistically significant change in nocturnal hypoglycaemia between the intervention groups (Bolli 2009). Five of the studies reported the number of participants with nocturnal hypoglycaemia in the publications (Fulcher 2005; Home 2005; Liu 2016; PRESCHOOL; Schober 2002). For two studies, we retrieved data from other sources (Chase 2008; Ratner 2000). Chase 2008 reported that no statistically significant change between the intervention groups was identified. Ratner 2000 reported nocturnal hypoglycaemia with confirmed blood glucose < 2 mmol/L and not just hypoglycaemia occurring at night as defined in the method section of the publication. In the CSR of this study, two different definitions of nocturnal hypoglycaemia were stated: hypoglycaemia at night and hypoglycaemia at night with blood glucose < 2 mmol/L. Nocturnal hypoglycaemia was reported for three different time periods in the CSR (after one month, from two months to the end of study, for the entire study period). From the tables in the CSR, it was apparent that the only analysis showing a statistically significant benefit of insulin glargine was nocturnal hypoglycaemia with confirmed blood glucose < 2 mmol/L from two months until the end of the study. This definition and time period were the ones reported in the full‐text publication.

A total of 713/1045 participants (68.2%) in the insulin glargine group compared with 693/1009 participants (68.7%) in the NPH insulin group experienced any nocturnal hypoglycaemia. There was no evidence of a difference in any nocturnal hypoglycaemia (RR 1.00, 95% CI 0.96 to 1.05; P = 0.96; 7 studies, 1054 participants; Analysis 2.16). One study investigated mild nocturnal hypoglycaemia as reported in the CSR (Fulcher 2005): 39/62 participants (62.9%) in the insulin glargine group compared with 47/63 participants (74.6%) in the NPH insulin group experienced mild nocturnal hypoglycaemia (RR 0.84, 95% CI 0.66 to 1.07; Analysis 2.17). Symptomatic nocturnal hypoglycaemia with or without blood glucose validation was reported in four studies (Fulcher 2005; Home 2005; Liu 2016; PRESCHOOL). There was no evidence of a difference in symptomatic nocturnal hypoglycaemia (RR 0.93, 95% CI 0.82 to 1.05; P = 0.26; 4 studies, 996 participants; Analysis 2.18). No study reported on asymptomatic nocturnal hypoglycaemia. Home 2005 reported that the proportion of participants who experienced nocturnal hypoglycaemia confirmed by a blood glucose level < 2.8 mmol/L and < 2.0 mmol/L did not differ significantly between interventions. We judged the overall risk of bias for all these outcomes except for severe nocturnal hypoglycaemia as 'some concerns' (data not shown).

Subgroup analysis and sensitivity analysis

Analysing studies for any nocturnal hypoglycaemia including only adults indicated an RR of 0.99, 95% CI 0.92 to 1.06; Analysis 2.16. Analysing studies for any nocturnal hypoglycaemia including only children indicated an RR of 1.01, CI 95% CI 0.95 to 1.08; Analysis 2.16. The test for subgroup differences did not indicate interaction (P = 0.65).

Analysing studies for symptomatic nocturnal hypoglycaemia including only adults indicated an RR of 0.97, 95% CI 0.88 to 1.08; Analysis 2.18. Analysing studies for symptomatic nocturnal hypoglycaemia including only children indicated an RR of 0.74, 95% CI 0.55 to 1.00; Analysis 2.18. The test for subgroup differences did not indicate interaction (P = 0.09).

Analysing studies for severe nocturnal hypoglycaemia including only adults indicated an RR of 0.87, 95% CI 0.60 to 1.27; Analysis 2.19. Analysing studies for severe nocturnal hypoglycaemia including only children indicated an RR of 0.77, 95% CI 0.47 to 1.25; Analysis 2.19. The test for subgroup differences did not indicate interaction (P = 0.68).

Restricting the analysis to only published data for any nocturnal hypoglycaemia indicated an RR of 1.00, 95% CI 0.95 to 1.06; Analysis 2.20. Restricting the analysis to only unpublished data for any nocturnal hypoglycaemia indicated an RR of 1.00, 95% CI 0.91 to 1.08; Analysis 2.20. The test for subgroup differences did not indicate interaction (P = 0.86).

Sensitivity analysis of any nocturnal hypoglycaemia excluding the largest study and the longest study indicated an RR of 1.00, 95% CI 0.95 to 1.05 (Home 2005).

Restricting the analysis to only published data for symptomatic nocturnal hypoglycaemia indicated an RR of 0.87, 95% CI 0.67 to 1.12; Analysis 2.21. Analysing only unpublished data for symptomatic nocturnal hypoglycaemia indicated an RR of 0.94, 95% CI 0.80 to 1.10; Analysis 2.21.

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Mild/moderate hypoglycaemia

We could retrieve data on mild/moderate hypoglycaemia from seven studies (Chase 2008; Fulcher 2005; Home 2005; Liu 2016; PRESCHOOL; Ratner 2000; Schober 2002).

A total of 951/1045 participants (91.0%) in the insulin glargine group compared with 898/1009 participants (89.0%) in the NPH insulin group experienced mild/moderate hypoglycaemia. There was no evidence of a difference in mild/moderate hypoglycaemia (RR 1.02, 95% CI 1.00 to 1.04; P = 0.09; 7 studies, 2054 participants; Analysis 2.22). We judged the overall risk of bias for this outcome as 'some concerns'.

One study only reported frequency of mild hypoglycaemia for the last month of treatment and not for the whole intervention period (12 months): there were 7.2 (SD 0.5) episodes/patient‐month in the insulin glargine group compared with 13.2 (SD 0.5) episodes/patient‐month in the NPH insulin group (Porcellati 2004). One study reported that there was no statistically significant change in hypoglycaemia between the intervention groups (Bolli 2009). Five studies reported mild/moderate hypoglycaemia in a format making the data suitable for meta‐analysis (Chase 2008; Home 2005; Liu 2016; PRESCHOOL; Schober 2002). For two studies, we retrieved the data from additional sources (Fulcher 2005; Ratner 2000).

Subgroup analysis and sensitivity analysis

Analysing studies including only adults indicated an RR of 1.02, 95% CI 0.99 to 1.06; Analysis 2.22. Analysing studies including only children indicated an RR of 1.01, CI 95% CI 0.99 to 1.04; Analysis 2.22. The test for subgroup differences did not indicate interaction (P = 0.68).

Restricting the analysis to only published data indicated an RR of 1.02, 95% CI 1.00 to 1.05; Analysis 2.23. Restricting the analysis to only unpublished data indicated an RR of 1.01, 95% CI 0.98 to 1.04; Analysis 2.23. The test for subgroup differences did not indicate interaction (P = 0.78).

Sensitivity analysis excluding the largest study and the longest study indicated an RR of 1.01, 95% CI 0.99 to 1.04 (Home 2005).

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Socioeconomic effects

We retrieved data for socioeconomic effects from CSRs of three studies (Fulcher 2005; Home 2005; Schober 2002). No studies reported an estimate of the costs of the intervention during the study period. One study had evaluated economic effects, but the supplemental CSR with these data could unfortunately not be retrieved (Ratner 2000). In the CSR for Fulcher 2005, it was reported that very few participants (three in each group) reported a loss of income because of diabetes during the treatment period. Approximately 30 participants in each intervention group reported seeking medical advice (ambulatory care) once or more during the treatment period (Fulcher 2005). Home 2005 could not evaluate all participants for economic data: 6/275 participants (2.1%) changed from employment status to non‐employment status during the study in the insulin glargine group compared with 7/265 participants (2.6%) in the NPH insulin group. Of the participants employed at baseline, 16/287 participants (7.5%) in the insulin glargine group compared with 23/283 participants (10.8%) in the NPH insulin group lost time for work during the study. Reasons for these changes during the study were not reported. Schober 2002 reported that nine of the caregivers (7.5%) employed at baseline had lost time for work during the study in the insulin glargine group compared with 12 of the caregivers (10.3%) in the NPH insulin group.

HbA1c

We retrieved data on HbA1c levels from all studies (Bolli 2009; Chase 2008; Fulcher 2005; Home 2005; Liu 2016; Porcellati 2004; PRESCHOOL; Ratner 2000; Schober 2002).

There was no evidence of a difference in HbA1c (MD 0.02%, 95% CI ‐0.1 to 0.1; P = 0.59; 9 studies, 2285 participants; moderate‐certainty evidence; Analysis 2.24). The 95% prediction interval ranged between ‐0.5% and 0.5%. We judged the overall risk of bias for this outcome as 'low'.

One study reported HbA1c at the end of follow‐up as adjusted least square means in the publication (Fulcher 2005). However, in this study, HbA1c at baseline was higher in the participants randomised to NPH insulin compared with insulin glargine (9.2% (SD 1.1) in the insulin glargine group compared with 9.7% (SD 1.3) in the NPH insulin group). In the CSR of this study, data with change from baseline were provided which we included in the meta‐analysis. Chase 2008 reported HbA1c for completers of the study only. However, in the CSR, HbA1c was reported for completers and for the intention‐to‐treat population.

Subgroup and sensitivity analysis

Five studies reported HbA1c in adults with a MD of ‐0.01%, 95% CI ‐0.2 to 0.1; Analysis 2.24. Four studies including only children reported HbA1c with a MD of 0.03%, 95% CI ‐0.1 to 0.2; Analysis 2.24. The test for subgroup differences did not indicate interaction (P = 0.67).

Analysing only published data indicated HbA1c with a MD of 0.02%, 95% CI ‐0.1 to 0.1, Analysis 2.25. Analysing only unpublished data indicated HbA1c with a MD of ‐0.04%, 95% CI ‐0.3 to 0.2; Analysis 2.25. The test for subgroup differences did not indicate interaction (P = 0.60).

Sensitivity analysis excluding the largest study and the longest study indicated a MD in HbA1c of 0.0%, 95% CI ‐0.1 to 0.1 (Home 2005).

All studies, except one had received funding from the pharmaceutical industry (Porcellati 2004). Porcellati 2004 applied NPH four times a day. Excluding this study from the analysis indicated a MD in HbA1c of 0.02%, 95% CI ‐0.1 to 0.1.

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Combined HbA1c and severe hypoglycaemia

None of the studies reported on combined HbA1c and severe hypoglycaemia.

Insulin detemir compared with insulin glargine

For an overview of main results for this comparison, see summary of findings Table 3.

Two studies compared insulin detemir with insulin glargine (Heller 2009; Pieber 2007). A total of 769 participants were randomised, 461 participants to insulin detemir and 308 participants to insulin glargine (Table 1). Both studies were published as full‐text articles in English. However, we retrieved unpublished information on outcomes for both studies from additional sources. One study administered insulin detemir once daily (evening dose). If necessary, a second dose could be administered in the morning (Heller 2009). One study applied insulin detemir twice daily (Pieber 2007). Insulin glargine was given once daily (evening dose) in both studies. Fast‐acting insulin was insulin aspart in both studies. Both studies included adults with T1DM. The duration of the intervention varied from 24 weeks to 52 weeks (see Table 1). Both studies were sponsored by Novo Nordisk.

Primary outcomes

All‐cause mortality

We retrieved data on all‐cause mortality from the clinical study synopsis of both studies. Heller 2009 reported that 0/299 participants died in the insulin detemir group compared with 1/144 participants (0.7%) in the insulin glargine group and Pieber 2007 reported that 0/161 participants died in the insulin detemir group compared with 0/159 participants in the insulin glargine group (low‐certainty evidence; Analysis 3.1). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Health‐related quality of life

No study reported scales evaluating health‐related quality of life. One study used the Diabetes Treatment Satisfaction Questionnaire and pain perception (Pieber 2007). One study used the Insulin Treatment Satisfaction Questionnaire (Heller 2009). Both treatment satisfaction questionnaires were reported in CSRs.

Severe hypoglycaemia

Heller 2009 reported the mean number of hypoglycaemic episodes in the insulin detemir group to be 146 in 299 participants in the insulin detemir group compared with 53 in 144 participants in the insulin glargine group. However, we could retrieve the number of participants experiencing one or more severe hypoglycaemic episodes from the associated CSR. Pieber 2007 reported severe hypoglycaemia in the publication.

A total of 57/460 participants (12.4%) in the insulin detemir group compared with 35/303 participants (11.6%) in the insulin glargine group experienced severe hypoglycaemia. There was no evidence of a difference in severe hypoglycaemia (RR 0.59, 95% CI 0.13 to 2.63; P = 0.49; 2 studies, 763 participants; very low‐certainty evidence; Analysis 3.2; Figure 5). We judged the overall risk of bias for this outcome as 'low'.

Figure 5

Severe hypoglycaemia

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Hypoglycaemia reported as a serious adverse event

A total of 13/460 participants (2.8%) in the insulin detemir group compared with 5/303 participants (1.7%) in the insulin glargine group experienced hypoglycaemia as a SAE. There was no evidence of a difference in hypoglycaemia reported as a SAE (RR 1.16, 95% CI 0.14 to 9.48; P = 0.89; 2 studies, 763 participants; Analysis 3.4). We judged the overall risk of bias for this outcome as 'low' (data not shown).

Subgroup and sensitivity analysis

Analysis according to publication status indicated interaction (P = 0.02; Analysis 3.3). However, this has to be interpreted with caution because the 95% CIs slightly overlapped. The remaining subgroup and sensitivity analyses could not be performed due to lack of data (Appendix 20).

Secondary outcomes

Cardiovascular mortality

We could retrieve data on cardiovascular mortality from additional sources for both studies. Heller 2009 reported that 0/299 participants died due to cardiovascular disease in the insulin detemir group compared with 1/144 participants (0.7%) in the insulin glargine group and Pieber 2007 reported that 0/161 participants in the insulin detemir group compared with 0/159 participants in the insulin glargine group died (Analysis 3.5). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐fatal myocardial infarction

Heller 2009 reported in the CSR that 1/299 participants (0.3%) in the insulin detemir group compared with 1/144 participants (0.7%) in the insulin glargine group experienced a non‐fatal myocardial infarction (low‐certainty evidence; Analysis 3.6). We judged the overall risk of bias for this outcome as 'low'.

Non‐fatal stroke

Heller 2009 reported in the CSR that 2/299 participants (0.6%) in the insulin detemir group compared with 0/144 participants in the insulin glargine group experienced a non‐fatal stroke (low‐certainty evidence; Analysis 3.7). We judged the overall risk of bias for this outcome as 'low'.

End‐stage renal disease

None of the studies for reported on end‐stage renal disease.

Blindness

None of the studies reported on blindness.

Serious adverse events

Both studies reported SAEs in the publications. A total of 49/460 participants (10.7%) in the insulin detemir group compared with 18/303 participants (5.9%) in the insulin glargine group experienced a SAE. There was no evidence of a difference in SAEs (RR 1.72, 95% CI 0.91 to 3.23; P = 0.24; 2 studies, 763 participants; low‐certainty evidence; Analysis 3.8). Analysing data in a fixed‐effect model showed beneficial effects of insulin glargine (RR 1.79, 95% CI 1.04 to 3.08; P = 0.04). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Diabetic ketoacidosis

Heller 2009 reported in the CSR that 1/299 participants (0.3%) in the insulin detemir group compared with 0/144 participants in the insulin glargine group experienced ketoacidosis (Analysis 3.9). We judged the overall risk of bias for this outcome as 'low'.

Non‐serious adverse events

We could retrieve data on non‐serious adverse events from both studies. Heller 2009 reported adverse events in the publication; we retrieved data for Pieber 2007 data from additional sources.

A total of 394/460 participants (85.7%) in the insulin detemir group compared with 250/303 participants (82.5%) in the insulin glargine group reported a non‐serious adverse event. There was no evidence of a difference in non‐serious adverse events (RR 1.01, 95% CI 0.93 to 1.09; 2 studies, 763 participants; Analysis 3.10). We judged the overall risk of bias for this outcome as 'some concerns'.

Subgroup and sensitivity analysis

Subgroup analysis according to published data compared with unpublished data did not indicate interaction (P = 0.28; Analysis 3.11). We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Withdrawals due to adverse events

A total of 9/460 participants (2.0%) in the insulin detemir group compared with 5/303 participants (1.7%) in the insulin glargine group withdrew because of adverse events. There was no evidence of a difference in withdrawals due to adverse events (RR 1.06, 95% CI 0.31 to 3.67; P = 0.92; 2 studies, 763 participants; Analysis 3.12).

Nocturnal hypoglycaemia

We could retrieve data on nocturnal hypoglycaemia from both studies. Pieber 2007 reported nocturnal hypoglycaemia in the publication. We retrieved data for Heller 2009 from additional sources. Both studies defined nocturnal hypoglycaemia as an episode occurring between 23.00 and 06.00.

A total of 27/460 participants (5.9%) in the insulin detemir group compared with 15/303 participants (5.0%) in the insulin glargine group experienced severe nocturnal hypoglycaemia. There was no evidence of a difference in severe nocturnal hypoglycaemia (RR 0.55, 95% CI 0.06 to 5.12; P = 0.60; 2 studies, 763 participants; very low‐certainty evidence; Analysis 3.16). We judged the overall risk of bias for this outcome as 'low'.

Pieber 2007 reported data on nocturnal hypoglycaemia according to different definitions in the publication. Heller 2009 reported there were no significant differences between the interventions in the risk of having a nocturnal hypoglycaemic episode, but the number of participants with an event in each intervention group was not provided in the publication. However, we could obtain these data from the CSR. Both studies had analysed nocturnal hypoglycaemia according to the same subclassifications: there was no evidence of a difference in any nocturnal hypoglycaemia (RR 1.01, 95% CI 0.93 to 1.09; P = 0.84; 2 studies, 763 participants; Analysis 3.13), in confirmed nocturnal hypoglycaemia (plasma glucose < 3.1 mmol/L and no assistance; RR 1.01, 95% CI 0.92 to 1.10; P = 0.90; 2 studies, 763 participants; Analysis 3.14); and in symptomatic nocturnal hypoglycaemia (plasma glucose ≤ 3.1 mmol/L or no plasma glucose, no assistance required; RR 1.02, 95% CI 0.81 to 1.29; P = 0.85; 2 studies, 763 participants; Analysis 3.15). We judged the overall risk of bias for all these outcomes except for severe nocturnal hypoglycaemia as 'some concerns' (data not shown).

Subgroup and sensitivity analysis

Analysis could only be performed according to published data compared with unpublished data: none of the definitions of nocturnal hypoglycaemia indicated interactions. We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Mild/moderate hypoglycaemia

We could retrieve data on mild/moderate hypoglycaemia from both studies. Pieber 2007 reported mild/moderate hypoglycaemia in the publication. Heller 2009 reported that the overall risk of having a hypoglycaemic episode during the treatment period was similar between the insulin detemir and the insulin glargine group with a relative risk (insulin detemir/insulin glargine) of 0.94; P = 0.57. The number of participants with mild/moderate hypoglycaemia was not reported in this publication. However, we could retrieve data from the CSR.

A total of 404/460 participants (87.8%) in the insulin detemir group compared with 243/303 participants (80.2%) in the insulin glargine group experienced mild/moderate hypoglycaemia. There was no evidence of a difference in mild/moderate hypoglycaemia (RR 1.04, 95% CI 0.94 to 1.14; P = 0.44; 2 studies, 763 participants; Analysis 3.17). We judged the overall risk of bias for this outcome as 'some concerns'.

Subgroup and sensitivity analysis

Analysis according to published data compared with unpublished data did not indicate interaction. We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Socioeconomic effects

No studies reported the costs of the intervention during the study period. One study published economic data based on simulation cohorts from a US healthcare system perspective (Pieber 2007).

HbA1c

We could retrieve data on HbA1c levels from both studies. There was no evidence of a difference in HbA1c (MD ‐0.01%, 95% CI ‐0.1 to 0.1; P = 0.89; 2 studies, 763 participants; low‐certainty evidence; Analysis 3.18). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Combined HbA1c and severe hypoglycaemia

Heller 2009 reported that HbA1c ≤ 7% was achieved without major hypoglycaemia during the last month of treatment for 91/285 participants (31.9%) in the insulin detemir group compared with 39/135 participants (28.9%) in the insulin glargine group (RR 1.11, 95% CI 0.81 to 1.51; P = 0.53; Analysis 3.19). Pieber 2007 did not report numerical data, but stated that the adjustment for HbA1c showed that the reduced risk of hypoglycaemia with insulin detemir was not due to differences in glycaemic control. We judged the overall risk of bias for this outcome as 'low' (data not shown).

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Insulin degludec compared with insulin detemir

For an overview of main results for this comparison, see summary of findings Table 4.

Two studies compared insulin degludec with insulin detemir (BEGIN Young; Davies 2014). A total of 806 participants were randomised, 477 participants to insulin degludec and 329 participants to insulin detemir (see 'Overview of study populations' Table 1). One study included children (BEGIN Young). The mean age of the children was 10 years. Both studies were published as full‐text articles in English. However, for both studies we could retrieve additional information on outcomes from additional sources. Both studies applied insulin degludec once daily and insulin detemir once or twice daily. Both studies applied insulin aspart as fast‐acting insulin. The duration of the intervention was 26 weeks in both studies and both studies had an extension period of 26 weeks. Both studies were sponsored by the same pharmaceutical company (Novo Nordisk).

Primary outcomes

All‐cause mortality

Both studies reported data on all‐cause mortality (BEGIN Young; Davies 2014). No participant died (0/475 participants in the insulin degludec group compared with 0/327 participants in the insulin detemir group; low‐certainty evidence; Analysis 4.1). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Health‐related quality of life

We retrieved data on health‐related quality of life from additional sources (Davies 2014). The applied questionnaire was the SF‐36. There was no evidence of a difference in health‐related quality of life for the physical health score (MD ‐0.60, 95% CI ‐1.83 to 0.63; P = 0.34; 1 study, 454 participants; low‐certainty evidence; Analysis 4.2) The health‐related quality of life for the mental health score favoured insulin detemir (MD ‐3.00, 95% CI ‐4.44 to ‐1.56; P < 0.001; 1 study, 454 participants; low‐certainty evidence; Analysis 4.2). The minimal important difference for the physical component score is two to three points and for the mental component score three points. We judged the overall risk of bias for this outcome as 'some concerns'.

Severe hypoglycaemia

Both studies reported data on severe hypoglycaemia in the publications. In the insulin degludec group, 63/475 participants (13.3%) experienced severe hypoglycaemia compared with 40/327 participants (12.2%) in the insulin detemir group. There was no evidence of a difference in severe hypoglycaemia (RR 1.17, 95% CI 0.81 to 1.69; P = 0.42; 2 studies, 802 participants; low‐certainty evidence; Analysis 4.3; Figure 6). We judged the overall risk of bias for this outcome as 'low'.

Figure 6

Severe hypoglycaemia

Subgroup and sensitivity analysis

Subgroup analysis including only adults compared with studies including only children did not indicate interaction (P = 0.51; Analysis 4.3). We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Hypoglycaemia reported as a serious adverse event

A total of 15/475 participants (3.2%) in the insulin degludec group compared with 10/327 participants (3.1%) in the insulin detemir group had a SAE due to hypoglycaemia. There was no evidence of a difference in SAEs (RR 0.92, 95% CI 0.37 to 2.32; P = 0.86; 2 studies, 802 participants; Analysis 4.4). We judged the overall risk of bias for this outcome as 'low' (data not shown).

The test for subgroup differences comparing adults with children did not indicate interaction (P = 0.27).

BEGIN Young in the SAE list of the CSR stated that 2/174 participants (1.1%) in the insulin degludec group compared with 4/175 participants (2.3%) in the insulin detemir experienced a hypoglycaemic seizure and 1/174 participants (0.6%) in the insulin degludec group compared with 1/175 participants (0.6%) in the insulin detemir group experienced hypoglycaemic unconsciousness. Davies 2014 in the SAE list of the CSR reported that 3/301 participants (1.0%) in the insulin degludec group compared with 1/152 participants (0.7%) in the insulin detemir group experienced a hypoglycaemic coma and 3/301 participants (1.0%) compared with 1/152 participants (0.7%) experienced hypoglycaemic unconsciousness.

Secondary outcomes

Cardiovascular mortality

Both studies reported data on cardiovascular mortality. No participant died (0/475 participants in the insulin degludec group compared with 0/327 participants in the insulin detemir group; Analysis 4.5). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐fatal myocardial infarction

Davies 2014 reported that no participant experienced a non‐fatal myocardial infarction (0/301 participants in the insulin degludec group compared with 0/152 participants in the insulin detemir group; ; low‐certainty evidence; Analysis 4.6). We retrieved these data from additional sources. We judged the overall risk of bias for this outcome as 'low'.

Non‐fatal stroke

Davies 2014 reported that no participant experienced a non‐fatal stroke (0/301 participants in the insulin degludec group compared with 0/152 participants in the insulin detemir group; Analysis 4.7; low‐certainty evidence). We retrieved these data from additional sources. We judged the overall risk of bias for this outcome as 'low'.

End‐stage renal disease

Davies 2014 reported that 0/301 participants in the insulin degludec group compared with 0/152 participants in the insulin detemir group experienced end‐stage renal disease (Analysis 4.8). We retrieved these data from additional sources. We judged the overall risk of bias for this outcome as 'low'.

Blindness

Davies 2014 reported that no participant experienced blindness (0/301 participants in the insulin degludec group compared with 0/152 participants in the insulin detemir group; Analysis 4.9). These data were retrieved from additional sources. We judged the overall risk of bias for this outcome as 'low'.

Serious adverse events

Both studies reported SAEs. In the insulin degludec group, 41/475 participants (8.6%) compared with 24/327 participants (7.3%) in the insulin detemir group experienced a SAE. There was no evidence of a difference in SAEs (RR 1.25, 95% CI 0.76 to 2.05; P = 0.38; 2 studies, 802 participants; low‐certainty evidence; Analysis 4.10). We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analysis

Subgroup analysis including only adults compared with studies including only children did not indicate interaction (P = 0.63; Analysis 4.10). We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Diabetic ketoacidosis

None of the studies reported on ketoacidosis in the publications. However, we retrieved data on diabetic ketoacidosis from additional sources. A total of 2/475 participants (0.4%) in the insulin degludec group compared with 0/327 participants in the insulin detemir group experienced diabetic ketoacidosis (Analysis 4.11). Both participants experiencing diabetic ketoacidosis were children (BEGIN Young). We judged the overall risk of bias for this outcome as 'low'.

We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐serious adverse events

BEGIN Young reported the number of children with non‐serious adverse events in the publication. For Davies 2014, we retrieved this information from additional sources. A total of 380/475 participants (80%) in the insulin degludec group compared with 269/327 participants (82.3%) in the insulin detemir group experienced a non‐serious adverse event. There was no evidence of a difference in non‐serious adverse events (RR 1.02, 95% CI 0.96 to 1.08; P = 0.48; 2 studies, 802 participants; Analysis 4.12). We judged the overall risk of bias for this outcome as 'some concerns'.

Subgroup and sensitivity analysis

Analyses including only adults compared with studies including only children and analyses comparing only published data with only unpublished data did not indicate interaction (P = 0.53; Analysis 4.12). We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Withdrawals due to adverse events

A total of 5/475 participants (1.1%) in the insulin degludec group compared with 1/327 participants (0.3%) in the insulin detemir group withdrew because of adverse events. There was no evidence of a difference in withdrawals due to adverse events (RR 2.32, 95% CI 0.38 to 14.18; P = 0.36; 2 studies, 802 participants; Analysis 4.13). We judged the overall risk of bias for this outcome as 'low' (data not shown).

Nocturnal hypoglycaemia

Both studies reported data on nocturnal hypoglycaemia. None of the studies reported on any nocturnal hypoglycaemia.

Severe nocturnal hypoglycaemia was reported in the CSRs of both studies. A total of 17/475 participants in the insulin degludec group (3.6%) compared with 10/327 participants (3.1%) in the insulin detemir group experienced severe nocturnal hypoglycaemia. There was no evidence of a difference in severe nocturnal hypoglycaemia (RR 1.12, 95% CI 0.51 to 2.46; P = 0.77; 2 studies, 802 participants; low‐certainty evidence; Analysis 4.18). We judged the overall risk of bias for this outcome as 'low'.

We retrieved data on nocturnal hypoglycaemia confirmed with blood glucose measurements. There was no evidence of a difference in confirmed nocturnal hypoglycaemia (RR 1.04, 95% CI 0.94 to 1.15; P = 0.40; 2 studies, 802 participants; Analysis 4.14). From the CSRs of both studies, data for mild documented nocturnal hypoglycaemia (plasma glucose ≤ 3.9 mmol/L, able to self‐treat) were available. There was no evidence of a difference in mild documented nocturnal hypoglycaemia (RR 0.97, 95% CI 0.86 to 1.10; P = 0.67; 2 studies, 802 participants; Analysis 4.15). There was no evidence of a difference in symptomatic nocturnal hypoglycaemia without blood glucose measurements (RR of 0.72, 95% CI 0.15 to 3.59; P = 0.69; 2 studies, 802 participants; Analysis 4.16). There was no evidence of a difference in asymptomatic nocturnal hypoglycaemia (RR 0.91, 95% CI 0.80 to 1.03; P = 0.13; 2 studies, 802 participants; Analysis 4.17). We judged the overall risk of bias for all these outcomes except for severe nocturnal hypoglycaemia as 'some concerns' (data not shown).

Subgroup and sensitivity analysis

Subgroup analysis including only adults only compared with studies including only children did not indicate subgroup interaction (P = 0.82; Analysis 4.18). We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Mild/moderate hypoglycaemia

Both studies reported data on mild/moderate hypoglycaemia in the publications. There was no evidence of a difference in mild/moderate hypoglycaemia (RR 1.02, 95% CI 0.99 to 1.05; P = 0.17; 2 studies, 802 participants; Analysis 4.19). We judged the overall risk of bias for this outcome as 'some concerns'.

Subgroup and sensitivity analysis

Subgroup analysis including only adults compared with studies including only children did not indicate interaction (P = 0.85; Analysis 4.19). We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Socioeconomic effects

No studies reported the costs of the intervention during the study period. One study reported economic predictions of the interventions based on simulation cohorts in an UK setting of children and adolescents (BEGIN Young).

HbA1c

Both studies had data for HbA1c. BEGIN Young reported data until the end of the extension period in the publication and not until the end of the intervention period. However, we could retrieve these data from ClinicalTrials.gov. There was no evidence of a difference in HbA1c (MD 0.05%, 95% CI ‐0.1 to 0.2; P = 0.44; 2 studies, 804 participants; low‐certainty evidence; Analysis 4.20). We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analysis

Subgroup analyses including only adults only compared with studies including only children and only published data compared with only unpublished data did not indicate interactions (P = 0.42; Analysis 4.20). We did not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Combined HbA1c and severe hypoglycaemia

Davies 2014 reported the combined outcome HbA1c and severe hypoglycaemia in the CSR. At the end of the intervention period, a total of 116/292 participants (39.7%) in the insulin degludec group compared with 53/145 participants (36.6%) in the insulin detemir group achieved an HbA1c < 7% without severe hypoglycaemia during the last 12 weeks of treatment (RR 1.09, 95% CI 0.84 to 1.41; P = 0.53; Analysis 4.21). We judged the overall risk of bias for this outcome as 'low' (data not shown).

Insulin degludec compared with insulin glargine

For an overview of main results for this comparison, see summary of findings Table 5.

Four studies compared insulin degludec with insulin glargine (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; SWITCH 1; Urakami 2017). A total of 1477 participants were randomised, 895 participants to insulin degludec and 582 participants to insulin glargine (see Table 1). One study included children (Urakami 2017). The mean age of the children was 10.5 years. All studies were published in full text in English. However, for all studies, we could retrieve additional information on outcomes from additional sources. All studies applied insulin degludec once daily and insulin glargine once daily. Urakami 2017 applied insulin aspart or insulin lispro before meals. The remaining studies applied insulin aspart before meals. The duration of the intervention ranged from 26 weeks to 52 weeks. SWITCH 1 and Urakami 2017 had a cross‐over design; the remaining studies were parallel‐group RCTs. Because of carryover effects, we evaluated outcomes before cross‐over. In SWITCH 1, each of the two treatment periods consisted of a 16‐week titration period and a 16‐week maintenance period; only data for health‐related quality of life and HbA1c were available before cross‐over. Three of the studies were sponsored by Novo Nordisk (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; SWITCH 1); one study did not report the funding source (Urakami 2017).

All‐cause mortality

Two studies reported on all‐cause mortality (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1). For one study, we retrieved this information from additional sources (BEGIN Flex T1). SWITCH 1 reported that four deaths occurred. However, these data could not be included in the meta‐analysis because it was not reported if the deaths occurred before or after cross‐over.

All studies reporting all‐cause mortality were performed in adults. A total of 3/646 participants (0.5%) in the insulin degludec group compared with 1/327 participants (0.3%) in the insulin glargine group died. There was no evidence of a difference in all‐cause mortality (Peto OR 1.34, 95% CI 0.15 to 11.93; P = 0.79; 2 studies, 955 participants; very low‐certainty evidence; Analysis 5.1). We judged the overall risk of bias for this outcome as 'low'.

Subgroup analysis and sensitivity analysis

Analysis according to only published data compared with only unpublished data did not indicate interaction (P = 0.46; Analysis 5.2). The remaining subgroup and sensitivity analyses could not be performed due to lack of data (Appendix 20).

Health‐related quality of life

SWITCH 1 reported health‐related quality of life before cross‐over in the CSR. BEGIN Basal‐Bolus Type 1 reported health‐related quality of life in an appendix to the publication. Both studies applied the SF‐36 questionnaire. There was no evidence of a difference in health‐related quality of life (MD for physical health score ‐0.04 points, 95% CI ‐1.21 to 1.13; P = 0.94; 2 studies, 1042 participants; very low‐certainty evidence; Analysis 5.3; and MD of mental health score ‐0.09 points, 95% CI ‐1.03 to 0.85; P = 0.85; 2 studies, 1042 participants; very low‐certainty evidence; Analysis 5.4). The minimal important difference for the physical component score is two to three points and for the mental component score three points. We judged the overall risk of bias for this outcome as 'some concerns'.

Subgroup analysis and sensitivity analysis

Analysis according to only published data compared with only unpublished data did not indicate subgroup interaction. The remaining subgroup and sensitivity analyses could not be performed due to lack of data (Appendix 20).

Severe hypoglycaemia

We could evaluate severe hypoglycaemia for three studies (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; Urakami 2017). Two studies reported severe hypoglycaemia in the main publication (BEGIN Basal‐Bolus Type 1; Urakami 2017) and for one study we retrieved data from an appendix to the publication (BEGIN Flex T1).

A total of 79/646 participants (12.3%) in the insulin degludec group compared with 32/324 participants (9.9%) in the insulin glargine group reported severe hypoglycaemia. There was no evidence of a difference in severe hypoglycaemia (RR 1.22, 95% CI 0.82 to 1.82; P = 0.32; 3 studies, 970 participants; low‐certainty evidence; Analysis 5.5; Figure 7). We judged the overall risk of bias for this outcome as 'low'.

Figure 7

Severe hypoglycaemia

Subgroup analysis and sensitivity analysis

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Hypoglycaemia reported as a serious adverse event

A total of 49/1100 participants (4.5%) in the insulin degludec group compared with 44/784 participants (5.6%) in the insulin glargine group had a SAE due to hypoglycaemia. There was no evidence of a difference in hypoglycaemia reported as a SAE (RR 0.81, 95% CI 0.40 to 1.66; P = 0.57; 4 studies, 1884 participants; Analysis 5.6). We judged the overall risk of bias for this outcome as 'low' (data not shown).

Cardiovascular mortality

We could retrieve data on cardiovascular mortality from two studies through additional sources (BEGIN Flex T1; Urakami 2017). One study reported the cause of death in the main publication (BEGIN Basal‐Bolus Type 1). Only BEGIN Basal‐Bolus Type 1 reported any deaths due to cardiovascular disease. In this study, 2/472 participants (0.4%) in the insulin degludec group compared with 1/154 participants (0.6%) in the insulin glargine group died due to cardiovascular disease (Analysis 5.7). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐fatal myocardial infarction

We could retrieve data on non‐fatal myocardial infarction for three studies from additional sources (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; Urakami 2017). However, only one study reported any participant experiencing a non‐fatal myocardial infarction. In this study, 1/472 participants (0.2%) in the insulin degludec group compared with 0/154 participants in the insulin glargine group experienced a non‐fatal myocardial infarction ( low‐certainty evidence; Analysis 5.8). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐fatal stroke

We could retrieve data on non‐fatal stroke for two studies from CSRs (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1). For one study, data were provided by the study author (Urakami 2017). BEGIN Flex T1 reported no event (0/165 participants in the insulin degludec group compared with 0/161 participants in the insulin glargine group). Urakami 2017 also reported no event (0/9 participants in both intervention groups). BEGIN Basal‐Bolus Type 1 reported that 1/472 participants (0.2%) in the insulin degludec group compared with 0/154 participants in the insulin glargine group experienced cerebral ischaemia (low‐certainty evidence; Analysis 5.9). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

End‐stage renal disease

For one study, the study author provided information that no participant developed end‐stage renal disease (Urakami 2017). None of the other studies reported on end‐stage renal disease. We judged the overall risk of bias for this outcome as 'low' (data not shown).

Blindness

For one study, the study author provided information that no participant developed blindness (Urakami 2017). None of the other studies reported on blindness. We judged the overall risk of bias for this outcome as 'low' (data not shown).

Serious adverse events

Three studies reported SAEs (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; Urakami 2017). Two studies reported data in the publications (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1). For one study, the investigator reported that no participant experienced a SAE (Urakami 2017).

A total of 56/646 participants (8.7%) in the insulin degludec group compared with 25/324 participants (7.7%) in the insulin glargine group experienced serious adverse events. There was no evidence of a difference in SAEs (RR 0.92, 95% CI 0.58 to 1.46; P = 0.73; 3 studies, 970 participants; low‐certainty evidence; Analysis 5.10). We judged the overall risk of bias for this outcome as 'low'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Diabetic ketoacidosis

Three studies reported diabetic ketoacidosis (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; Urakami 2017). One study reported data in the publication (BEGIN Basal‐Bolus Type 1). For two studies, we retrieved data from additional sources (BEGIN Flex T1; Urakami 2017).

A total of 3/646 participants (0.5%) in the insulin degludec group compared with 3/324 participants (0.9%) in the insulin glargine group experienced diabetic ketoacidosis. There was no evidence of a difference in diabetic ketoacidosis (RR 0.57, 95% CI 0.05 to 6.89; P = 0.66; 3 studies, 970 participants; Analysis 5.11). We judged the overall risk of bias for this outcome as 'low'.

Subgroup analysis and sensitivity analysis

Analysis according to only published data compared with only unpublished data did not indicate subgroup interaction.

We could not perform the remaining subgroup and sensitivity analyses due to lack of data (Appendix 20).

Non‐serious adverse events

Three studies reported non‐serious adverse events (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; Urakami 2017). Two studies reported data in the publications (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1). For one study, the investigator reported that no participant experienced a non‐serious adverse event (Urakami 2017).

A total of 522/646 participants (80.8%) in the insulin degludec group compared with 244/324 participants (75.3%) in the insulin glargine group experienced a non‐serious adverse event. There was no evidence of a difference in non‐serious adverse events (RR 1.02, 95% CI 0.95 to 1.10; P = 0.52; 3 studies, 970 participants; Analysis 5.13). We judged the overall risk of bias for this outcome as 'some concerns'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Nocturnal hypoglycaemia

Three studies reported nocturnal hypoglycaemia (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; Urakami 2017). Two studies reported data for one or more nocturnal hypoglycaemic outcomes in the publications (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1). For one study, the investigator reported that no participant experienced nocturnal hypoglycaemia (Urakami 2017).

A total of 23/646 participants (3.6%) in the insulin degludec group compared with 8/324 participants (2.5%) in the insulin glargine group experienced severe nocturnal hypoglycaemia. There was no evidence of a difference in severe hypoglycaemia (RR 1.39, 95% CI 0.59 to 3.27; P = 0.46; 3 studies, 970 participants; low‐certainty evidence; Analysis 5.19). We judged the overall risk of bias for this outcome as 'low'.

A total of 464/646 participants (71.8%) in the insulin degludec group compared with 235/324 participants (72.5%) in the insulin glargine group experienced nocturnal hypoglycaemia. There was no evidence of a difference in nocturnal hypoglycaemia (RR 0.99, 95% CI 0.91 to 1.07; P = 0.76; 3 studies, 970 participants; Analysis 5.15). We retrieved data on mild nocturnal hypoglycaemia from additional sources. There was no evidence of a difference in mild nocturnal hypoglycaemia (RR 0.98, 95% CI 0.90 to 1.07; P = 0.63; 2 studies, 952 participants; Analysis 5.16). Asymptomatic nocturnal hypoglycaemia was reported in the CSRs of two studies. There was no evidence of a difference in asymptomatic nocturnal hypoglycaemia (RR 0.84, 95% CI 0.71 to 1.00; P = 0.05; 2 studies, 952 participants; Analysis 5.17). There was no evidence of a difference in symptomatic nocturnal hypoglycaemia (RR 1.22, 95% CI 0.72 to 2.07; P = 0.46; 2 studies, 952 participants; Analysis 5.18). We judged the overall risk of bias for all these outcomes except for severe nocturnal hypoglycaemia as 'some concerns' (data not shown).

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Mild/moderate hypoglycaemia

Three studies reported on mild/moderate hypoglycaemia (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; Urakami 2017). Two studies reported data in the publications (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1). For one study, the investigator reported that no participant experienced mild/moderate hypoglycaemia (Urakami 2017).

A total of 624/646 participants (96.6%) in the insulin degludec group compared with 312/324 participants (96.3%) in the insulin glargine group experienced mild/moderate hypoglycaemia. There was no evidence of a difference in mild/moderate hypoglycaemia (RR 1.02, 95% CI 0.99 to 1.04; P = 0.18; 3 studies, 970 participants; Analysis 5.20). We judged the overall risk of bias for this outcome as 'some concerns'.

We did not perform subgroup and sensitivity analyses due to lack of data (Appendix 20).

Socioeconomic effects

No studies reported the costs of the intervention during the study period. One co‐publication analysed the cost‐effectiveness based on applying assumptions from two studies to a UK National Health Service perspective (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1).

HbA1c

Four studies reported HbA1c levels (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; SWITCH 1; Urakami 2017). Three studies reported data in the publications (BEGIN Flex T1; SWITCH 1; Urakami 2017). BEGIN Basal‐Bolus Type 1 only reported HbA1c after the extension period and not after the end of the regular intervention in the publication. However, we could retrieve these data from the CSR.

There was a reduction in HbA1c in favour of insulin glargine (MD 0.1%, 95% CI 0.0 to 0.2; P = 0.05; 1388 participants; 4 studies; Analysis 5.21; low‐certainty evidence). We judged the overall risk of bias for this outcome as 'low'.

Subgroup and sensitivity analyses

Three studies reported HbA1c in adults with a MD of 0.1%, 95% CI 0.0 to 0.2; Analysis 5.21. One study reported HbA1c in children with a MD of 0%, 95% CI ‐0.6 to 0.6; Analysis 5.21. The test for subgroup differences did not indicate interaction (P = 0.71).

Analysing only published data indicated a MD in HbA1c of 0.1%, 95% 0.02 to 0.3; Analysis 5.22. Analysing only unpublished data indicated a MD in HbA1c of 0.0%, 95% CI ‐0.2 to 0.2; Analysis 5.22. The test for subgroup differences did not indicate interaction (P = 0.26).

Sensitivity analysis excluding the largest study and the longest study indicated a MD in HbA1c of 0.1%, 95% CI 0.02 to 0.3 (BEGIN Basal‐Bolus Type 1).

The remaining subgroup analyses could not be performed due to lack of data (Appendix 20).

Combined HbA1c and severe hypoglycaemia

A combined measure of HbA1c and severe hypoglycaemia was available from two studies through the CSRs (BEGIN Basal‐Bolus Type 1; BEGIN Flex T1). BEGIN Basal‐Bolus Type 1 reported that 174/453 participants (38.4%) in the insulin degludec group compared with 63/149 participants (42.3%) in the insulin glargine group achieved the HbA1c target < 7% without severe hypoglycaemia during the last 12 weeks of treatment. BEGIN Flex T1 reported that 56/153 participants (36.6%) in the insulin degludec group compared with 60/156 participants (38.5%) in the insulin glargine group achieved the HbA1c target < 7% without severe hypoglycaemia during the last 12 weeks of treatment.

There was no evidence of a difference in people achieving HbA1c < 7% without severe hypoglycaemia (RR 0.92, 95% CI 0.78 to 1.10; 2 studies, 911 participants; Analysis 5.23). We judged the overall risk of bias for this outcome as 'low' (data not shown).

Assessment of reporting bias

We did not draw funnel plots due to limited number of studies per outcome included in the analyses.

Ongoing studies

We did not identify ongoing trials of interest for this review.

Studies awaiting assessment

We identified 13 studies with 20 records which we classified as awaiting classification (Agesen 2019; Basal Analog Study; ChiCTR2000032703; EudraCT 2007‐004144‐74; EudraCT 2009‐012317‐22; INEOX; IRCT201203079224N1; J‐Collection; Mianowska 2007; NCT00564018; Sherif 2014; UMIN000020521; UMIN000021046); for details please see 'Studies awaiting classification'.

Three studies randomising 474 participants compared insulin degludec with insulin glargine (Agesen 2019; ChiCTR2000032703; INEOX). Four studies randomising 253 participants compared insulin detemir with insulin glargine (Basal Analog Study; EudraCT 2007‐004144‐74; EudraCT 2009‐012317‐22; J‐Collection). Three studies randomising 154 participants compared insulin glargine with NPH insulin (IRCT201203079224N1; Mianowska 2007; Sherif 2014).

Two studies had more than two intervention groups: one study randomising 33 participants had three intervention groups comparing insulin glargine with insulin detemir with NPH insulin (NCT00564018), one study randomising 100 participants compared insulin degludec with insulin glargine and with continuing existing basal insulin treatment (UMIN000020521).

One study compared insulin degludec with another unspecified long‐acting insulin analogue (UMIN000021046). This study randomising 200 participants included people with T1DM and T2DM.

Seven studies were marked as awaiting classification, as they were listed as completed, but no publications were yet available (Agesen 2019; EudraCT 2007‐004144‐74; EudraCT 2009‐012317‐22; INEOX; IRCT201203079224N1; UMIN000020521; UMIN000021046).

Two studies were published as abstracts (Basal Analog Study; Sherif 2014). One study had results available in the trials register – however, it was stated in the trials register that the trial was ended prematurely. It was not possible through correspondence with authors to clarify how long the trial continued (EudraCT 2007‐004144‐74). One study was listed as completed and prematurely ended with no study data (NCT00564018).

One cross‐over study had a full‐text publication available. No data could be retrieved before cross‐over from the publication (Mianowska 2007).

Investigators were contacted, if this was possible, in order to get the status of the studies clarified (See Appendix 19).

Discussion

Summary of main results

This Cochrane Review is the first systematic review investigating the effects of (ultra‐)long acting insulin analogues in people with T1DM with substantial amounts of information from CSRs and clinical study synopses. We included 26 studies with 8784 participants: 2428 participants were randomised to NPH insulin, 2889 participants to insulin detemir, 2095 participants to insulin glargine and 1372 participants to insulin degludec. Eight studies contributing 21% of all participants included children.

The amount of evidence on patient‐important outcomes was limited from full‐text publications. However, we could retrieve substantial data on patient‐important outcomes from the CSRs. There was moderate‐certainty evidence comparing insulin detemir with NPH insulin for T1DM showing a lower risk of severe hypoglycaemia in favour of insulin detemir. However, the 95% prediction interval indicated inconsistency of this result. Insulin detemir or insulin glargine compared with NPH insulin did not show benefits or harms for severe nocturnal hypoglycaemia. For all other main outcomes, with overall low risk of bias and comparing insulin analogues with each other, there were no clear differences. Data on patient‐important outcomes such as health‐related quality of life, macrovascular and microvascular diabetic complications were sparse or missing.

Comparing the insulin analogues detemir and glargine with NPH insulin, we are moderately confident about the results for all‐cause mortality, severe (nocturnal) hypoglycaemia, SAEs and HbA1c. We are uncertain about the effects on non‐fatal myocardial infarction, non‐fatal stroke and health‐related quality of life, mainly because data were sparse or there were only a few studies which did not last long enough to investigate these outcomes.

There was no evidence of a difference in any outcome between children and adults.

Overall completeness and applicability of evidence

We conducted an extensive search for studies, included publications in all languages, and tried to obtain additional data on all studies. We identified two unpublished studies (NCT00595374; NCT00605137). We managed to retrieve additional unpublished information on all studies, except for three studies which were only available as full‐text publications (Bolli 2009; Porcellati 2004; Urakami 2017). Two study authors provided personal information on their studies (Home 2005; Urakami 2017), One unpublished study did not have a CSR but some data could be retrieved from a clinical study synopsis (NCT00595374). Two Japanese studies had CSRs, but we were unable to obtain the complete version of these (Kobayashi 2007; NCT00605137). Two studies had a cross‐over design and not all data could be analysed or were reported before cross‐over which we needed because of potential carryover effects (SWITCH 1; Urakami 2017). We looked for additional studies and cross‐checked our data with the data from other meta‐analyses of relevance (Laranjeira 2018; Tricco 2014; Tricco 2018). The information obtained from CSRs was clearly the best to establish an adequate 'Risk of bias assessment' and to maximise the yield of information for our prespecified outcomes (Appendix 22; Appendix 23; Appendix 24; Appendix 25; Appendix 26; Appendix 27; Appendix 28; Appendix 29; Appendix 30; Appendix 31; Appendix 32; Appendix 33; Appendix 34; Appendix 35; Appendix 36; Appendix 37; Appendix 38; Appendix 39; Appendix 40; Appendix 41; Appendix 42). We noticed major differences between reported outcomes in publications and CSRs, e.g. all‐cause mortality was documented in 25% of publications compared to 91% in CSRs (Appendix 41). SAEs and non‐serious adverse events were documented in 54% of publications compared to 91% in CSRs (Appendix 41). However, the amount of information within the CSRs varied substantially and we probably did not have access to a single full CSR (Appendix 7).

We investigated a broad spectrum of people with T1DM as both children and adults were included. However, we did not include pregnant women with T1DM, as we anticipated these women would have pronounced fluctuating insulin requirements and a specific hypoglycaemia risk profile. All studies were performed in white or Asian people. Data on people of African origin were lacking. None of the studies was performed in low‐ or middle‐income settings.

Quality of the evidence

Depending on the outcome measures, we judged the certainty of the evidence as moderate for all‐cause mortality, severe hypoglycaemia, severe nocturnal hypoglycaemia, SAEs and HbA1c. For most comparisons, we judged the certainty of the evidence as low for non‐fatal myocardial infarction, non‐fatal stroke and health‐related quality of life. No information or only few data were available for blindness, end‐stage renal disease, combined HbA1c with severe hypoglycaemia and socioeconomic effects.

For all studies, we contacted one or more study authors to obtain supplemental information on baseline data, 'Risk of bias' domains and outcomes (see Appendix 19). However, several investigators advised us to contact the pharmaceutical company of the study, as they did not have access to the full dataset.

All studies but six had a non‐inferiority RCT design which is often required for regulatory approval (Bolli 2009; Home 2005; Porcellati 2004; Ratner 2000; Schober 2002; Urakami 2017). The usual primary endpoint was change in HbA1c which does not minimise the reliability of analysing other outcomes such as hypoglycaemia by means of meta‐analysis, because with a potential benefit of newer compounds in reducing HbA1c, a benefit of the number of hypoglycaemic episodes could be expected. Adjustments of hypoglycaemic events for HbA1c levels or achievement of certain HbA1c thresholds without hypoglycaemia would provide important information. Unfortunately, only a few of our included studies reported on this combined endpoint, and, if done, no clear differences were recorded.

All studies except one had an open‐label design (SWITCH 1). This could have influenced some of the subjective outcome measures, especially health‐related quality of life, non‐serious adverse events, mild/moderate hypoglycaemia and some measures of nocturnal hypoglycaemia. Another factor influencing findings could have been investigators being more careful when adjusting the newer insulin analogues due to less clinical experience with these compounds. Also, some participants might have been more prone to measure blood glucose as they might have anticipated experiencing more hypoglycaemic episodes with human insulin preparations, thereby even affecting hypoglycaemia confirmed with blood glucose measurements.

Improving and maintaining glycaemic control in T1DM is a key objective. However, hypoglycaemia is a serious problem affecting health‐related quality of life and treatment satisfaction of people with diabetes, making it difficult to achieve near‐normal glucose levels in T1DM. Therefore, for any proclaimed benefit of an intervention on hypoglycaemia, it is vital to evaluate the risk of bias in order to establish reliable results. 'Risk of bias' assessment depends considerably on the definition of hypoglycaemia. It appears low if severe hypoglycaemia is also reported as a serious adverse event (SAE) because there is a standard definition of SAEs, or if the combined endpoint of HbA1c levels with severe hypoglycaemia is reported. Unfortunately, no data were available for the combined endpoint HbA1c with severe hypoglycaemia for the comparisons insulin detemir versus NPH insulin and insulin glargine versus NPH insulin. Of note, only about one third of participants being treated with insulin glargine, insulin detemir or insulin degludec achieved an HbA1c < 7% without severe hypoglycaemia. Other definitions of severe hypoglycaemia like hypoglycaemia‐induced coma or convulsions, necessity for intubation or intensive‐care unit stay also reflect hard clinical endpoints. However, the included studies most often defined severe hypoglycaemia as a hypoglycaemic event which needed "third party assistance". This is prone to bias because third party assistance might encompass a broad range of interventions, e.g. giving food or a drink by a relative or friend, subcutaneous glucagon injection or intravenous glucose administration. Only Thalange 2013 made an effort to define third party assistance in a way that minimised risk of bias (the child had to be semiconscious or unconscious or in coma with or without convulsions and may have required parenteral treatment with glucagon or intravenous glucose). A Cochrane Review associated to this systematic review will establish an in‐depth analysis of the definitions and reporting of hypoglycaemia in trials of long‐acting insulin analogues in people with type 1 diabetes mellitus (Ørskov Ipsen 2020).

An overview of the reported definitions of hypoglycaemic episodes in our included studies found no evidence of differences between the various interventions on these outcomes with the exception of insulin detemir compared with NPH insulin, demonstrating a benefit for severe hypoglycaemia, any/mild/symptomatic nocturnal hypoglycaemia and mild/moderate hypoglycaemia (Appendix 42). With the exception of severe hypoglycaemia, we judged the risk of bias as 'some concerns' for measurement of these outcomes. There was no benefit or risk of insulin detemir for hypoglycaemia reported as a SAE or severe nocturnal hypoglycaemia event (Appendix 42).

Long‐term complications of diabetes were sparsely reported. Long‐term complications of diabetes develop over years, and therefore the duration of the included studies might have been too short to identify if an intervention had beneficial or harmful effects. Data on all‐cause mortality were most often retrieved from CSRs and few deaths were observed in the studies. However, to our knowledge, no data from long‐term observational studies indicate that the type of intermediate or (ultra‐)long‐acting insulin influences the risk of death or macrovascular and microvascular complications of diabetes. However, long‐term follow‐up from interventional studies has shown that good glycaemic control in people with T1DM is an important factor for preventing complications (DCCT/EDIC 2016).

No studies reported the direct costs of insulin treatment during the study period. Several studies had co‐publications with economic analyses in different country settings based on assumptions derived from the clinical study (Bartley 2008; BEGIN Basal‐Bolus Type 1; BEGIN Flex T1; BEGIN Young; Pieber 2007). However, these assumptions do not seem to be supported by our meta‐analyses of the clinical trial data. Furthermore, other studies have shown that the direct costs of the long‐acting insulin analogues often are substantially higher than the costs of NPH insulin (Ewen 2019).

Only one study had not received free drugs or financial funding from the pharmaceutical industry (Porcellati 2004). It is known that studies receiving funding or provision of free drugs or devices from a pharmaceutical company lead to more favourable results and conclusions compared to studies sponsored by other sources (Lundh 2017).

Potential biases in the review process

We were unable to draw funnel plots to assess small‐study bias due to lack of data. We tried to explore inconsistency of results and the reasons for it through subgroup and sensitivity analyses. The only factor, comparing insulin detemir with NPH insulin, that indicated an influence on the effect estimate for the sensitivity analysis of one outcome (severe hypoglycaemia) was publication status. This has to be interpreted with caution because the subgroup of studies with unpublished data consisted of two studies only.

We identified 13 studies as 'awaiting classification'. Data from these studies would have added information on an additional 1194 participants. Most of the studies were listed as completed in trials registers, but data, publications or both were not available. For some of the studies, these data might not yet have been analysed, but other studies were completed years ago and are still not published (Basal Analog Study; EudraCT 2007‐004144‐74; J‐Collection; NCT00564018; UMIN000021046). For most studies awaiting classification, we contacted the investigators for clarification.

We were dealing with a heterogeneous group of studies. Our meta‐analyses, when performed, were limited by the inability to use individual participant data to assess whether distinct clinical characteristics may have influenced the effect estimates of the interventions. Many of the included studies were designed and powered to detect changes in HbA1c but, for all studies, we were able to extract most of our predefined outcomes.

Several studies were published in more than one publication which, for some studies, made it difficult to separate the primary publication from companion papers (for details, see Included studies).

Two review authors carried out data extraction. However, the review authors extracting the data were not blinded as to from which study they were extracting data.

We only included studies with a duration of 24 weeks or more to get some information on patient‐relevant outcomes; by not including studies with a shorter duration, we might have underestimated the short‐term risks of the interventions.

A potential selection bias exists as more healthy and motivated people may participate in a clinical study. However, a Cochrane Review observed that clinical outcomes in people participating in RCTs are not substantially different to outcomes in comparable individuals outside the RCT context (Vist 2008).

We requested CSRs and other information from EMA. EMA replied that it "is currently operating within the fourth phase of its business continuity plan to ensure operational continuity during its relocation to Amsterdam. Whilst every effort is being made to process all requests as soon as possible, you should be aware that due to these exceptional circumstances from October 2019 requests cannot be processed immediately and will be dealt in a chronological order from the time they were received". At the moment of publication of this Cochrane Review, the first pieces of information from EMA are arriving. Because we do not know when the last information package of EMA will be available, we plan to make full use of EMA data in a future update of our review. In case of very important EMA data, we will publish an interim updated version of our review as soon as possible.

Agreements and disagreements with other studies or reviews

Other reviews of insulin analogues in people with T1DM have been published. The most recent systematic review was performed for refinement of the WHO Essential Medicine List (EML), which was an update of a systematic review published in 2014 (Tricco 2018). The review for WHO EML included adults with T1DM, but also included pregnant women with T1DM. We did not chose to include the latter cohort as pregnancy causes considerable changes in insulin sensitivity. Tricco 2018 included studies irrespective of study duration. We required a minimum duration of 24 weeks to get more reliable information on patient‐relevant outcome measures. Short‐term studies usually evaluate surrogate markers and often have shorter intervention periods than the titration periods of the longer‐term studies. Tricco 2018 analysed insulin glargine and insulin detemir together; we chose to perform separate analyses. Tricco 2018 included 62 RCTs according to the abstract. However, they missed identifying co‐publications of primary publications. Therefore, several studies were included more than once and handled as independent studies. Another difference to our review is the lack of identification of unpublished data, especially CSRs which provided substantial information to all our analyses including 'Risk of bias' assessment. Tricco 2018 reported a statistically significant decrease in HbA1c with insulin analogues compared with NPH insulin and a statistically significant lower risk of severe hypoglycaemia, which we could not verify in our analyses. One umbrella review of reviews compared long‐acting insulin analogues with NPH insulin (Laranjeira 2018). Eleven systematic reviews were identified and a total of 25 RCTs were included irrespective of age of participants or duration of the intervention. The conclusion of this overview, based on data for all systematic reviews, was that long‐acting insulin analogues were more effective than NPH insulin concerning lowering HbA1c. No statistically significant differences were found for severe hypoglycaemia (Laranjeira 2018).

Figure 1

Screen4Me: Cochrane´s screening service.

Figure 2

Study flow diagram

CSR: clinical study report; EMA: European Medicines Agency; HTA: health technology assessment; Screen4Me: Cochrane's screening service.

Figure 3

Severe hypoglycaemia

Figure 4

Severe hypoglycaemia

Figure 5

Severe hypoglycaemia

Figure 6

Severe hypoglycaemia

Figure 7

Severe hypoglycaemia

Analysis 1.1

Comparison 1: Insulin detemir versus NPH insulin, Outcome 1: All‐cause mortality

Analysis 1.2

Comparison 1: Insulin detemir versus NPH insulin, Outcome 2: All‐cause mortality (published vs. unpublished data)

Analysis 1.3

Comparison 1: Insulin detemir versus NPH insulin, Outcome 3: Severe hypoglycaemia

Analysis 1.4

Comparison 1: Insulin detemir versus NPH insulin, Outcome 4: Severe hypoglycaemia (published vs. unpublished data)

Analysis 1.5

Comparison 1: Insulin detemir versus NPH insulin, Outcome 5: Hypoglycaemia reported as a serious adverse event

Analysis 1.6

Comparison 1: Insulin detemir versus NPH insulin, Outcome 6: Cardiovascular mortality

Analysis 1.7

Comparison 1: Insulin detemir versus NPH insulin, Outcome 7: Non‐fatal myocardial infarction

Analysis 1.8

Comparison 1: Insulin detemir versus NPH insulin, Outcome 8: Serious adverse events

Analysis 1.9

Comparison 1: Insulin detemir versus NPH insulin, Outcome 9: Serious adverse events (published vs. unpublished data)

Analysis 1.10

Comparison 1: Insulin detemir versus NPH insulin, Outcome 10: Diabetic ketoacidosis

Analysis 1.11

Comparison 1: Insulin detemir versus NPH insulin, Outcome 11: Diabetic ketoacidosis (published vs. unpublished data)

Analysis 1.12

Comparison 1: Insulin detemir versus NPH insulin, Outcome 12: Non‐serious adverse events

Analysis 1.13

Comparison 1: Insulin detemir versus NPH insulin, Outcome 13: Non‐serious adverse events (published vs unpublished data)

Analysis 1.14

Comparison 1: Insulin detemir versus NPH insulin, Outcome 14: Withdrawals due to adverse events

Analysis 1.15

Comparison 1: Insulin detemir versus NPH insulin, Outcome 15: Any nocturnal hypoglycaemia

Analysis 1.16

Comparison 1: Insulin detemir versus NPH insulin, Outcome 16: Mild nocturnal hypoglycaemia

Analysis 1.17

Comparison 1: Insulin detemir versus NPH insulin, Outcome 17: Nocturnal hypoglycaemia (symptoms)

Analysis 1.18

Comparison 1: Insulin detemir versus NPH insulin, Outcome 18: Severe nocturnal hypoglycaemia

Analysis 1.19

Comparison 1: Insulin detemir versus NPH insulin, Outcome 19: Any nocturnal hypoglycaemia (published vs. unpublished data)

Analysis 1.20

Comparison 1: Insulin detemir versus NPH insulin, Outcome 20: Mild nocturnal hypoglycaemia (published vs. unpublished data)

Analysis 1.21

Comparison 1: Insulin detemir versus NPH insulin, Outcome 21: Nocturnal hypoglycaemia, symptoms only (published vs. unpublished data)

Analysis 1.22

Comparison 1: Insulin detemir versus NPH insulin, Outcome 22: Severe nocturnal hypoglycaemia (published vs. unpublished data)

Analysis 1.23

Comparison 1: Insulin detemir versus NPH insulin, Outcome 23: Nocturnal hypoglycaemia, asymptomatic (children vs. adults)

Analysis 1.24

Comparison 1: Insulin detemir versus NPH insulin, Outcome 24: Mild/moderate hypoglycaemia

Analysis 1.25

Comparison 1: Insulin detemir versus NPH insulin, Outcome 25: Mild/moderate hypoglycaemia (published vs. unpublished data)

Analysis 1.26

Comparison 1: Insulin detemir versus NPH insulin, Outcome 26: HbA1c

Analysis 1.27

Comparison 1: Insulin detemir versus NPH insulin, Outcome 27: HbA1c (published vs. unpublished data)

Analysis 2.1

Comparison 2: Insulin glargine versus NPH insulin, Outcome 1: All‐cause mortality

Analysis 2.2

Comparison 2: Insulin glargine versus NPH insulin, Outcome 2: Health‐realted quality of life

Analysis 2.3

Comparison 2: Insulin glargine versus NPH insulin, Outcome 3: Severe hypoglycaemia

Analysis 2.4

Comparison 2: Insulin glargine versus NPH insulin, Outcome 4: Severe hypoglycaemia (published vs. unpublished data)

Analysis 2.5

Comparison 2: Insulin glargine versus NPH insulin, Outcome 5: Hypoglycaemia reported as a serious adverse event

Analysis 2.6

Comparison 2: Insulin glargine versus NPH insulin, Outcome 6: Cardiovascular mortality

Analysis 2.7

Comparison 2: Insulin glargine versus NPH insulin, Outcome 7: Non‐fatal myocardial infarction

Analysis 2.8

Comparison 2: Insulin glargine versus NPH insulin, Outcome 8: Non‐fatal stroke

Analysis 2.9

Comparison 2: Insulin glargine versus NPH insulin, Outcome 9: Serious adverse events

Analysis 2.10

Comparison 2: Insulin glargine versus NPH insulin, Outcome 10: Serious adverse events (published vs. unpublished data)

Analysis 2.11

Comparison 2: Insulin glargine versus NPH insulin, Outcome 11: Diabetic ketoacidosis

Analysis 2.12

Comparison 2: Insulin glargine versus NPH insulin, Outcome 12: Diabetic ketoacidosis (published vs. unpublished data)

Analysis 2.13

Comparison 2: Insulin glargine versus NPH insulin, Outcome 13: Non‐serious adverse events

Analysis 2.14

Comparison 2: Insulin glargine versus NPH insulin, Outcome 14: Non‐serious adverse events (published vs. unpublished data)

Analysis 2.15

Comparison 2: Insulin glargine versus NPH insulin, Outcome 15: Withdrawals due to adverse events

Analysis 2.16

Comparison 2: Insulin glargine versus NPH insulin, Outcome 16: Nocturnal hypoglycaemia

Analysis 2.17

Comparison 2: Insulin glargine versus NPH insulin, Outcome 17: Mild nocturnal hypoglycaemia

Analysis 2.18

Comparison 2: Insulin glargine versus NPH insulin, Outcome 18: Nocturnal hypoglycaemia (symptoms)

Analysis 2.19

Comparison 2: Insulin glargine versus NPH insulin, Outcome 19: Severe nocturnal hypoglycaemia

Analysis 2.20

Comparison 2: Insulin glargine versus NPH insulin, Outcome 20: Nocturnal hypoglycaemia (published vs. unpublished data)

Analysis 2.21

Comparison 2: Insulin glargine versus NPH insulin, Outcome 21: Symptomatic nocturnal hypoglycaemia (published vs. unpublished data)

Analysis 2.22

Comparison 2: Insulin glargine versus NPH insulin, Outcome 22: Mild/moderate hypoglycaemia

Analysis 2.23

Comparison 2: Insulin glargine versus NPH insulin, Outcome 23: Mild/moderate hypoglycaemia (published vs. unpublished data)

Analysis 2.24

Comparison 2: Insulin glargine versus NPH insulin, Outcome 24: HbA1c

Analysis 2.25

Comparison 2: Insulin glargine versus NPH insulin, Outcome 25: HbA1c (published vs unpublished data)

Analysis 2.26

Comparison 2: Insulin glargine versus NPH insulin, Outcome 26: HbA1c (NPH < 2x/day vs ≥ 2x/day)

Analysis 3.1

Comparison 3: Insulin detemir versus insulin glargine, Outcome 1: All‐cause mortality

Analysis 3.2

Comparison 3: Insulin detemir versus insulin glargine, Outcome 2: Severe hypoglycaemia

Analysis 3.3

Comparison 3: Insulin detemir versus insulin glargine, Outcome 3: Severe hypoglycaemia (published vs. unpublished data)

Analysis 3.4

Comparison 3: Insulin detemir versus insulin glargine, Outcome 4: Hypoglycaemia reported as a serious adverse event

Analysis 3.5

Comparison 3: Insulin detemir versus insulin glargine, Outcome 5: Cardiovascular mortality

Analysis 3.6

Comparison 3: Insulin detemir versus insulin glargine, Outcome 6: Non‐fatal myocardial infarction

Analysis 3.7

Comparison 3: Insulin detemir versus insulin glargine, Outcome 7: Non‐fatal stroke

Analysis 3.8

Comparison 3: Insulin detemir versus insulin glargine, Outcome 8: Serious adverse events

Analysis 3.9

Comparison 3: Insulin detemir versus insulin glargine, Outcome 9: Diabetic ketoacidosis

Analysis 3.10

Comparison 3: Insulin detemir versus insulin glargine, Outcome 10: Non‐serious adverse events

Analysis 3.11

Comparison 3: Insulin detemir versus insulin glargine, Outcome 11: Non‐serious adverse events (published vs. unpublished data)

Analysis 3.12

Comparison 3: Insulin detemir versus insulin glargine, Outcome 12: Withdrawals due to adverse events

Analysis 3.13

Comparison 3: Insulin detemir versus insulin glargine, Outcome 13: Any nocturnal hypoglycaemia

Analysis 3.14

Comparison 3: Insulin detemir versus insulin glargine, Outcome 14: Confirmed nocturnal hypoglycaemia (PG < 3.1 mmol/L and no assistance)

Analysis 3.15

Comparison 3: Insulin detemir versus insulin glargine, Outcome 15: Symptomatic nocturnal hypoglycaemia (PG ≥ 3.1 or no PG and no assistance required)

Analysis 3.16

Comparison 3: Insulin detemir versus insulin glargine, Outcome 16: Severe nocturnal hypoglycaemia

Analysis 3.17

Comparison 3: Insulin detemir versus insulin glargine, Outcome 17: Mild/moderate hypoglycaemia

Analysis 3.18

Comparison 3: Insulin detemir versus insulin glargine, Outcome 18: HbA1c

Analysis 3.19

Comparison 3: Insulin detemir versus insulin glargine, Outcome 19: Individuals with HbA1c < 7% without severe hypoglycaemia

Analysis 4.1

Comparison 4: Insulin degludec versus insulin detemir, Outcome 1: All‐cause mortality

Analysis 4.2

Comparison 4: Insulin degludec versus insulin detemir, Outcome 2: Health‐related quality of life

Analysis 4.3

Comparison 4: Insulin degludec versus insulin detemir, Outcome 3: Severe hypoglycaemia

Analysis 4.4

Comparison 4: Insulin degludec versus insulin detemir, Outcome 4: Hypoglycaemia reported as a serious adverse event

Analysis 4.5

Comparison 4: Insulin degludec versus insulin detemir, Outcome 5: Cardiovascular mortality

Analysis 4.6

Comparison 4: Insulin degludec versus insulin detemir, Outcome 6: Non‐fatal myocardial infarction

Analysis 4.7

Comparison 4: Insulin degludec versus insulin detemir, Outcome 7: Non‐fatal stroke

Analysis 4.8

Comparison 4: Insulin degludec versus insulin detemir, Outcome 8: End stage renal disease

Analysis 4.9

Comparison 4: Insulin degludec versus insulin detemir, Outcome 9: Blindness

Analysis 4.10

Comparison 4: Insulin degludec versus insulin detemir, Outcome 10: Serious adverse events

Analysis 4.11

Comparison 4: Insulin degludec versus insulin detemir, Outcome 11: Diabetic ketoacidosis

Analysis 4.12

Comparison 4: Insulin degludec versus insulin detemir, Outcome 12: Non‐serious adverse events

Analysis 4.13

Comparison 4: Insulin degludec versus insulin detemir, Outcome 13: Withdrawals due to adverse events

Analysis 4.14

Comparison 4: Insulin degludec versus insulin detemir, Outcome 14: Nocturnal hypoglycaemia

Analysis 4.15

Comparison 4: Insulin degludec versus insulin detemir, Outcome 15: Mild nocturnal hypoglycaemia

Analysis 4.16

Comparison 4: Insulin degludec versus insulin detemir, Outcome 16: Nocturnal hypoglycaemia (symptomatic)

Analysis 4.17

Comparison 4: Insulin degludec versus insulin detemir, Outcome 17: Nocturnal hypoglycaemia (asymptomatic)

Analysis 4.18

Comparison 4: Insulin degludec versus insulin detemir, Outcome 18: Severe nocturnal hypoglycaemia

Analysis 4.19

Comparison 4: Insulin degludec versus insulin detemir, Outcome 19: Mild/moderate hypoglycaemia

Analysis 4.20

Comparison 4: Insulin degludec versus insulin detemir, Outcome 20: HbA1c

Analysis 4.21

Comparison 4: Insulin degludec versus insulin detemir, Outcome 21: Individuals with HbA1c < 7% without severe hypoglycaemia

Analysis 5.1

Comparison 5: Insulin degludec versus insulin glargine, Outcome 1: All‐cause mortality

Analysis 5.2

Comparison 5: Insulin degludec versus insulin glargine, Outcome 2: All‐cause mortality (published vs. unpublished data)

Analysis 5.3

Comparison 5: Insulin degludec versus insulin glargine, Outcome 3: Health‐related quality of life (physical health)

Analysis 5.4

Comparison 5: Insulin degludec versus insulin glargine, Outcome 4: Health‐related quality of life (mental health)

Analysis 5.5

Comparison 5: Insulin degludec versus insulin glargine, Outcome 5: Severe hypoglycaemia

Analysis 5.6

Comparison 5: Insulin degludec versus insulin glargine, Outcome 6: Hypoglycaemia reported as a serious adverse event

Analysis 5.7

Comparison 5: Insulin degludec versus insulin glargine, Outcome 7: Cardiovascular mortality

Analysis 5.8

Comparison 5: Insulin degludec versus insulin glargine, Outcome 8: Non‐fatal myocardial infarction

Analysis 5.9

Comparison 5: Insulin degludec versus insulin glargine, Outcome 9: Non‐fatal stroke

Analysis 5.10

Comparison 5: Insulin degludec versus insulin glargine, Outcome 10: Serious adverse events

Analysis 5.11

Comparison 5: Insulin degludec versus insulin glargine, Outcome 11: Diabetic ketoacidosis

Analysis 5.12

Comparison 5: Insulin degludec versus insulin glargine, Outcome 12: Diabetic ketoacidosis (published vs. unpublished data)

Analysis 5.13

Comparison 5: Insulin degludec versus insulin glargine, Outcome 13: Non‐serious adverse events

Analysis 5.14

Comparison 5: Insulin degludec versus insulin glargine, Outcome 14: Withdrawals due to adverse events

Analysis 5.15

Comparison 5: Insulin degludec versus insulin glargine, Outcome 15: Nocturnal hypoglycaemia

Analysis 5.16

Comparison 5: Insulin degludec versus insulin glargine, Outcome 16: Mild nocturnal hypoglycaemia

Analysis 5.17

Comparison 5: Insulin degludec versus insulin glargine, Outcome 17: Nocturnal hypoglycaemia (asymptomatic)

Analysis 5.18

Comparison 5: Insulin degludec versus insulin glargine, Outcome 18: Nocturnal hypoglycaemia (symptomatic)

Analysis 5.19

Comparison 5: Insulin degludec versus insulin glargine, Outcome 19: Severe nocturnal hypoglycaemia

Analysis 5.20

Comparison 5: Insulin degludec versus insulin glargine, Outcome 20: Mild/moderate hypoglycaemia

Analysis 5.21

Comparison 5: Insulin degludec versus insulin glargine, Outcome 21: HbA1c

Analysis 5.22

Comparison 5: Insulin degludec versus insulin glargine, Outcome 22: HbA1c (published vs. unpublished data)

Analysis 5.23

Comparison 5: Insulin degludec versus insulin glargine, Outcome 23: Individuals with HbA1c < 7% without severe hypoglycaemia

Summary of findings 1. Summary of findings: insulin detemir versus NPH insulin

Outcomes	NPH insulin	Insulin detemir	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin detemir compared with NPH insulin for T1DM
Patients: people with T1DM Settings: outpatients Intervention: insulin detemir Comparison: NPH insulin
All‐cause mortality Follow‐up: 24‐104 weeks	See comment		Peto OR4.97 (0.79 to 31.38)	3334 (9)	⊕⊕⊕⊝ moderate^a	All 5 deaths reported in 2 studies including adults occurred in the insulin detemir group
Health‐related quality of life Description: diabetes health profile; insulin therapy‐related quality of life at night (scale not specified) Follow‐up: 26‐48 weeks	See comment			870 (3)	⊕⊕⊝⊝ low^b	No study reported health‐related quality of life in a format making it suitable for meta‐analysis 1 study including adults reported higher scores in the insulin detemir group vs the NPH insulin group (Kobayashi 2007) 2 studies did not show evidence of a difference between intervention groups (NCT00595374 included children; Standl 2004 included adults)
Severe hypoglycaemia (n/N) Definition: hypoglycaemia requiring third party assistance (Bartley 2008; Kobayashi 2007; NCT00605137; Robertson 2007; Russell‐Jones 2004; Standl 2004; Thalange 2013; Vague 2003); episodes where the children were semi‐conscious, unconscious or in a coma, with or without convulsions (Thalange 2013) Follow‐up: 24‐104 weeks	115 per 1000	79 per 1000 (60 to 106)	RR 0.69 (0.52 to 0.92)	3219 (8)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between 0.34 and 1.39 5 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
Non‐fatal myocardial infarction/stroke Definition: myocardial infarction Follow‐up: 24 months	See comment			495 (1)	⊕⊕⊝⊝ low^d	1/331 participants in the insulin detemir group vs 0/164 participants in the NPH insulin group experienced a non‐fatal myocardial infarction (Bartley 2008) Stroke was not reported Study included adults
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring 23:00‐06:00 (Bartley 2008; NCT00605137; Russell‐Jones 2004; Standl 2004; Vague 2003); occurring 22:00‐07:00 (Robertson 2007; Thalange 2013) Follow‐up: 24 weeks ‐ 24 months	54 per 1000	36 per 1000 (21 to 64)	RR 0.67 (0.39 to 1.17)	2925 (7)	⊕⊕⊕⊝ moderate^e	The 95% prediction interval ranged between 0.16 and 2.87 4 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
Serious adverse events (n/N) Follow‐up: 24‐104 weeks	82 per 1000	78 per 1000 (62 to 100)	RR 0.95 (0.75 to 1.21)	3332 (9)	⊕⊕⊕⊝ moderate^e	The 95% prediction interval ranged between 0.71 and 1.27 6 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
HbA1c (%) Follow‐up: 24 weeks ‐ 24 months	The mean HbA1c ranged across the NPH insulin groups from 7.3% to 8.6%	The mean HbA1c in the insulin detemir groups was 0.01% higher (0.1% lower to 0.1% higher)	—	3122 (8)	⊕⊕⊕⊝ moderate^e	The 95% prediction interval ranged between ‐0.1% and 0.1% 5 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; CSR: clinical study report; HbA1c: glycosylated haemoglobin A1c; n/N: number of people experiencing an event; NPH: neutral protamine Hagedorn; OR: odds ratio RR: risk ratio; T1DM: type 1 diabetes mellitus.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by one level because of indirectness (insufficient time frame) ‐ see Appendix 1. ^bDowngraded by two levels because of overall risk of bias ('some concerns') and imprecision (few studies) ‐ see Appendix 1. ^cDowngraded by one level because of inconsistency (95% prediction interval consistent with benefit and harm) ‐ see Appendix 1. ^dDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 1. ^eDowngraded by one level because of imprecision (CI consistent with benefit and harm) ‐ see Appendix 1.

Summary of findings 1. Summary of findings: insulin detemir versus NPH insulin

Summary of findings 2. Summary of findings: insulin glargine versus NPH insulin

Outcomes	NPH insulin	Insulin glargine	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin glargine compared with NPH insulin for T1DM
Patients: people with T1DM Settings: outpatients Intervention: insulin glargine Comparison: NPH insulin
All‐cause mortality Follow‐up: 24‐52 weeks	See comment		Peto OR0.14 (0.00 to 6.98)	2175 (8)	⊕⊕⊕⊝ moderate^a	1 study including adults reported 0/1207 participants died in the insulin glargine group vs 1/1068 participants in the NPH insulin group 4 studies included adults, 4 studies included children (the test for subgroup differences could not be performed)
Health‐related quality of life Scales: Well‐Being Enquiry for Diabetics; General Well‐being; Diabetes Quality of Life for Youth and Parents' Diabetes Quality of Life Follow‐up: 24‐28 weeks	See comment			1013 (4)	⊕⊕⊝⊝ low^b	1 study including adults (Bolli 2009) reported greater improvements in the insulin glargine group compared with NPH insulin in one domain (diabetes‐related worries) There was no evidence of a difference in 3 studies (Chase 2008 included children; Home 2005 and Ratner 2000 included adults)
Severe hypoglycaemia (n/N) Definition: symptomatic hypoglycaemia requiring third party assistance, with either a blood glucose level < 2.8 mmol/L or prompt recovery after administration of oral carbohydrate, iv glucose or glucagon (Fulcher 2005; Home 2005; Schober 2002); requiring third party assistance and associated with either blood glucose < 2.0 mmol/L or prompt recovery after oral carbohydrate, iv glucose, or intramuscular or subcutaneous glucagon administration (Chase 2008); hypoglycaemia requiring third party assistance or involving a seizure, coma, unconsciousness or the use of glucagon (Liu 2016); hypoglycaemia requiring third party assistance (Porcellati 2004; PRESCHOOL; Ratner 2000) Follow‐up: 24‐52 weeks	125 per 1000	105 per 1000 (84 to 130)	RR 0.84 (0.67 to 1.04)	2350 (9)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between 0.65 and 1.09 5 studies included adults, 4 studies included children (the test for subgroup differences did not indicate interaction)
Non‐fatal myocardial infarction/stroke Definition: myocardial infarction/cerebral ischaemia Follow‐up: 28 weeks	See comment			585 (1)	⊕⊕⊝⊝ low^d	No participant experienced a non‐fatal myocardial infarction 1 study including adults reported 0/292 participants in the insulin glargine group vs 1/293 participants in the NPH insulin group experienced cerebral ischaemia (Home 2005)
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring 23:00‐07:00 (PRESCHOOL); severe hypoglycaemia occurring after the evening insulin injection and before the morning insulin dose (Fulcher 2005); severe hypoglycaemia occurring during sleep between bedtime and rising in the morning, or before the morning pre‐breakfast self‐blood glucose measurement and the morning insulin injection (Home 2005); severe hypoglycaemia occurring while asleep after the bedtime insulin dose and before the morning insulin dose and before the morning blood glucose measurement (Ratner 2000); severe hypoglycaemia while the participant was sleeping between bedtime and after the evening injection and before getting up in the morning (Schober 2002); severe hypoglycaemia occurring 00:00‐06:00 (Chase 2008) Follow‐up: 24‐28 weeks	87 per 1000	72 per 1000 (54 to 97)	RR 0.83 (0.62 to 1.12)	1893 (6)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between 0.54 and 1.27 3 studies included adults, 3 studies included children (the test for subgroup differences did not indicate interaction)
Serious adverse events (n/N) Follow‐up: 24‐30 weeks	100 per 1000	108 per 1000 (63 to 184)	RR 1.08 (0.63 to 1.84)	2229 (8)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between 0.22 and 5.21 4 studies included adults, 4 studies included children (the test for subgroup differences did not indicate interaction)
HbA1c (%) Follow‐up: 24 weeks ‐ 1 year	The mean HbA1c ranged across the NPH insulin groups from 7.1% to 7.3%	The mean HbA1c in the insulin glargine groups was 0.02% higher (0.1% lower to 0.1% higher)	—	2285 (9)	⊕⊕⊕⊝ moderate^c	The 95% prediction interval ranged between ‐0.5% and 0.5% 5 studies included adults, 4 studies included children (the test for subgroup differences did not indicate interaction)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). a.m.: ante meridiem; CI: confidence interval; HbA1c: glycosylated haemoglobin A1c; iv: intravenous; n/N: number of people experiencing an event; NPH: neutral protamine Hagedorn; RR: risk ratio; T1DM: type 1 diabetes mellitus.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by one level because of indirectness (insufficient time frame) ‐ see Appendix 2. ^bDowngraded by two levels because of overall risk of bias ('some concerns') and imprecision (few studies) ‐ see Appendix 2. ^cDowngraded by one level because of imprecision (CI consistent with benefit and harm) ‐ see Appendix 2. ^dDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 2.

Summary of findings 2. Summary of findings: insulin glargine versus NPH insulin

Summary of findings 3. Summary of findings: insulin detemir versus insulin glargine

Outcomes	Insulin glargine	Insulin detemir	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin detemir compared with insulin glargine for T1DM
Patients: people with T1DM Settings: outpatients Intervention: insulin detemir Comparison: insulin glargine
All‐cause mortality Follow‐up: 26 and 52 weeks	See comment			763 (2)	⊕⊕⊝⊝ low^a	No participant died 2 studies included adults
Health‐related quality of life	Not reported
Severe hypoglycaemia (n/N) Definition: hypoglycaemia requiring third party assistance Follow‐up: 26 and 52 weeks	116 per 1000	68 per 1000 (15 to 304)	RR 0.59 (0.13 to 2.63)	763 (2)	⊕⊝⊝⊝ very low^b	2 studies included adults
Non‐fatal myocardial infarction/stroke Definition: non‐fatal myocardial infarction/stroke Follow‐up: 52 weeks	See comment			443 (1)	⊕⊕⊝⊝ low^a	1 study including adults reported 1/299 participants in the insulin detemir group vs 1/144 participants in the insulin glargine group experienced a non‐fatal myocardial infarction One study including adults reported 2/299 participants in the insulin detemir group vs 0/144 participants in the insulin glargine group experienced a non‐fatal stroke
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring from 11 p.m. to 6 a.m. Follow‐up: 26 and 52 weeks	50 per 1000	27 per 1000 (3 to 253)	RR 0.55 (0.06 to 5.12)	763 (2)	⊕⊝⊝⊝ very low^b	2 studies included adults
Serious adverse events (n/N) Follow‐up: 26 and 52 weeks	59 per 1000	102 per 1000 (54 to 195)	RR 1.72 (0.91 to 3.28)	763 (2)	⊕⊕⊝⊝ low^c	The fixed‐effect statistical model showed an RR of 1.79 (1.04 to 3.08) in favour of insulin glargine 2 studies included adults
HbA1c (%) Follow‐up: 26 and 52 weeks	The mean HbA1c ranged across the insulin glargine groups from 7.6% to 8.2%	The mean HbA1c in the insulin detemir groups was 0.01% lower (0.1% lower to 0.1% higher)	—	763 (2)	⊕⊕⊝⊝ low^c	2 studies included adults
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). a.m.: ante meridiem; CI: confidence interval; HbA1c: glycosylated haemoglobin A1c; n/N: number of people experiencing an event; p.m.: post meridiem; RR: risk ratio; T1DM: type 1 diabetes mellitus.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 3. ^bDowngraded by three levels because of inconsistency (point estimates varied widely, non‐consistent direction of effect) and serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 3. ^cDowngraded by two levels because of serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 3.

Summary of findings 3. Summary of findings: insulin detemir versus insulin glargine

Summary of findings 4. Summary of findings: insulin degludec versus insulin detemir

Outcomes	Insulin detemir	Insulin degludec	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin degludec compared with insulin detemir for T1DM
Patients people with T1DM Settings: outpatients Intervention: insulin degludec Comparison: insulin detemir
All‐cause mortality Follow‐up: 26 weeks	See comment			802 (2)	⊕⊕⊝⊝ low^a	No participant died 1 study included adults, 1 study included children
Health‐related quality of life Scale: Short‐Form 36 version 2 (higher values mean better health‐related quality of life) Follow‐up: 26 weeks	Physical health score: the mean score in the insulin detemir group was 52.5 Mental health score: the mean score in the insulin detemir group was 52.5	Physical health score: the mean score in the insulin degludec group was 0.60 points lower (1.83 points lower to 0.63 points higher) Mental health score: the mean score in the insulin degludec group was 3.00 points lower (4.44 points lower to 1.56 points lower)	—	454 (1)	⊕⊕⊝⊝ low^b	Physical health score: MID is 2‐3 points Mental health score: MID is 3 points Study included adults
Severe hypoglycaemia (n/N) Definition: hypoglycaemia requiring third party assistance (Davies 2014) or altered mental status and cannot assist in their own care, is semiconscious or unconscious, or in a coma ± convulsions and may require parenteral therapy (glucagon or iv glucose) (BEGIN Young) Follow‐up: 26 weeks	122 per 1000	143 per 1000 (99 to 207)	RR 1.17 (0.81 to 1.69)	802 (2)	⊕⊕⊝⊝ low^c	1 study included adults, 1 study included children (the test for subgroup differences did not indicate interaction)
Non‐fatal myocardial infarction/stroke Definition: non‐fatal myocardial infarction/stroke Follow‐up: 26 weeks	See comment			453 (1)	⊕⊕⊝⊝ low^a	No participant experienced a non‐fatal myocardial infarction or stroke Study included adults
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring 00:01‐05:59 (Davies 2014) or 23:00‐07:00 (BEGIN Young) Follow‐up: 26 weeks	31 per 1000	34 per 1000 (16 to 75)	RR 1.12 (0.51 to 2.46)	802 (2)	⊕⊕⊝⊝ low^c	1 study included children, 1 study included adults (the test for subgroup differences did not indicate interaction)
Serious adverse events (n/N) Follow‐up: 26 weeks	73 per 1000	92 per 1000 (56 to 150)	RR 1.25 (0.76 to 2.05)	802 (2)	⊕⊕⊝⊝ low^c	1 study included children, 1 study included adults (the test for subgroup differences did not indicate interaction)
HbA1c (%) Follow‐up: 26 weeks	The mean HbA1c in the insulin glargine groups was 7.3%	The mean HbA1c in the insulin detemir groups was 0.05% lower (0.1% lower to 0.2% higher)	—	802 (2)	⊕⊕⊝⊝ low^c	1 study included children, 1 study included adults (the test for subgroup differences did not indicate interaction)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). a.m.: ante meridiem; CI: confidence interval; HbA1c: glycosylated haemoglobin A1c; iv: intravenous; MID: minimal important difference; n/N: number of people experiencing an event; p.m.: post meridiem; RR: risk ratio; T1DM: type 1 diabetes mellitus.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 4. ^bDowngraded by two levels because of overall risk of bias ('some concerns') and imprecision (few studies) ‐ see Appendix 4. ^cDowngraded by two levels because of serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 4.

Summary of findings 4. Summary of findings: insulin degludec versus insulin detemir

Summary of findings 5. Summary of findings: insulin degludec versus insulin glargine

Outcomes	Insulin glargine	Insulin degludec	Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
Insulin degludec compared with insulin glargine for T1DM
Patients: people with T1DM Settings: outpatients Intervention: insulin degludec Comparison: insulin glargine
All‐cause mortality Follow‐up: 26 ‐ 52 weeks	3 per 1000	4 per 1000 (0 to 36)	Peto OR 1.34 (0.15 to 11.93)	973 (3)	⊕⊝⊝⊝ very low^a	A total of 3/646 participants in the insulin degludec group vs 1/327 participants in the insulin glargine group died 2 studies included adults 1 study included children
Health‐related quality of life Scale: Short‐Form 36 version 2 (higher values mean better health‐related quality of life) Follow‐up: 32 and 52 weeks	Physical health score: the mean score ranged across the insulin glargine groups from 50.6 to 51.8 Mental health score: the mean score ranged across the insulin glargine groups from 49.9 to 50.4	Physical health score: the mean score in the insulin degludec groups was 0.04 points lower (1.21 points lower to 1.13 points higher) Mental health score: the mean score in the insulin degludec groups was 0.09 points lower (1.03 points lower to 0.85 points higher)	—	1042 (2)	⊕⊝⊝⊝ very low^b	Physical health score: MID is 2‐3 points Mental health score: MID is 3 points 2 studies included adults
Severe hypoglycaemia (n/N) Definition: hypoglycaemia requiring third party assistance (BEGIN Flex T1; BEGIN Young) or an event associated with impaired consciousness or seizure (Urakami 2017) Follow‐up: 24 and 52 weeks	102 per 1000	124 per 1000 (83 to 185)	RR 1.22 (0.82 to 1.82)	970 (3)	⊕⊕⊝⊝ low^c	2 studies included adults 1 study including children reported no child experienced severe hypoglycaemia (Urakami 2017)
Non‐fatal myocardial infarction/stroke Definition: non‐fatal myocardial infarction/cerebral ischaemia Follow‐up: 24 and 52 weeks	See comment			970 (3)/970 (3)	⊕⊕⊝⊝ low^d	2 studies including adults reported 1/637 participants in the insulin degludec group vs 0/315 participants in the insulin glargine group experienced a non‐fatal myocardial infarction; there were no events in 1 study including children (Urakami 2017) 2 studies including adults reported 1/637 participants in the insulin degludec group vs 0/315 in the insulin glargine group experienced cerebral ischaemia; there were no events in 1 study including children (Urakami 2017)
Severe nocturnal hypoglycaemia (n/N) Definition: severe hypoglycaemia occurring from 22:00 to 06:59 h Follow‐up: 24 ‐ 52 weeks	25 per 1000	35 per 1000 (15 to 83)	RR 1.39 (0.59 to 3.27)	970 (3)	⊕⊕⊝⊝ low^c	2 studies included adults 1 study include children
Serious adverse events (n/N) Follow‐up: 24 and 52 weeks	77 per 1000	71 per 1000 (45 to 113)	RR 0.92 (0.58 to 1.46)	970 (3)	⊕⊕⊝⊝ low^c	2 studies included adults 1 study including children reported no child experienced a serious adverse event (Urakami 2017)
HbA1c (%) Follow‐up: 24 and 52 weeks	The mean HbA1c ranged across the insulin glargine groups from 6.9% to 7.8%	The mean HbA1c in the insulin degludec groups was 0.1% higher (0% lower to 0.2% higher)	—	1388 (4)	⊕⊕⊝⊝ low^c	The 95% prediction interval ranged between ‐0.1% and 0.3% 3 studies included adults, 1 study included children (the test for subgroup differences did not indicate interaction)
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; CSR: clinical study report; HbA1c: glycosylated haemoglobin A1c; MID: minimal important difference; n/N: number of people experiencing an event; OR: odds ratio; RR: risk ratio; T1DM: type 1 diabetes mellitus:
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^*Assumed risk was derived from the event rates in the comparator groups. ^aDowngraded by three levels because of indirectness (insufficient time frame) and serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 5. ^bDowngraded by three levels because of overall risk of bias ('some concerns') and serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 5. ^cDowngraded by two levels because of serious imprecision (CI consistent with benefit and harm, few studies) ‐ see Appendix 5. ^dDowngraded by two levels because of indirectness (insufficient time frame) and imprecision (few studies) ‐ see Appendix 5.

Summary of findings 5. Summary of findings: insulin degludec versus insulin glargine

Table 1. Overview of study populations

Study ID (study design)	Intervention(s) and comparator(s)	Description of power and sample size calculation	Screened/eligible (n)	Randomised (n)	Analysed primary outcome) (n)	Finishing study (n)	Randomised finishing study (%)	Follow‐up (extended follow‐up)^a
Bartley 2008 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "A total of 489 patients were needed to obtain 245 evaluable patients on detemir and 123 on NPH to detect a clinically relevant difference of 0.4% in HbA1c with a power of 85%, assuming a standard deviation (SD) for HbA1c of 1.2 and an expected drop‐out rate of 25%"	557	331	320	278	84.3	24 months
	C: NPH insulin		557	166	159	144	86.7
	total:			497	479	422	85.0
BEGIN Basal‐Bolus Type 1^b (parallel‐group non‐inferiority RCT)	I: insulin degludec	Quote: "Sample size was determined by the primary objective with the assumption of a one sided t test at a significance level of 2·5%, a zero mean treatment difference, and an SD of 1·1% for HbA1c. A total of 624 participants were needed for at least 95% power after adjustment for a 15% dropout rate"	722	472	472	404	85.6	52 weeks (104 weeks)
	C: insulin glargine		722	157	157	137	87.0
	total:			629	629	541	86.0
BEGIN Flex T1^c (parallel‐group non‐inferiority RCT)	I: insulin degludec	Quote: "Sample size was determined on the basis of the primary objective under the assumption of a 1‐sided t test of size 2.5%, a zero mean treatment difference, and standard deviation of 1.1% for HbA1c"	549	165	165	139	84.2	26 weeks (52 weeks)
	C: insulin glargine		549	164	164	152	92.7
	total:			329	329	291	88.4
BEGIN Young^d (parallel‐group non‐inferiority RCT)	I: insulin degludec	Quote: "The sample size was determined using a t‐statistic under the assumption of a one‐sided test of size 2.5%, a zero mean treatment difference and standard deviation (SD) of 1.25% for HbA1c. A total of 346 participants had to be randomized to achieve at least 80% or greater power in the evaluation of the per protocol (PP) analysis set, after adjustment for a 10% dropout rate"	363	174	174	170	97.7	26 weeks (52 weeks)
	C: insulin detemir		363	176	176	163	93.7
	total:			350	350	333	95.1
Bolli 2009 (parallel‐group superiority RCT)	I: insulin glargine	Quote: "The expected FBG difference in the two groups at the end of the study treatment was estimated to be 30+/‐60 mg/dL. Using a two‐sided test with ɑ = 0.01 and ß = 0.1 (i.e., power: 1‐ß = 0.9), 240 evaluable patients were to be included. Due to an expected dropout rate of 20% and to the randomization schedule, which was restricted and stratified by centre (26 centres), 312 patients were planned to be enrolled"	213	85	85	78	91.8	24 weeks (30 weeks)
	C: NPH insulin		213	90	90	74	82.2
	total:			175	175	152	86.7
Chase 2008 (parallel‐group non‐inferiority RCT)	I: insulin glargine	Quote: "The primary clinical outcome (the mean change in A1C from baseline [week 0] to endpoint [week 24 or last post randomization assessment]) was compared in the 2 treatment groups using analysis of covariance (ANCOVA), with treatment group, study centre (pooled), CGMS values, sex, and baseline value as covariates (α = 0.05; 2‐sided test). The 95% confidence intervals (CIs) were computed for the adjusted mean difference between treatment groups from the ANCOVA to test for noninferiority (defined as an upper bound of the 95% CI for the mean difference in A1C of ≤ 0.4%)"	235	85	84	81	95.3	24 weeks (25 weeks)
	C: NPH insulin/Lente		235	90	84	76	84.4
	total:			175	168	157	89.7
Davies 2014^e (parallel‐group non‐inferiority RCT)	I: insulin degludec	Quote: "Assuming a standard deviation (SD) of 1.1% for the primary endpoint, the trial had 90% power with 360 participants randomized 2:1"	512	303	302	283	93.4	26 weeks (52 weeks)
	C: insulin detemir		512	153	153	138	90.2
	total:			456	455	421	92.5
Fulcher 2005 (parallel‐group non‐inferiority RCT)	I: insulin glargine	Quote: "The sample size was calculated assuming a 20% dropout rate, so that 118 patients (59 in each group) were enrolled in order to have 96 patients (48 in each group) available for evaluation at end‐point. Assuming a SD of 1.2 for HbA1c (based on previous Phase IIIa studies), the study had 80% power to detect a 0.7% difference in HbA1c"	173^f	62	62	58	94	30 weeks
	C: NPH insulin		173^f	63	62	49	78
	total:			125	124^f	107	85.6
Heller 2009 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The sample size was determined for 2:1 (detemir:glargine) randomization and based on a 1‐sided t test at a 2.5% significance level. Assuming an SD of 1.0% for HbA1c and a dropout rate of 15%, a sample size of 435 patients gave 95% power to demonstrate noninferiority"	515	300	299	263	87.7	52 weeks
	C: insulin glargine		515	147	144	122	83.0
	total:			447	443	385	86.1
Home 2005 (parallel‐group superiority RCT)	I: insulin glargine	Quote from CSR: "It was planned to treat 520 subjects, 260 subjects in each group. Each investigation site was to randomise 10‐20 subjects.The primary efficacy variable for the comparison between HOE 901 and NPH insulin was the change from baseline in GHb at the study endpoint for the individual subject ... The standard deviation for change from baseline in GHb at endpoint was estimated to be 1.6%. Based on 1:1 randomization and using a t‐test, a total number of 440 subjects (220 subjects for each group) was required to detect a mean difference of 0.5% GHb between HOE 901 and NPH with a type I error of α = 5% and a statistical power of 90%. With an expected drop‐out rate of 15% during the course of the study, a total number of 520 subjects (260 subjects in each group) were to be enrolled in order to have 440 subjects (220 subjects in each group) evaluable at week 28"	655	298	292	276	94.5	28 weeks
	C: NPH insulin		655	305	293	272	92.8
	total:			603^g	585^h	548	93.6
Kobayashi 2007 (parallel‐group non‐inferiority RCT)	I: insulin detemir	—	454ⁱ	197	195	183	93.4	48 weeks
	C: NPH insulin	—	454ⁱ	99	98	91	92.9
	total:			296	293	274	92.6
Liu 2016 (parallel‐group non‐inferiority RCT)	I: insulin glargine	Quote from CSR: "The planned sample size was reduced from 366 to 150 patients in view of extremely difficult recruitment progress over the 2 years since first patient’s enrolment..."	196	107	108	106	99.1	24 weeks (25 weeks)
	C: NPH insulin		196	55	54	50	90.9
	total:			162	161	156	96.3
NCT00595374^f (parallel‐group non‐inferiority RCT)	I: insulin detemir	—	124	75	—	70	93.3	26 weeks
	C: NPH insulin	—	124	38	—	34	92.1
	total:			113	—	104	92.0
NCT00605137^f (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote from trial protocol: "This power calculation is based on a two‐sample poisson test at a significance level of 5% for the comparison of the mean rate of nocturnal episodes per four weeks although nocturnal episodes will be analysed as recurrent events using gamma frailty model in the trial analysis"	88	57	55	55	96.5	24 weeks
	C: NPH insulin		88	29	27	27	93.1
	total:			86	82	82	95.3
Pieber 2007 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The sample size was determined in order to test non‐inferiority in a 1:1 randomization. Assuming a standard deviation for HbA1c of 1.2% and a clinically relevant, absolute difference in HbA1c of 0.4%, a total of 286 randomized participants were needed to achieve a power of 80%. Assuming a 10% drop‐out rate, 159 randomized participants were needed in each group"	415	161	161	147	91.3	26 weeks
	C: insulin glargine		415	161	159	146	90.7
	total:			322	319	293	91.0
Porcellati 2004 (parallel‐group superiority RCT)	I: insulin glargine	Quote: "In this design, a total of 120 participants were required to achieve 90% power to detect a difference of 0.3% among the means with group standard deviations of 0.4 at the significance level (alpha) of 5%"	130	61	61	61	100	1 year
	C: NPH insulin		130	60	60	60	100
	total:			121	121	121	100
PRESCHOOL (parallel‐group non‐inferiority RCT)	I: insulin glargine	Quote: "Sample size calculation was based on an expected composite hypoglycemia rate of 0.8 events/100 patient‐yr of exposure to insulin glargine or to NPH insulin. The sample size and novel composite outcome was planned to ensure sufficient power so that the upper bound of the two‐sided 95% confidence interval (CI) for the insulin glargine:NPH ratio of the mean composite hypoglycemia rates for the comparison of treatment groups would not exceed 1.15. A sample size of 35 completed patients per treatment group was to provide 96% power to demonstrate noninferiority of insulin glargine vs. NPH"	165	61	61	57	93.4	24 weeks (26 weeks)
	C: NPH insulin		165	64	64	54	84.4
	total:			125^j	125	111	88.8
Ratner 2000 (parallel‐group superiority RCT)	I: insulin glargine	Quote: "An estimated 440 participants (220 in each treatment group) were required to detect a mean difference of 0.5% in GHb levels between treatment with a type 1 error of α = 5% and a statistical power of 90%"	677^f	266	256	233	88.3	28 weeks
	C: NPH insulin		677^f	274	262	248	91.9
	total:			540^k	518	481^f	90.1
Robertson 2007 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The sample size was determined for a 2: 1 randomization based on a two‐sided t ‐test on a 5% significance level. Assuming a SD for HbA 1c of 1.1% and a clinically relevant difference in HbA 1c of 0.4% (absolute), 270 children were needed to achieve a power of 80%. With an expected drop‐out rate of 20%, 338 children were to be allocated to study treatment"	363^f	232	232	226	97.4	26 weeks
	C: NPH insulin		363^f	115	114	109	94.8
	total:			347	347	335^f	96.5
Russell‐Jones 2004 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "Sample size was based on an SD for HbA1c of 1.4% and the assumption that a 0.4% absolute difference in HbA1c represents a clinically relevant difference" and " All comparisons were 2‐tailed tests with a 5% level of significance"	838^f	492	491	465	94.7	6 months
	C: NPH insulin		838^f	257	256	235	91.8
	total:			749^f	747	700	93.5
Schober 2002 (parallel‐group superiority RCT)	I: insulin glargine	Quote: "The sample size was calculated to detect a mean difference in HbA1C from baseline to endpoint of 0.5% with a statistical power of 90%. Assuming a 20% dropout rate, the minimum sample size required was 360 patients"	385	180	155	169	93.9	28 weeks
	C: NPH insulin		385	181	156	168	92.8
	total:			361^l	311	337^f	93.4
Standl 2004^m (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote from CSR: "A total of 440 type 1 participants were planned for randomisation in order to obtain 400 evaluable participants, assuming a dropout rate of approximately 10%"	505^f	237	210	212	89.5	6 months (12 months)
	C: NPH insulin		505^f	224	206	209	93.3
	total:			461^f	416^f	421	91.3
SWITCH 1ⁿ (cross‐over non‐inferiority RCT)	I: insulin degludec	Quote: "The trial was powered to show noninferiority of the primary end point. Based on the assumption that up to 10% of the randomised patients may not contribute to the analysis, 400 patients needed to contribute to the analysis if 446 patients were randomised to ensure a power of 94%, to demonstrate noninferiority with an expected rate of overall symptomatic hypoglycemia of 5.0 episodes per patient‐years’ exposure (PYE)"	634	249	249	209	83.9	32 weeks
	C: insulin glargine		634	252	251	205	81.3
	total:			501	414	414	82.6
Thalange 2013 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The power calculation was analysed on this basis: using a two‐sided t‐test with a one‐sided significance level of 2.5%, assuming SD of 1.1, a non‐inferiority criterion of 0.4%, a power of 85% and an expected dropout rate of 20%, a total of 344 children were to be randomized"	381	177	171	164	92.7	52 weeks (104 weeks)
	C: NPH insulin		381	171	168	161	94.2
	total:			348	339	325	93.4
Urakami 2017^p (cross‐over superiority RCT)	I: insulin degludec	—	—	9	9	9	100	24 weeks
	C: insulin glargine	—	—	9	9	9	100
	total:			18	18	18	100
Vague 2003 (parallel‐group non‐inferiority RCT)	I: insulin detemir	Quote: "The initial cohort size was calculated to achieve a power of 85% on the basis of non‐inferiority testing at the 5% significance level and a 2:1 randomization"	471^f	301	280	284	94.4	6 months (12 months)
	C: NPH insulin		471^f	147	139	141	96.6
	total:			448	419	425	95.1
Overall total	All insulin detemir			2889		2648
	All insulin degludec			1372		1214
	All insulin glargine			2095		1890
	All NPH insulin			2428		2202
	All interventions and comparators			8784		7954
— denotes not reported ^aFollow‐up under randomised conditions until end of study (= duration of intervention + follow‐up post‐intervention or identical to duration of intervention); extended follow‐up refers to follow‐up of participants once the original study was terminated as specified in the power calculation. ^bData in the table are for the main period.After 52 weeks, the participants of the initial study were invited to an extension study. 74% in the degludec and 75% in the glargine participated. Of the one included in the extension period, 94% (330/351) participants completed in the degludec group and 96% (113/118) participants in the glargine group. ^cAn additional study arm existed, which was not included in this review. ^dData in the table are for the main period. In the insulin degludec group, 152 participants entered the extension study and 151 participants completed; in the insulin detemir group, 128 participants entered the extension study and 122 participants completed. ^eData in the table are for the main period. In the insulin degludec group, 248 participants entered the extension study and 242 participants completed (79.9% of those initially randomised); in the insulin detemir group, 122 participants entered the extension study and 115 participants completed (75.2% of those initially randomised). ^fData from clinical study report/synopsis. ^gIn the publication, it was only mentioned that 602 participants were randomised, but not explained how these were divided between the intervention groups. This was reported in the clinical study report. In the publication, there was only information about the allocation of the 585 participants who received the intervention. ^hIn the main publication, the number of participants analysed was not clearly described; this number was provided by the clinical study report. ⁱBoth people with type 1 diabetes mellitus and type 2 diabetes mellitus were screened. ^jOne participant randomised to NPH insulin was actually treated with insulin glargine, thus the safety population comprised 62 participants for insulin glargine and 63 participants for NPH insulin. ^kIn the main publication, it was stated that 534 participants were randomised (264 participants allocated to insulin glargine; 270 participants allocated to NPH insulin). In the clinical study report, it was stated that a total of 540 participants were randomised, but six were never treated (2 participants in the insulin glargine group; 4 participants in the NPH insulin group). ^lOf the 361 participants randomised, 12 withdrew their consent before being treated, therefore a total 349 participants were treated: 174 participants in the glargine group compared with 175 participants in the NPH group. ^mData in the table are for the main period. In the insulin detemir group, 154 participants entered the extension study and 118 participants completed (49.8% of those initially randomised); in the NPH insulin group, 135 participants entered the extension study and 134 participants completed (59.8% of those initially randomised). ⁿData from first treatment period before cross‐over (32 weeks). ^oExtension only performed for the detemir group. ^pNot reported if any participant dropped out during the study. All randomised participants were included in all analyses. A1c: glycosylated haemoglobin A1c ANCOVA: analysis of covariance C: comparator CGMS: continuous glucose monitoring system CI: confidence interval CSR: clinical study report FBG: fasting blood glucose GHb: glycated haemoglobin HbA1c: glycosylated haemoglobin A1c HOE 901: insulin glargine I: intervention NPH: neutral protamine Hagedorn PYE: patient‐years’ exposure RCT: randomised controlled trial SD: standard deviation vs: versus

Table 1. Overview of study populations

Comparison 1. Insulin detemir versus NPH insulin

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 All‐cause mortality Show forest plot	9	3334	Peto Odds Ratio (Peto, Fixed, 95% CI)	4.97 [0.79, 31.38]

1.1.1 Adults	6	2558	Peto Odds Ratio (Peto, Fixed, 95% CI)	4.97 [0.79, 31.38]
1.1.2 Children	3	776	Peto Odds Ratio (Peto, Fixed, 95% CI)	Not estimable
1.2 All‐cause mortality (published vs. unpublished data) Show forest plot	9	3334	Risk Ratio (M‐H, Random, 95% CI)	3.64 [0.42, 31.40]

1.2.1 Published	2	842	Risk Ratio (M‐H, Random, 95% CI)	4.47 [0.24, 82.58]
1.2.2 Unpublished	7	2492	Risk Ratio (M‐H, Random, 95% CI)	2.85 [0.12, 69.55]
1.3 Severe hypoglycaemia Show forest plot	8	3219	Risk Ratio (M‐H, Random, 95% CI)	0.69 [0.52, 0.92]

1.3.1 Adults	5	2443	Risk Ratio (M‐H, Random, 95% CI)	0.71 [0.49, 1.03]
1.3.2 Children	3	776	Risk Ratio (M‐H, Random, 95% CI)	0.61 [0.30, 1.23]
1.4 Severe hypoglycaemia (published vs. unpublished data) Show forest plot	8	3175	Risk Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.93]

1.4.1 Published	6	2677	Risk Ratio (M‐H, Random, 95% CI)	0.62 [0.50, 0.78]
1.4.2 Unpublished	2	498	Risk Ratio (M‐H, Random, 95% CI)	1.42 [0.77, 2.62]
1.5 Hypoglycaemia reported as a serious adverse event Show forest plot	8	3219	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.51, 1.71]

1.5.1 Adults	5	2443	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.48, 1.86]
1.5.2 Children	3	776	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.16, 5.57]
1.6 Cardiovascular mortality Show forest plot	9		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

1.6.1 Adults	6		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
1.6.2 Children	3		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
1.7 Non‐fatal myocardial infarction Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

1.8 Serious adverse events Show forest plot	9	3332	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.75, 1.21]

1.8.1 Adults	6	2556	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.73, 1.28]
1.8.2 Children	3	776	Risk Ratio (M‐H, Random, 95% CI)	0.89 [0.56, 1.43]
1.9 Serious adverse events (published vs. unpublished data) Show forest plot	9	3332	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.75, 1.21]

1.9.1 Published	2	641	Risk Ratio (M‐H, Random, 95% CI)	0.66 [0.40, 1.09]
1.9.2 Unpublished	7	2691	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.80, 1.39]
1.10 Diabetic ketoacidosis Show forest plot	6	2012	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.36, 1.76]

1.10.1 Adults	3	1236	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.24, 2.92]
1.10.2 Children	3	776	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.27, 2.15]
1.11 Diabetic ketoacidosis (published vs. unpublished data) Show forest plot	6	2012	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.36, 1.76]

1.11.1 Published data	2	694	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.27, 2.52]
1.11.2 Unpublished data	4	1318	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.25, 2.38]
1.12 Non‐serious adverse events Show forest plot	9	3332	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.95, 1.01]

1.12.1 Adults	6	2556	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.95, 1.03]
1.12.2 Children	3	776	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.90, 1.02]
1.13 Non‐serious adverse events (published vs unpublished data) Show forest plot	9	3332	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.94, 1.01]

1.13.1 Published data	3	1141	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.90, 1.01]
1.13.2 Unpublished data	6	2191	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.95, 1.04]
1.14 Withdrawals due to adverse events Show forest plot	8	3222	Risk Ratio (M‐H, Random, 95% CI)	2.23 [0.98, 5.05]

1.14.1 Adults	5	2445	Risk Ratio (M‐H, Random, 95% CI)	2.25 [0.94, 5.41]
1.14.2 Children	3	777	Risk Ratio (M‐H, Random, 95% CI)	2.08 [0.22, 19.90]
1.15 Any nocturnal hypoglycaemia Show forest plot	8	3219	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.87, 0.95]

1.15.1 Adults	5	2443	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.88, 0.98]
1.15.2 Children	3	776	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.81, 0.94]
1.16 Mild nocturnal hypoglycaemia Show forest plot	7	3073	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.85, 0.96]

1.16.1 Adults	4	2149	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.83, 1.00]
1.16.2 Children	3	924	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.78, 1.00]
1.17 Nocturnal hypoglycaemia (symptoms) Show forest plot	6	2578	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.79, 0.98]

1.17.1 Adults	4	2149	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.82, 1.01]
1.17.2 Children	2	429	Risk Ratio (M‐H, Random, 95% CI)	0.55 [0.19, 1.61]
1.18 Severe nocturnal hypoglycaemia Show forest plot	7	2925	Risk Ratio (M‐H, Random, 95% CI)	0.67 [0.39, 1.17]

1.18.1 Adults	4	2149	Risk Ratio (M‐H, Random, 95% CI)	0.57 [0.35, 0.93]
1.18.2 Children	3	776	Risk Ratio (M‐H, Random, 95% CI)	0.64 [0.13, 3.17]
1.19 Any nocturnal hypoglycaemia (published vs. unpublished data) Show forest plot	8	3219	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.87, 0.95]

1.19.1 Published	6	2677	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.86, 0.95]
1.19.2 Unpublished	2	542	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.80, 1.04]
1.20 Mild nocturnal hypoglycaemia (published vs. unpublished data) Show forest plot	7	3073	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.85, 0.96]

1.20.1 Published	4	2084	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.85, 0.98]
1.20.2 Unpublished	3	989	Risk Ratio (M‐H, Random, 95% CI)	0.89 [0.75, 1.07]
1.21 Nocturnal hypoglycaemia, symptoms only (published vs. unpublished data) Show forest plot	6	2578	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.79, 0.98]

1.21.1 Published	3	1589	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.81, 0.99]
1.21.2 Unpublished	3	989	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.57, 1.08]
1.22 Severe nocturnal hypoglycaemia (published vs. unpublished data) Show forest plot	7	2925	Risk Ratio (M‐H, Random, 95% CI)	0.67 [0.39, 1.17]

1.22.1 Published	5	2383	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.32, 1.25]
1.22.2 Unpublished	2	542	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.33, 2.45]
1.23 Nocturnal hypoglycaemia, asymptomatic (children vs. adults) Show forest plot	2	429	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.78, 1.18]

1.24 Mild/moderate hypoglycaemia Show forest plot	8	3219	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.94, 0.99]

1.24.1 Adults	5	2443	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.93, 1.02]
1.24.2 Children	3	776	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.93, 1.01]
1.25 Mild/moderate hypoglycaemia (published vs. unpublished data) Show forest plot	8	3219	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.94, 0.99]

1.25.1 Published	6	2677	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.93, 1.00]
1.25.2 Unpublished	2	542	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.92, 1.05]
1.26 HbA1c Show forest plot	8	3122	Mean Difference (IV, Random, 95% CI)	0.01 [‐0.08, 0.10]

1.26.1 Adults	5	2354	Mean Difference (IV, Random, 95% CI)	‐0.03 [‐0.14, 0.07]
1.26.2 Children	3	768	Mean Difference (IV, Random, 95% CI)	0.13 [‐0.04, 0.31]
1.27 HbA1c (published vs. unpublished data) Show forest plot	8	3122	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.09, 0.09]

1.27.1 Published	6	2624	Mean Difference (IV, Random, 95% CI)	‐0.02 [‐0.13, 0.09]
1.27.2 Unpublished	2	498	Mean Difference (IV, Random, 95% CI)	0.10 [‐0.08, 0.28]

Comparison 1. Insulin detemir versus NPH insulin

Comparison 2. Insulin glargine versus NPH insulin

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 All‐cause mortality Show forest plot	8	2175	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.14 [0.00, 6.98]

2.1.1 Adults	4	1365	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.14 [0.00, 6.98]
2.1.2 Children	4	810	Peto Odds Ratio (Peto, Fixed, 95% CI)	Not estimable
2.2 Health‐realted quality of life Show forest plot	2	880	Mean Difference (IV, Random, 95% CI)	0.62 [‐0.71, 1.96]

2.3 Severe hypoglycaemia Show forest plot	9	2350	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.67, 1.04]

2.3.1 Adults	5	1540	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.58, 1.05]
2.3.2 Children	4	810	Risk Ratio (M‐H, Random, 95% CI)	1.14 [0.59, 2.21]
2.4 Severe hypoglycaemia (published vs. unpublished data) Show forest plot	9	2350	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.67, 1.04]

2.4.1 Published	7	1691	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.63, 1.22]
2.4.2 Unpublished	2	659	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.56, 1.25]
2.5 Hypoglycaemia reported as a serious adverse event Show forest plot	8	2229	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.64, 1.39]

2.5.1 Adults	4	1419	Risk Ratio (M‐H, Random, 95% CI)	0.89 [0.57, 1.37]
2.5.2 Children	4	810	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.32, 2.87]
2.6 Cardiovascular mortality Show forest plot	8		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

2.6.1 Adults	4		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.6.2 Children	4		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.7 Non‐fatal myocardial infarction Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

2.8 Non‐fatal stroke Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

2.9 Serious adverse events Show forest plot	8	2229	Risk Ratio (M‐H, Random, 95% CI)	1.08 [0.63, 1.84]

2.9.1 Adults	4	1419	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.72, 1.35]
2.9.2 Children	4	810	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.28, 3.64]
2.10 Serious adverse events (published vs. unpublished data) Show forest plot	8	2229	Risk Ratio (M‐H, Random, 95% CI)	1.08 [0.63, 1.84]

2.10.1 Published	4	1284	Risk Ratio (M‐H, Random, 95% CI)	1.11 [0.45, 2.70]
2.10.2 Unpublished	4	945	Risk Ratio (M‐H, Random, 95% CI)	1.10 [0.46, 2.60]
2.11 Diabetic ketoacidosis Show forest plot	7	2054	Risk Ratio (M‐H, Random, 95% CI)	0.53 [0.19, 1.44]

2.11.1 Adults	3	1244	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.11, 9.58]
2.11.2 Children	4	810	Risk Ratio (M‐H, Random, 95% CI)	0.45 [0.15, 1.39]
2.12 Diabetic ketoacidosis (published vs. unpublished data) Show forest plot	7	2054	Risk Ratio (M‐H, Random, 95% CI)	0.53 [0.19, 1.44]

2.12.1 Published	3	685	Risk Ratio (M‐H, Random, 95% CI)	0.39 [0.11, 1.31]
2.12.2 Unpublished	4	1369	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.18, 5.77]
2.13 Non‐serious adverse events Show forest plot	8	2229	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.96, 1.06]

2.13.1 Adults	4	1419	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.95, 1.07]
2.13.2 Children	4	810	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.93, 1.12]
2.14 Non‐serious adverse events (published vs. unpublished data) Show forest plot	8	2229	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.96, 1.06]

2.14.1 Published	5	1308	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.94, 1.05]
2.14.2 Unpublished	3	921	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.94, 1.14]
2.15 Withdrawals due to adverse events Show forest plot	8	2230	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.24, 2.81]

2.15.1 Adults	4	1419	Risk Ratio (M‐H, Random, 95% CI)	1.74 [0.29, 10.39]
2.15.2 Children	4	811	Risk Ratio (M‐H, Random, 95% CI)	0.30 [0.06, 1.53]
2.16 Nocturnal hypoglycaemia Show forest plot	7	2054	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.96, 1.05]

2.16.1 Adults	3	1244	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.92, 1.06]
2.16.2 Children	4	810	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.95, 1.08]
2.17 Mild nocturnal hypoglycaemia Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

2.18 Nocturnal hypoglycaemia (symptoms) Show forest plot	4	996	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.82, 1.05]

2.18.1 Adults	2	710	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.88, 1.08]
2.18.2 Children	2	286	Risk Ratio (M‐H, Random, 95% CI)	0.74 [0.55, 1.00]
2.19 Severe nocturnal hypoglycaemia Show forest plot	6	1893	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.62, 1.12]

2.19.1 Adults	3	1244	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.60, 1.27]
2.19.2 Children	3	649	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.47, 1.25]
2.20 Nocturnal hypoglycaemia (published vs. unpublished data) Show forest plot	7	2054	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.96, 1.05]

2.20.1 Published	5	1345	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.95, 1.06]
2.20.2 Unpublished	2	709	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.91, 1.08]
2.21 Symptomatic nocturnal hypoglycaemia (published vs. unpublished data) Show forest plot	4	996	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.82, 1.05]

2.21.1 Published	3	871	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.67, 1.12]
2.21.2 Unpublished	1	125	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.80, 1.10]
2.22 Mild/moderate hypoglycaemia Show forest plot	7	2054	Risk Ratio (M‐H, Random, 95% CI)	1.02 [1.00, 1.04]

2.22.1 Adults	3	1244	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.99, 1.06]
2.22.2 Children	4	810	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.99, 1.04]
2.23 Mild/moderate hypoglycaemia (published vs. unpublished data) Show forest plot	7	2054	Risk Ratio (M‐H, Random, 95% CI)	1.02 [1.00, 1.04]

2.23.1 Published	5	1395	Risk Ratio (M‐H, Random, 95% CI)	1.02 [1.00, 1.05]
2.23.2 Unpublished	2	659	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.98, 1.05]
2.24 HbA1c Show forest plot	9	2285	Mean Difference (IV, Random, 95% CI)	0.02 [‐0.06, 0.11]

2.24.1 Adults	5	1523	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.16, 0.13]
2.24.2 Children	4	762	Mean Difference (IV, Random, 95% CI)	0.03 [‐0.13, 0.20]
2.25 HbA1c (published vs unpublished data) Show forest plot	9	2285	Mean Difference (IV, Random, 95% CI)	0.02 [‐0.06, 0.11]

2.25.1 Published	6	1868	Mean Difference (IV, Random, 95% CI)	0.02 [‐0.09, 0.14]
2.25.2 Unpublished	3	417	Mean Difference (IV, Random, 95% CI)	‐0.04 [‐0.26, 0.18]
2.26 HbA1c (NPH < 2x/day vs ≥ 2x/day) Show forest plot	9	2285	Mean Difference (IV, Random, 95% CI)	0.02 [‐0.06, 0.11]

2.26.1 NPH up to twice a day	8	2164	Mean Difference (IV, Random, 95% CI)	0.05 [‐0.03, 0.13]
2.26.2 NPH more than twice a day	1	121	Mean Difference (IV, Random, 95% CI)	‐0.50 [‐0.93, ‐0.07]

Comparison 2. Insulin glargine versus NPH insulin

Comparison 3. Insulin detemir versus insulin glargine

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 All‐cause mortality Show forest plot	2		Peto Odds Ratio (Peto, Fixed, 95% CI)	Totals not selected

3.2 Severe hypoglycaemia Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	0.59 [0.13, 2.63]

3.3 Severe hypoglycaemia (published vs. unpublished data) Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	0.59 [0.13, 2.63]

3.3.1 Published	1	320	Risk Ratio (M‐H, Random, 95% CI)	0.25 [0.07, 0.86]
3.3.2 Unpublished	1	443	Risk Ratio (M‐H, Random, 95% CI)	1.13 [0.72, 1.77]
3.4 Hypoglycaemia reported as a serious adverse event Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.16 [0.14, 9.48]

3.5 Cardiovascular mortality Show forest plot	2		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

3.6 Non‐fatal myocardial infarction Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

3.7 Non‐fatal stroke Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

3.8 Serious adverse events Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.72 [0.91, 3.28]

3.9 Diabetic ketoacidosis Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

3.10 Non‐serious adverse events Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.93, 1.09]

3.11 Non‐serious adverse events (published vs. unpublished data) Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.93, 1.09]

3.11.1 Published	1	443	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.97, 1.10]
3.11.2 Unpublished	1	320	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.84, 1.09]
3.12 Withdrawals due to adverse events Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.31, 3.67]

3.13 Any nocturnal hypoglycaemia Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.93, 1.09]

3.13.1 Published	1	320	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.75, 1.17]
3.13.2 Unpublished	1	443	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.94, 1.11]
3.14 Confirmed nocturnal hypoglycaemia (PG < 3.1 mmol/L and no assistance) Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.92, 1.10]

3.14.1 Published	1	320	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.71, 1.16]
3.14.2 Unpublished	1	443	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.93, 1.12]
3.15 Symptomatic nocturnal hypoglycaemia (PG ≥ 3.1 or no PG and no assistance required) Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.81, 1.29]

3.15.1 Published	1	320	Risk Ratio (M‐H, Random, 95% CI)	1.29 [0.78, 2.12]
3.15.2 Unpublished	1	443	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.77, 1.21]
3.16 Severe nocturnal hypoglycaemia Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	0.55 [0.06, 5.12]

3.16.1 Published	1	320	Risk Ratio (M‐H, Random, 95% CI)	0.11 [0.01, 2.02]
3.16.2 Unpublished	1	443	Risk Ratio (M‐H, Random, 95% CI)	1.18 [0.60, 2.32]
3.17 Mild/moderate hypoglycaemia Show forest plot	2	763	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.94, 1.14]

3.17.1 Published	1	320	Risk Ratio (M‐H, Random, 95% CI)	1.10 [0.95, 1.26]
3.17.2 Unpublished	1	443	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.96, 1.06]
3.18 HbA1c Show forest plot	2	717	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.13, 0.12]

3.19 Individuals with HbA1c < 7% without severe hypoglycaemia Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

Comparison 3. Insulin detemir versus insulin glargine

Comparison 4. Insulin degludec versus insulin detemir

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 All‐cause mortality Show forest plot	2		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

4.1.1 Adults	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.1.2 Children	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.2 Health‐related quality of life Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

4.2.1 Physical health score	1	454	Mean Difference (IV, Fixed, 95% CI)	‐0.60 [‐1.83, 0.63]
4.2.2 Mental health score	1	454	Mean Difference (IV, Fixed, 95% CI)	‐3.00 [‐4.44, ‐1.56]
4.3 Severe hypoglycaemia Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	1.17 [0.81, 1.69]

4.3.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.57, 1.78]
4.3.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	1.30 [0.80, 2.12]
4.4 Hypoglycaemia reported as a serious adverse event Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.37, 2.32]

4.4.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	0.69 [0.29, 1.69]
4.4.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	2.01 [0.37, 10.84]
4.5 Cardiovascular mortality Show forest plot	2		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

4.5.1 Adults	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.5.2 Children	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.6 Non‐fatal myocardial infarction Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

4.7 Non‐fatal stroke Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

4.8 End stage renal disease Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

4.9 Blindness Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

4.10 Serious adverse events Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	1.25 [0.76, 2.05]

4.10.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	1.45 [0.67, 3.17]
4.10.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	1.13 [0.60, 2.15]
4.11 Diabetic ketoacidosis Show forest plot	2		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

4.11.1 Adults	1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
4.11.2 Children	1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
4.12 Non‐serious adverse events Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.96, 1.08]

4.12.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.88, 1.11]
4.12.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.97, 1.10]
4.13 Withdrawals due to adverse events Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	2.32 [0.38, 14.18]

4.13.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	1.51 [0.16, 14.44]
4.13.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	5.03 [0.24, 103.99]
4.14 Nocturnal hypoglycaemia Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.94, 1.15]

4.14.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.85, 1.18]
4.14.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.94, 1.21]
4.15 Mild nocturnal hypoglycaemia Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.86, 1.10]

4.15.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.85, 1.16]
4.15.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.79, 1.15]
4.16 Nocturnal hypoglycaemia (symptomatic) Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	0.72 [0.15, 3.59]

4.16.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	0.38 [0.20, 0.72]
4.16.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	2.01 [0.37, 10.84]
4.17 Nocturnal hypoglycaemia (asymptomatic) Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.80, 1.03]

4.17.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.70, 1.23]
4.17.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.79, 1.04]
4.18 Severe nocturnal hypoglycaemia Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.51, 2.46]

4.18.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	1.21 [0.43, 3.38]
4.18.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.30, 3.41]
4.19 Mild/moderate hypoglycaemia Show forest plot	2	802	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.99, 1.05]

4.19.1 Adults	1	453	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.96, 1.08]
4.19.2 Children	1	349	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.99, 1.06]
4.20 HbA1c Show forest plot	2	805	Mean Difference (IV, Random, 95% CI)	0.05 [‐0.08, 0.18]

4.20.1 Adults	1	455	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.18, 0.18]
4.20.2 Children	1	350	Mean Difference (IV, Random, 95% CI)	0.11 [‐0.08, 0.30]
4.21 Individuals with HbA1c < 7% without severe hypoglycaemia Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

Comparison 4. Insulin degludec versus insulin detemir

Comparison 5. Insulin degludec versus insulin glargine

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
5.1 All‐cause mortality Show forest plot	3	973	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.34 [0.15, 11.93]

5.2 All‐cause mortality (published vs. unpublished data) Show forest plot	3	973	Risk Ratio (M‐H, Random, 95% CI)	1.13 [0.17, 7.65]

5.2.1 Published	1	626	Risk Ratio (M‐H, Random, 95% CI)	0.65 [0.06, 7.15]
5.2.2 Unpublished	2	347	Risk Ratio (M‐H, Random, 95% CI)	2.98 [0.12, 72.67]
5.3 Health‐related quality of life (physical health) Show forest plot	2	1043	Mean Difference (IV, Random, 95% CI)	‐0.04 [‐1.21, 1.13]

5.3.1 Published	1	629	Mean Difference (IV, Random, 95% CI)	0.50 [‐0.93, 1.93]
5.3.2 Unpublished	1	414	Mean Difference (IV, Random, 95% CI)	‐0.70 [‐2.30, 0.90]
5.4 Health‐related quality of life (mental health) Show forest plot	2	1539	Mean Difference (IV, Random, 95% CI)	‐0.09 [‐1.03, 0.85]

5.4.1 Published	1	629	Mean Difference (IV, Random, 95% CI)	0.40 [‐1.33, 2.13]
5.4.2 Unpublished	1	910	Mean Difference (IV, Random, 95% CI)	‐0.30 [‐1.42, 0.82]
5.5 Severe hypoglycaemia Show forest plot	3	970	Risk Ratio (M‐H, Random, 95% CI)	1.22 [0.82, 1.82]

5.5.1 Adults	2	952	Risk Ratio (M‐H, Random, 95% CI)	1.22 [0.82, 1.82]
5.5.2 Children	1	18	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
5.6 Hypoglycaemia reported as a serious adverse event Show forest plot	4	1884	Risk Ratio (M‐H, Random, 95% CI)	0.81 [0.40, 1.66]

5.6.1 Adults	3	1866	Risk Ratio (M‐H, Random, 95% CI)	0.81 [0.40, 1.66]
5.6.2 Children	1	18	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
5.7 Cardiovascular mortality Show forest plot	3		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

5.7.1 Adults	2		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
5.7.2 Children	1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
5.8 Non‐fatal myocardial infarction Show forest plot	3		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

5.8.1 Adults	2		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
5.8.2 Children	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
5.9 Non‐fatal stroke Show forest plot	3		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

5.10 Serious adverse events Show forest plot	3	970	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.58, 1.46]

5.10.1 Adults	2	952	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.58, 1.46]
5.10.2 Children	1	18	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
5.11 Diabetic ketoacidosis Show forest plot	3	970	Risk Ratio (M‐H, Random, 95% CI)	0.57 [0.05, 6.89]

5.11.1 Adults	2	952	Risk Ratio (M‐H, Random, 95% CI)	0.57 [0.05, 6.89]
5.11.2 Children	1	18	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
5.12 Diabetic ketoacidosis (published vs. unpublished data) Show forest plot	3	970	Risk Ratio (M‐H, Random, 95% CI)	0.57 [0.05, 6.89]

5.12.1 Published	1	626	Risk Ratio (M‐H, Random, 95% CI)	0.22 [0.04, 1.29]
5.12.2 Unpublished	2	344	Risk Ratio (M‐H, Random, 95% CI)	2.93 [0.12, 71.34]
5.13 Non‐serious adverse events Show forest plot	3	970	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.95, 1.10]

5.13.1 Adults	2	952	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.95, 1.10]
5.13.2 Children	1	18	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
5.14 Withdrawals due to adverse events Show forest plot	2	955	Risk Ratio (M‐H, Random, 95% CI)	2.47 [0.72, 8.43]

5.15 Nocturnal hypoglycaemia Show forest plot	3	970	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.91, 1.07]

5.15.1 Adults	2	952	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.91, 1.08]
5.15.2 Chlidren	1	18	Risk Ratio (M‐H, Random, 95% CI)	0.50 [0.12, 2.08]
5.16 Mild nocturnal hypoglycaemia Show forest plot	2	952	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.90, 1.07]

5.17 Nocturnal hypoglycaemia (asymptomatic) Show forest plot	2	952	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.71, 1.00]

5.18 Nocturnal hypoglycaemia (symptomatic) Show forest plot	2	952	Risk Ratio (M‐H, Random, 95% CI)	1.22 [0.72, 2.07]

5.19 Severe nocturnal hypoglycaemia Show forest plot	3	970	Risk Ratio (M‐H, Random, 95% CI)	1.39 [0.59, 3.27]

5.20 Mild/moderate hypoglycaemia Show forest plot	3	970	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.99, 1.04]

5.20.1 Adults	2	952	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.99, 1.04]
5.20.2 Children	1	18	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.82, 1.22]
5.21 HbA1c Show forest plot	4	1388	Mean Difference (IV, Random, 95% CI)	0.10 [0.00, 0.21]

5.21.1 Adults	3	1370	Mean Difference (IV, Random, 95% CI)	0.11 [0.00, 0.21]
5.21.2 Children	1	18	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.55, 0.55]
5.22 HbA1c (published vs. unpublished data) Show forest plot	4	1388	Mean Difference (IV, Random, 95% CI)	0.10 [0.00, 0.21]

5.22.1 Published	3	847	Mean Difference (IV, Random, 95% CI)	0.14 [0.02, 0.25]
5.22.2 Unpublished	1	541	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.21, 0.21]
5.23 Individuals with HbA1c < 7% without severe hypoglycaemia Show forest plot	2	911	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.78, 1.10]

Comparison 5. Insulin degludec versus insulin glargine

Risk of bias for analysis 1.1 All‐cause mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.1.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Kobayashi 2007

	Randomised, multi‐centre, open‐labelled, parallel‐group trial. No details were provided in the clinical study synopsis. Because this study was performed in cooperation with Novo Nordisk it is likely that randomisation and concealment of allocation were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS). The FAS consisted of all randomised participants who had any available efficacy data after receiving the trial product. The last observation carried forward (LOCF) approach was used for all endpoints at week 48 for participants who had at least one valid post‐baseline measurement.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and some translated pages from the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
NCT00595374

	Randomisation method was not specified. Because this trial was conducted in cooperation with Novo Nordisk we assumed that randomisation and allocation concealment were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All analyses were conducted for the full analysis set (FAS).

	Almost all randomised participants completed the study. All analyses were conducted for the full analysis set (FAS).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and ClinicalTrials.gov. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 1.1.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.1 All‐cause mortality

Risk of bias for analysis 1.3 Severe hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.3.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Kobayashi 2007

	Randomised, multi‐centre, open‐labelled, parallel‐group trial. No details were provided in the clinical study synopsis. Because this study was performed in cooperation with Novo Nordisk it is likely that randomisation and concealment of allocation were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS). The FAS consisted of all randomised participants who had any available efficacy data after receiving the trial product. The last observation carried forward (LOCF) approach was used for all endpoints at week 48 for participants who had at least one valid post‐baseline measurement.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and some translated pages from the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 1.3.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.3 Severe hypoglycaemia

Risk of bias for analysis 1.6 Cardiovascular mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.6.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Kobayashi 2007

	Randomised, multi‐centre, open‐labelled, parallel‐group trial. No details were provided in the clinical study synopsis. Because this study was performed in cooperation with Novo Nordisk it is likely that randomisation and concealment of allocation were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS). The FAS consisted of all randomised participants who had any available efficacy data after receiving the trial product. The last observation carried forward (LOCF) approach was used for all endpoints at week 48 for participants who had at least one valid post‐baseline measurement.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and some translated pages from the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
NCT00595374

	Randomisation method was not specified. Because this trial was conducted in cooperation with Novo Nordisk we assumed that randomisation and allocation concealment were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All analyses were conducted for the full analysis set (FAS).

	Almost all randomised participants completed the study. All analyses were conducted for the full analysis set (FAS).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and ClinicalTrials.gov. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. Protocol deviations were explained in the clinical study report. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 1.6.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. . A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.6 Cardiovascular mortality

Risk of bias for analysis 1.7 Non‐fatal myocardial infarction

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.7 Non‐fatal myocardial infarction

Risk of bias for analysis 1.8 Serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.8.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Kobayashi 2007

	Randomised, multi‐centre, open‐labelled, parallel‐group trial. No details were provided in the clinical study synopsis. Because this study was performed in cooperation with Novo Nordisk it is likely that randomisation and concealment of allocation were performed adequately. There were no relevant baseline imbalances.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS). The FAS consisted of all randomised participants who had any available efficacy data after receiving the trial product. The last observation carried forward (LOCF) approach was used for all endpoints at week 48 for participants who had at least one valid post‐baseline measurement.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and some translated pages from the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
NCT00595374

	Randomisation method was not specified. Because this trial was conducted in cooperation with Novo Nordisk we assumed that randomisation and allocation concealment were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All analyses were conducted for the full analysis set (FAS).

	Almost all randomised participants completed the study. All analyses were conducted for the full analysis set (FAS).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and ClinicalTrials.gov. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 1.8.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design with outcome measure unlikely influenced by lack of blinding

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.8 Serious adverse events

Risk of bias for analysis 1.10 Diabetic ketoacidosis

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.10.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Kobayashi 2007

	Randomised, multi‐centre, open‐labelled, parallel‐group trial. No details were provided in the clinical study synopsis. Because this study was performed in cooperation with Novo Nordisk it is likely that randomisation and concealment of allocation were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS). The FAS consisted of all randomised participants who had any available efficacy data after receiving the trial product. The last observation carried forward (LOCF) approach was used for all endpoints at week 48 for participants who had at least one valid post‐baseline measurement.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and some translated pages from the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 1.10.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. Protocol deviations were explained in the clinical study report. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. Protocol deviations were explained in the clinical study report. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.10 Diabetic ketoacidosis

Risk of bias for analysis 1.12 Non‐serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.12.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Kobayashi 2007

	Randomised, multi‐centre, open‐labelled, parallel‐group trial. No details were provided in the clinical study synopsis. Because this study was performed in cooperation with Novo Nordisk it is likely that randomisation and concealment of allocation were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS). The FAS consisted of all randomised participants who had any available efficacy data after receiving the trial product. The last observation carried forward (LOCF) approach was used for all endpoints at week 48 for participants who had at least one valid post‐baseline measurement.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from clinical study synopsis and some translated pages from the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
NCT00595374

	Randomisation method was not specified. Because this trial was conducted in cooperation with Novo Nordisk we assumed that randomisation and allocation concealment were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All analyses were conducted for the full analysis set (FAS).

	Almost all randomised participants completed the study. All analyses were conducted for the full analysis set (FAS).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from clinical study synopsis and ClinicalTrials.gov. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 1.12.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 1.12 Non‐serious adverse events

Risk of bias for analysis 1.18 Severe nocturnal hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.18.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 1.18.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.18 Severe nocturnal hypoglycaemia

Risk of bias for analysis 1.22 Severe nocturnal hypoglycaemia (published vs. unpublished data)

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.22.1 Published
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 1.22.2 Unpublished
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.22 Severe nocturnal hypoglycaemia (published vs. unpublished data)

Risk of bias for analysis 1.24 Mild/moderate hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.24.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Kobayashi 2007

	Randomised, multi‐centre, open‐labelled, parallel‐group trial. No details were provided in the clinical study synopsis. Because this study was performed in cooperation with Novo Nordisk it is likely that randomisation and concealment of allocation were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS). The FAS consisted of all randomised participants who had any available efficacy data after receiving the trial product. The last observation carried forward (LOCF) approach was used for all endpoints at week 48 for participants who had at least one valid post‐baseline measurement.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from clinical study synopsis and some translated pages from the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 1.24.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 1.24 Mild/moderate hypoglycaemia

Risk of bias for analysis 1.26 HbA1c

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.26.1 Adults
Bartley 2008

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) was used as the primary analysis for both efficacy and safety. Missing observations were considered missing at random in all analyses. Last observation carried forward (LOCF) was used for missing values on HbA1c, if participants had been treated for a minimum of 12 weeks and had a measurement after this period of time.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Kobayashi 2007

	Randomised, multi‐centre, open‐labelled, parallel‐group trial. No details were provided in the clinical study synopsis. Because this study was performed in cooperation with Novo Nordisk it is likely that randomisation and concealment of allocation were performed adequately. There were no relevant baseline imbalances.

	Open‐label trial design. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All participants who received at least one dose of trial product were included in the safety analysis. For all efficacy endpoints the analysis was performed on the full analysis set (FAS). The FAS consisted of all randomised participants who had any available efficacy data after receiving the trial product. The last observation carried forward (LOCF) approach was used for all endpoints at week 48 for participants who had at least one valid post‐baseline measurement.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from clinical study synopsis and some translated pages from the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Russell‐Jones 2004

	Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS). Block randomisation within individual trial sites was handled by the system. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis (ITT) set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. All primary and secondary efficacy analyses were based on the ITT analysis set.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Standl 2004

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Vague 2003

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 1.26.2 Children
NCT00605137

	Randomisation codes were prepared by a responsible person for randomisation and carried out by the registration centre. Randomisation was stratified by type of bolus insulin. Minor baseline differences did not indicate a problem with randomisation.

	Open‐label trial design. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. For all efficacy endpoints the analyses were performed on a full analysis set (FAS), i.e. all randomised participants who had any available efficacy data after receiving the trial product.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from study protocol, clinical study synopsis and some translated pages of the clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Robertson 2007

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. In the primary analysis, the missing HbA1c values at the end of trial were substituted by last observation carried forward (LOCF) if the measurement after 18 weeks was available.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Thalange 2013

	Randomisation was carried out using a telephone/web based randomisation system called Interactive Voice Response System (IVRS)/Interactive Web Response System (IWRS). Stratification was applied by age group (2‐5 and 6‐16 years old). There were no relevant baseline imbalances.

	Open‐label trial design. P study report. A modified intention‐to‐treat analysis was applied (all participants exposed to at least one dose of trial product with a post‐baseline observation).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. Last‐observation carried forward (LOCF) was used for HbA1c.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 1.26 HbA1c

Risk of bias for analysis 2.1 All‐cause mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.1.1 Adults
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Porcellati 2004

	A simple randomisation was used based on computer‐generated random numbers by a person who was not involved in establishing eligibility and entry of patients. Concealment of the randomisation was insured by having the allocation codes in a locked unreadable computer file handled by a person not involved in the recruitment of patients. The randomisation schedule was restricted and stratiﬁed by centre. There were no relevant baseline imbalances.

	Open‐label trial design. All participants randomised finished the study.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce information on analysis plan. Only one measurement and result provided for the time point selected by review authors.

	Open‐label trial design. Scarce information on analysis plan.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 2.1.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Liu 2016

	For efficacy analyses, patients were analysed in the treatment group allocated by the Interactive Voice Response System/Interactive Web Response System (IVRS/IWRS) at randomisation (as randomised). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy analyses were based on the modified intention‐to‐treat (mITT) population, corresponding to all randomised patients who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable. The safety population was defined as all randomised patients who took at least one dose or partial dose. In the event of patients having received treatments that differed from those assigned according to the randomisation schedule, then the safety analyses were to be conducted according to the treatment received rather than according to the randomisation groups.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy population consisted of all randomised patients who received at least one dose of the study medication (modified intent‐to‐treat [mITT] population).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.1 All‐cause mortality

Risk of bias for analysis 2.2 Health‐realted quality of life

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 2.2 Health‐realted quality of life

Risk of bias for analysis 2.3 Severe hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.3.1 Adults
Bolli 2009

	This was a randomised, parallel group, open‐label, multicentre, single country study. The randomisation schedule was restricted and stratiﬁed by centre. No more details available. There were no relevant baseline imbalances.

	Open‐label trial design. All efficacy analyses were determined using a modified intention‐to‐treat (ITT) population (taking at least one dose of study drug and providing data at baseline and at least one on‐treatment visit).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All efficacy analyses were determined using a modified intention‐to‐treat (ITT) population (taking at least one dose of study drug and providing data at baseline and at least one on‐treatment visit).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Only one measurement and result provided for the time point selected by review authors.

	Open‐label trial. No details of the randomisation process.
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Porcellati 2004

	A simple randomisation was used based on computer‐generated random numbers by a person who was not involved in establishing eligibility and entry of patients. Concealment of the randomisation was insured by having the allocation codes in a locked unreadable computer file handled by a person not involved in the recruitment of patients. The randomisation schedule was restricted and stratiﬁed by centre. There were no relevant baseline imbalances.

	Open‐label trial design. All participants randomised finished the study.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce information on analysis plan. Only one measurement and result provided for the time point selected by review authors.

	Open‐label trial design. Scarce information on analysis plan.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 2.3.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Liu 2016

	For efficacy analyses, patients were analysed in the treatment group allocated by the Interactive Voice Response System/Interactive Web Response System (IVRS/IWRS) at randomisation (as randomised). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy analyses were based on the modified intention‐to‐treat (mITT) population, corresponding to all randomised patients who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable. The safety population was defined as all randomised patients who took at least one dose or partial dose. In the event of patients having received treatments that differed from those assigned according to the randomisation schedule, then the safety analyses were to be conducted according to the treatment received rather than according to the randomisation groups.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy population consisted of all randomised patients who received at least one dose of the study medication (modified intent‐to‐treat [mITT] population).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. Protocol deviations were explained in the clinical study report. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.3 Severe hypoglycaemia

Risk of bias for analysis 2.6 Cardiovascular mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.6.1 Adults
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Porcellati 2004

	A simple randomisation was used based on computer‐generated random numbers by a person who was not involved in establishing eligibility and entry of patients. Concealment of the randomisation was insured by having the allocation codes in a locked unreadable computer file handled by a person not involved in the recruitment of patients. The randomisation schedule was restricted and stratiﬁed by centre. There were no relevant baseline imbalances.

	Open‐label trial design. All participants randomised finished the study.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce information on analysis plan. Only one measurement and result provided for the time point selected by review authors.

	Open‐label trial design. Scarce information on analysis plan.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 2.6.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Liu 2016

	For efficacy analyses, patients were analysed in the treatment group allocated by the Interactive Voice Response System/Interactive Web Response System (IVRS/IWRS) at randomisation (as randomised). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy analyses were based on the modified intention‐to‐treat (mITT) population, corresponding to all randomised patients who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable. The safety population was defined as all randomised patients who took at least one dose or partial dose. In the event of patients having received treatments that differed from those assigned according to the randomisation schedule, then the safety analyses were to be conducted according to the treatment received rather than according to the randomisation groups.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy population consisted of all randomised patients who received at least one dose of the study medication (modified intent‐to‐treat [mITT] population).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.6 Cardiovascular mortality

Risk of bias for analysis 2.7 Non‐fatal myocardial infarction

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.7 Non‐fatal myocardial infarction

Risk of bias for analysis 2.8 Non‐fatal stroke

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.8 Non‐fatal stroke

Risk of bias for analysis 2.9 Serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.9.1 Adults
Bolli 2009

	This was a randomised, parallel group, open‐label, multicentre, single country study. The randomisation schedule was restricted and stratiﬁed by centre. No more details available. There were no relevant baseline imbalances.

	Open‐label trial design. All efficacy analyses were determined using a modified intention‐to‐treat (ITT) population (taking at least one dose of study drug and providing data at baseline and at least one on‐treatment visit).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All efficacy analyses were determined using a modified intention‐to‐treat (ITT) population (taking at least one dose of study drug and providing data at baseline and at least one on‐treatment visit).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Only one measurement and result provided for the time point selected by review authors.

	Open‐label trial. No details of the randomisation process.
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Only participants with at least one measurement after baseline were included in the intent‐to‐treat (ITT) analysis.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 2.9.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Liu 2016

	For efficacy analyses, patients were analysed in the treatment group allocated by the Interactive Voice Response System/Interactive Web Response System (IVRS/IWRS) at randomisation (as randomised). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy analyses were based on the modified intention‐to‐treat (mITT) population, corresponding to all randomised patients who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable. The safety population was defined as all randomised patients who took at least one dose or partial dose. In the event of patients having received treatments that differed from those assigned according to the randomisation schedule, then the safety analyses were to be conducted according to the treatment received rather than according to the randomisation groups.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.9 Serious adverse events

Risk of bias for analysis 2.11 Diabetic ketoacidosis

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.11.1 Adults
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 2.11.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Liu 2016

	For efficacy analyses, patients were analysed in the treatment group allocated by the Interactive Voice Response System/Interactive Web Response System (IVRS/IWRS) at randomisation (as randomised). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy analyses were based on the modified intention‐to‐treat (mITT) population, corresponding to all randomised patients who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable. The safety population was defined as all randomised patients who took at least one dose or partial dose. In the event of patients having received treatments that differed from those assigned according to the randomisation schedule, then the safety analyses were to be conducted according to the treatment received rather than according to the randomisation groups.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy population consisted of all randomised patients who received at least one dose of the study medication (modified intent‐to‐treat [mITT] population).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.11 Diabetic ketoacidosis

Risk of bias for analysis 2.13 Non‐serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.13.1 Adults
Bolli 2009

	This was a randomised, parallel group, open‐label, multicentre, single country study. The randomisation schedule was restricted and stratiﬁed by centre. No more details available. There were no relevant baseline imbalances.

	Open‐label trial design. All efficacy analyses were determined using a modified intention‐to‐treat (ITT) population (taking at least one dose of study drug and providing data at baseline and at least one on‐treatment visit).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All efficacy analyses were determined using a modified intention‐to‐treat (ITT) population (taking at least one dose of study drug and providing data at baseline and at least one on‐treatment visit).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions. No details of the randomisation process.
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 2.13.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Liu 2016

	For efficacy analyses, patients were analysed in the treatment group allocated by the Interactive Voice Response System/Interactive Web Response System (IVRS/IWRS) at randomisation (as randomised). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy analyses were based on the modified intention‐to‐treat (mITT) population, corresponding to all randomised patients who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable. The safety population was defined as all randomised patients who took at least one dose or partial dose. In the event of patients having received treatments that differed from those assigned according to the randomisation schedule, then the safety analyses were to be conducted according to the treatment received rather than according to the randomisation groups.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy population consisted of all randomised patients who received at least one dose of the study medication (modified intent‐to‐treat [mITT] population).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 2.13 Non‐serious adverse events

Risk of bias for analysis 2.19 Severe nocturnal hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.19.1 Adults
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 2.19.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy population consisted of all randomised patients who received at least one dose of the study medication (modified intent‐to‐treat [mITT] population).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.19 Severe nocturnal hypoglycaemia

Risk of bias for analysis 2.22 Mild/moderate hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.22.1 Adults
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 2.22.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Liu 2016

	For efficacy analyses, patients were analysed in the treatment group allocated by the Interactive Voice Response System/Interactive Web Response System (IVRS/IWRS) at randomisation (as randomised). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy analyses were based on the modified intention‐to‐treat (mITT) population, corresponding to all randomised patients who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable. The safety population was defined as all randomised patients who took at least one dose or partial dose. In the event of patients having received treatments that differed from those assigned according to the randomisation schedule, then the safety analyses were to be conducted according to the treatment received rather than according to the randomisation groups.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy population consisted of all randomised patients who received at least one dose of the study medication (modified intent‐to‐treat [mITT] population).

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 2.22 Mild/moderate hypoglycaemia

Risk of bias for analysis 2.24 HbA1c

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 2.24.1 Adults
Bolli 2009

	This was a randomised, parallel group, open‐label, multicentre, single country study. The randomisation schedule was restricted and stratiﬁed by centre. No more details available. There were no relevant baseline imbalances.

	Open‐label trial design. All efficacy analyses were determined using a modified intention‐to‐treat (ITT) population (taking at least one dose of study drug and providing data at baseline and at least one on‐treatment visit).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. All efficacy analyses were determined using a modified intention‐to‐treat (ITT) population (taking at least one dose of study drug and providing data at baseline and at least one on‐treatment visit).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Only one measurement and result provided for the time point selected by review authors.

	Open‐label trial. No details of the randomisation process.
Fulcher 2005

	The randomisation schedule for treatment assignments was prepared by centre on a 1:1 basis. A randomisation list was provided to each centre (pharmacy department or appropriate study personal but not to investigator blinded to the study). There were no relevant baseline imbalances.

	Single‐blind trial design: the investigator responsible for insulin dosage adjustment was blinded to the treatment allocated to the participant and was not involved in any aspect relating to study supplies. The study coordinator and the participant were aware of the treatment allocation. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The primary efficacy analysis was carried out with a modified intention‐to‐treat population (ITT), which consisted of all randomised patients with at least one dose of study medication. All secondary efficacy variables were also analysed on an ITT basis. The safety population consisted of all randomised participants who received at least one dose of study medication.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Home 2005

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (participants with at least one measurement after baseline).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (efficacy was analysed using those intention‐to‐treat (ITT) participants for whom both a pre‐treatment and an on‐treatment value were available, with the exception of hypoglycaemia, for which the entire ITT population was analysed).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Porcellati 2004

	A simple randomisation was used based on computer‐generated random numbers by a person who was not involved in establishing eligibility and entry of patients. Concealment of the randomisation was insured by having the allocation codes in a locked unreadable computer file handled by a person not involved in the recruitment of patients. The randomisation schedule was restricted and stratiﬁed by centre. There were no relevant baseline imbalances.

	Open‐label trial design. All participants randomised finished the study.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce information on analysis plan. Only one measurement and result provided for the time point selected by review authors.

	Open‐label trial design. Scarce information on analysis plan.
Ratner 2000

	The randomisation schedule was stratified according to centre and pre‐randomisation basal insulin regimen of once versus twice daily. Forty randomisation numbers were allocated to each basal insulin regimen for each centre. To ensure a balanced number of participants for each treatment group the randomisation schedule was generated using block sizes of 4. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pretreatment and a during‐treatment value.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated and having both a pre‐treatment and a during‐treatment value. An ITT analysis was performed for all variables. Participants with missing baseline values or no value during treatment were excluded from the statistical analysis of the variable in question.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 2.24.2 Children
Chase 2008

	Two randomisation schedules with a 1:1 randomisation ratio, one for males another for females, were generated by the sponsor. At the point of randomisation both the investigator/study coordinator and the participant were blinded to the treatment allocation (to randomisation). The randomisation schedule was stored with the randomisation code administrator at Aventis Pharmaceuticals, Bridgewater, NJ. There were no relevant baseline imbalances.

	Open‐label trial design. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (participants taking at least one dose of study medication with a baseline measurement and at least one follow‐up measure; the safety population consisted of all randomised participants who received at least one dose of study medication).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Liu 2016

	For efficacy analyses, patients were analysed in the treatment group allocated by the Interactive Voice Response System/Interactive Web Response System (IVRS/IWRS) at randomisation (as randomised). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy analyses were based on the modified intention‐to‐treat (mITT) population, corresponding to all randomised patients who received at least one dose, and had both a baseline assessment and at least one post‐baseline assessment for at least one efficacy variable. The safety population was defined as all randomised patients who took at least one dose or partial dose. In the event of patients having received treatments that differed from those assigned according to the randomisation schedule, then the safety analyses were to be conducted according to the treatment received rather than according to the randomisation groups.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
PRESCHOOL

	Participants were centrally randomised (utilising permuted block randomisation schedule) via Interactive Voice Response System (IVRS) at a 1:1 ratio to one of the two treatment groups. There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The efficacy population consisted of all randomised patients who received at least one dose of the study medication (modified intent‐to‐treat [mITT] population).

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Schober 2002

	A 1:1 randomisation schedule was generated by the sponsor. This schedule paired sequential numbers with treatment codes allocated at random. The schedule was prepared by centre on a 1:1 basis. An independent agency was used for central telephone randomisation. There were no relevant baseline imbalances.

	Open‐label trial design. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The modified intention‐to‐treat (ITT) population was defined as all participants randomised and treated. The safety population compromised all participants who were randomised and received study medication. Efficacy was analysed using those ITT participants for whom both a pre‐treatment and an on‐treatment value were available.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 2.24 HbA1c

Risk of bias for analysis 3.1 All‐cause mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Pieber 2007

	Participants were symmetrically randomised (1:1) to either insulin detemir or insulin glargine. Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS).

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to trial products).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.1 All‐cause mortality

Risk of bias for analysis 3.2 Severe hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Pieber 2007

	Participants were symmetrically randomised (1:1) to either insulin detemir or insulin glargine. Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS).

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to trial products).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.2 Severe hypoglycaemia

Risk of bias for analysis 3.5 Cardiovascular mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Pieber 2007

	Participants were symmetrically randomised (1:1) to either insulin detemir or insulin glargine. Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS).

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to trial products).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.5 Cardiovascular mortality

Risk of bias for analysis 3.6 Non‐fatal myocardial infarction

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.6 Non‐fatal myocardial infarction

Risk of bias for analysis 3.7 Non‐fatal stroke

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.7 Non‐fatal stroke

Risk of bias for analysis 3.8 Serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Pieber 2007

	Participants were symmetrically randomised (1:1) to either insulin detemir or insulin glargine. Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS).

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to trial products).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.8 Serious adverse events

Risk of bias for analysis 3.9 Diabetic ketoacidosis

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.9 Diabetic ketoacidosis

Risk of bias for analysis 3.10 Non‐serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Pieber 2007

	Participants were symmetrically randomised (1:1) to either insulin detemir or insulin glargine. Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS).

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to trial products).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 3.10 Non‐serious adverse events

Risk of bias for analysis 3.16 Severe nocturnal hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 3.16.1 Published
Pieber 2007

	Participants were symmetrically randomised (1:1) to either insulin detemir or insulin glargine. Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS).

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to trial products).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 3.16.2 Unpublished
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.16 Severe nocturnal hypoglycaemia

Risk of bias for analysis 3.17 Mild/moderate hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 3.17.1 Published
Pieber 2007

	Participants were symmetrically randomised (1:1) to either insulin detemir or insulin glargine. Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS).

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to trial products).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 3.17.2 Unpublished
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 3.17 Mild/moderate hypoglycaemia

Risk of bias for analysis 3.18 HbA1c

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Heller 2009

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to at least one dose of trial product, bolus or basal insulin).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Participants needed to have at least one post‐treatment measurement obtained after 3 month of treatment and a valid baseline measurement in order to qualify for the analysis of HbA1c. Last‐observation‐carried‐forward (LOCF) method was generally used for missing values on the endpoints that were measured after initiation of treatment and on more than one occasion.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Pieber 2007

	Participants were symmetrically randomised (1:1) to either insulin detemir or insulin glargine. Randomisation was carried out using a telephone randomisation system, the Interactive Voice Response System (IVRS).

	Open‐label trial design. A modified intention‐to‐treat analysis was applied (all randomised participants exposed to trial products).

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Modified intention‐to‐treat analysis set was used (all exposed participants were analysed). Missing observations were considered missing at random in all analyses. Missing HbA1c measurement at the end of the trial (after 26 weeks) was extrapolated by last‐observation‐carried‐forward (LOCF) if a measurement after 20 weeks was available. Otherwise, the measurement was considered as missing.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 3.18 HbA1c

Risk of bias for analysis 4.1 All‐cause mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.1.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 4.1.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.1 All‐cause mortality

Risk of bias for analysis 4.2 Health‐related quality of life

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.2.1 Physical health score
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 4.2.2 Mental health score
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 4.2 Health‐related quality of life

Risk of bias for analysis 4.3 Severe hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.3.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 4.3.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method. External classification of severe hypoglycaemia was performed blinded.

	External classification of severe hypoglycaemia was performed blinded and outcome measure unlikely influenced by potential lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.3 Severe hypoglycaemia

Risk of bias for analysis 4.5 Cardiovascular mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.5.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 4.5.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.5 Cardiovascular mortality

Risk of bias for analysis 4.6 Non‐fatal myocardial infarction

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.6 Non‐fatal myocardial infarction

Risk of bias for analysis 4.7 Non‐fatal stroke

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.7 Non‐fatal stroke

Risk of bias for analysis 4.8 End stage renal disease

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.8 End stage renal disease

Risk of bias for analysis 4.9 Blindness

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.9 Blindness

Risk of bias for analysis 4.10 Serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.10.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 4.10.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.10 Serious adverse events

Risk of bias for analysis 4.11 Diabetic ketoacidosis

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.11.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 4.11.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.11 Diabetic ketoacidosis

Risk of bias for analysis 4.12 Non‐serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.12.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 4.12.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 4.12 Non‐serious adverse events

Risk of bias for analysis 4.18 Severe nocturnal hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.18.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 4.18.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. External classification of severe hypoglycaemia was performed blinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.18 Severe nocturnal hypoglycaemia

Risk of bias for analysis 4.19 Mild/moderate hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.19.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 4.19.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

Risk of bias for analysis 4.19 Mild/moderate hypoglycaemia

Risk of bias for analysis 4.20 HbA1c

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 4.20.1 Adults
Davies 2014

	At the randomisation visit (visit 2) the participants were to be randomised to either insulin degludec or insulin detemir, both in combination with insulin aspart. The randomisation was to be carried out in a 2:1 manner using the Interactive Voice/Web Response System IV/WRS. The trial was stratified according to region with 4 levels: Europe (Italy, UK, Macedonia and Finland), Japan, India and South America (Brazil). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. Full Analysis Set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. Safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Participants in the safety set were to contribute to the evaluation “as treated”. Unless otherwise specified missing values (including intermittent missing values) were imputed using the last observation carried forward (LOCF) method. Missing observations were considered missing at random in all analyses.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 4.20.2 Children
BEGIN Young

	Participants were allocated to treatment with insulin degludec or insulin detemir in a 1:1 ratio. Randomisation was carried out using a central interactive voice/web response system (IV/WRS). Randomisation was stratified according to 3 age groups: 1 to less than 6 years; 6 to less than 12 years; 12 to less than 18 years of age. There were no relevant baseline imbalances.

	Open‐label trial design. The internal Novo Nordisk safety committee and the external data monitoring committee (DMC) reviewed safety data on an ongoing basis. The internal safety committee was blinded and the DMC was unblinded. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The full analysis set (FAS) included all randomised participants. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants contributed to the evaluation “as randomised”. The safety analysis set included all participants receiving at least one dose of the investigational product. Participants in the safety set were to contribute to the evaluation “as treated”. All analyses and summary of efficacy endpoints, and formal statistical analyses related to safety endpoints were based on the full analysis set (FAS). Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Unless otherwise specified, missing values (including intermittent missing values) were imputed using the last‐observation‐carried‐forward (LOCF) method.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 4.20 HbA1c

Risk of bias for analysis 5.1 All‐cause mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants finished the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure unlikely influenced by lack of blinding. Scarce information.

Risk of bias for analysis 5.1 All‐cause mortality

Risk of bias for analysis 5.3 Health‐related quality of life (physical health)

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.3.1 Published
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 5.3.2 Unpublished
SWITCH 1

	Participants were randomised in a 1:1 manner to one of the two treatment sequences, using the Interactive Voice/Web Response System IV/WRS. Within each treatment sequence participants were randomised 1:1 to morning or evening dosing. There were no relevant baseline imbalances.

	Double‐blind crossover study (participants, the clinical study group and the investigator remained blinded throughout the trial). Analyses of all endpoints were based on the full analysis set (FAS). Efficacy endpoints and patient‐reported outcome endpoints were summarised using the FAS. Safety endpoints were summarised using the safety analysis set.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Analyses of all endpoints were based on the FAS. Efficacy endpoints and patient‐reported outcome endpoints were summarised using the FAS. Safety endpoints were summarised using the SAS.

	Double‐blind design (participants, the clinical study group and the investigator remained blinded throughout the trial).

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 5.3 Health‐related quality of life (physical health)

Risk of bias for analysis 5.4 Health‐related quality of life (mental health)

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.4.1 Published
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 5.4.2 Unpublished
SWITCH 1

	Participants were randomised in a 1:1 manner to one of the two treatment sequences, using the Interactive Voice/Web Response System IV/WRS. Within each treatment sequence participants were randomised 1:1 to morning or evening dosing. There were no relevant baseline imbalances.

	Double‐blind crossover study (participants, the clinical study group and the investigator remained blinded throughout the trial). Analyses of all endpoints were based on the full analysis set (FAS). Efficacy endpoints and patient‐reported outcome endpoints were summarised using the FAS. Safety endpoints were summarised using the safety analysis set.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Analyses of all endpoints were based on the FAS. Efficacy endpoints and patient‐reported outcome endpoints were summarised using the FAS. Safety endpoints were summarised using the SAS.

	Double‐blind design (participants, the clinical study group and the investigator remained blinded throughout the trial).

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.

Risk of bias for analysis 5.4 Health‐related quality of life (mental health)

Risk of bias for analysis 5.5 Severe hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.5.1 Adults
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 5.5.2 Children
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure unlikely influenced by lack of blinding. Scarce information.

Risk of bias for analysis 5.5 Severe hypoglycaemia

Risk of bias for analysis 5.7 Cardiovascular mortality

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.7.1 Adults
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 5.7.2 Children
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants finished the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure unlikely influenced by lack of blinding. Scarce information.

Risk of bias for analysis 5.7 Cardiovascular mortality

Risk of bias for analysis 5.8 Non‐fatal myocardial infarction

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.8.1 Adults
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 5.8.2 Children
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure unlikely influenced by lack of blinding. Scarce information.

Risk of bias for analysis 5.8 Non‐fatal myocardial infarction

Risk of bias for analysis 5.9 Non‐fatal stroke

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure unlikely influenced by lack of blinding. Scarce information.

Risk of bias for analysis 5.9 Non‐fatal stroke

Risk of bias for analysis 5.10 Serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.10.1 Adults
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 5.10.2 Children
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure unlikely influenced by lack of blinding. Scarce information.

Risk of bias for analysis 5.10 Serious adverse events

Risk of bias for analysis 5.11 Diabetic ketoacidosis

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.11.1 Adults
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 5.11.2 Children
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure unlikely influenced by lack of blinding. Scarce information.

Risk of bias for analysis 5.11 Diabetic ketoacidosis

Risk of bias for analysis 5.13 Non‐serious adverse events

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.13.1 Adults
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 5.13.2 Children
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions. Scarce information.

Risk of bias for analysis 5.13 Non‐serious adverse events

Risk of bias for analysis 5.19 Severe nocturnal hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure unlikely influenced by lack of blinding. Scarce information.

Risk of bias for analysis 5.19 Severe nocturnal hypoglycaemia

Risk of bias for analysis 5.20 Mild/moderate hypoglycaemia

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.20.1 Adults
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.
Subgroup 5.20.2 Children
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label design. Outcome measure could have been influenced by knowledge of the intervention received. It it unlikely that there were strong beliefs in beneficial or harmful effects of the interventions. Scarce information.

Risk of bias for analysis 5.20 Mild/moderate hypoglycaemia

Risk of bias for analysis 5.21 HbA1c

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 5.21.1 Adults
BEGIN Basal‐Bolus Type 1

	Participants were allocated to treatment with insulin degludec or insulin glargine in a 3:1 ratio. The Interactive Voice/Web Response System IV/WRS allocated the trial product to the participant at each dispensing and randomisation visit. There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the FAS was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 52 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
BEGIN Flex T1

	Randomisation was carried out using a telephone randomisation system (Interactive Voice Response System (IVRS)). There were no relevant baseline imbalances.

	Open‐label trial design. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Only endpoints derived after 26 weeks of treatment were to be analysed statistically. Missing values were imputed by last‐observation‐carried‐forward (LOCF) for all endpoints.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
SWITCH 1

	Participants were randomised in a 1:1 manner to one of the two treatment sequences, using the Interactive Voice/Web Response System IV/WRS. Within each treatment sequence participants were randomised 1:1 to morning or evening dosing. There were no relevant baseline imbalances.

	Double‐blind crossover study (participants, the clinical study group and the investigator remained blinded throughout the trial). Analyses of all endpoints were based on the full analysis set (FAS). Efficacy endpoints and patient‐reported outcome endpoints were summarised using the FAS. Safety endpoints were summarised using the safety analysis set.

	Withdrawals and reasons for withdrawal were documented, did not differ substantially between intervention groups and did not appear to be related to health status. The statistical evaluation of the full analysis set (FAS) was to follow the intention‐to‐treat (ITT) principle and participants were to contribute to the evaluation “as randomised”. The safety analysis set included all participants who received at least one dose of the investigational product or its comparator. Analyses of all endpoints were based on the FAS. Efficacy endpoints and patient‐reported outcome endpoints were summarised using the FAS. Safety endpoints were summarised using the SAS.

	Double‐blind design (participants, the clinical study group and the investigator remained blinded throughout the trial).

	Data from full clinical study report. Only one measurement and result provided for the time point selected by review authors.

	No risk of bias identified.
Subgroup 5.21.2 Children
Urakami 2017

	No details on the randomisation process. There were no relevant baseline imbalances.

	Open‐label trial design. Scarce information.

	All randomised participants completed the study.

	Open‐label design with outcome measure unlikely influenced by lack of blinding.

	Scarce data from publication and study author. Probably only one measurement and result provided for the time point selected by review authors.

	No details on the randomisation process. Open‐label trial design with outcome measure. and no protocol available.

Risk of bias for analysis 5.21 HbA1c