Treatment of newly diagnosed glioblastoma in the elderly: a network meta‐analysis

Summary of findings for the main comparison. Summary of findings on overall survival comparing treatments to supportive care only

Estimates of effects, certainty assessment and rankings of different treatment options compared with supportive care only on overall survival in elderly people with glioblastoma
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: radiotherapy with 40 Gy in 15 fractions (RT40); radiotherapy with 60 Gy in 30 fractions (RT60); chemoradiotherapy (CRT); chemoradiotherapy plus bevacizumab (BEV_CRT); radiotherapy plus bevacizumab (BEV_RT); temozolomide (TMZ) Comparison: supportive care only Outcome: overall survival
All intervention options (7 RCTs; 1540 participants in total)*	Relative effect (network estimate) ** (95% CI)	Illustrative absolute effects for death at 6 months (95% CIs)	Certainty of the evidence (GRADE)	Ranking¥
Supportive care only*** (1 RCT, 81 participants)	Reference comparator	76 per 100***	Reference comparator	7.0 (worst)
RT60 (5 RCTs; 713 participants)	HR 0.47 (0.29 to 0.76)	49 per 100 (34 to 66)	⊕⊕⊕⊝ Moderate¹	5.0
BEV_RT (1 RCT; 75 participants)	HR 0.48 (0.23 to 1.00)	50 per 100 (28 to 76)	Not graded²	4.7
RT40 (4 RCTs; 930 participants)	HR 0.44 (0.25 to 0.77)	47 per 100 (30 to 67)	⊕⊕⊝⊝ Low³	4.3
TMZ (3 RCTs, 538 participants)	HR 0.42 (0.25 to 0.71)	45 per 100 (30 to 64)	⊕⊕⊝⊝ Low³	3.8
CRT (2 RCTs; 635 participants)	HR 0.30 (0.17 to 0.53)	35 per 100 (22 to 53)	Not graded²	1.8
BEV_CRT (1 RCT; 73 participants)	HR 0.25 (0.11 to 0.54)	30 per 100 (15 to 54)	Not graded²	1.4
NMA‐'Summary of findings' table definitions Estimates are reported as HR: hazard ratio, CI: confidence interval,NMA: network meta‐analysis. * This refers to the number of studies in the network evaluating the given intervention and the number of participants involved in these studies. All NMA effect estimates in this 'Summary of findings' table are derived 100% from indirect evidence, except for the comparison of RT60 versus supportive care, which was directly compared in one study. Where there was no common comparator for the comparison we did not grade the certainty of the evidence. * The assumed effect of supportive care is based on Keime‐Guibert 2007 data, which may be slightly overestimated because a high proportion of patients underwent biopsy only. ¥These rankings do not take into account the certainty of the evidence and should be interpreted with caution. The estimates of ungraded evidence are very uncertain.
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.
¹ Downgraded ‐1 as the evidence was derived from a single small study. ² There was no common comparator for the comparison (i.e. the intervention was not connected in a loop in the evidence network), therefore we did not grade the certainty of the evidence. ³ Contributing direct evidence was of moderate or low certainty. Abbreviations BEV_CRT; chemoradiotherapy plus bevacizumab; CI: confidence interval; CRT; chemoradiotherapy; HR: hazard ratio: RT40; radiotherapy (40 Gy in 15 fractions); RT60; radiotherapy (60 Gy in 30 fractions); SC; supportive care; TMZ; temozolomide; TTF_AC; tumour treating fields plus adjuvant chemotherapy) after concomitant CRT)

Summary of findings 2. Summary of findings on overall survival comparing treatments to hypofractionated radiotherapy

Estimates of effects and certainty assessments compared with hypofractionated radiotherapy (40 Gy/15 fractions) on overall survival in elderly people with newly diagnosed glioblastoma
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: chemoradiotherapy (CRT); chemoradiotherapy plus bevacizumab (BEV_CRT); radiotherapy plus bevacizumab (BEV_RT); temozolomide (TMZ) Comparison: hypofractionated radiotherapy (RT 40) Outcome: overall survival
All intervention options (7 RCTs; 1540 participants in total)*	Relative effect (network estimate) ** (95% CI)	Illustrative absolute effects for death at 12 months (95% CIs)	Certainty of the evidence (GRADE)
RT 40 (4 RCTs; 930 participants)	Reference comparator	78 per 100***	Reference comparator
BEV_RT (1 RCT; 75 participants)	HR 1.08 (0.66 to 1.78)	81 per 100 (63 to 93)	⊕⊕⊝⊝ Low¹
TMZ (3 RCTs, 538 participants)	HR 0.95 (0.71 to 1.26)	76 per 100 (66 to 85)	⊕⊕⊝⊝ Low¹
CRT (2 RCTs; 635 participants)	HR 0.67 (0.56 to 0.80)	64 per 100 (57 to 70)	⊕⊕⊕⊕ High
BEV_CRT (1 RCT; 73 participants)	HR 0.56 (0.31 to 0.99)	57 per 100 (37 to 78)	⊕⊕⊕⊝ Moderate²
For the comparison with standard radiotherapy (60 Gy/30 fractions) see summary of findings Table 3.
NMA‐'Summary of findings' table definitions Estimates are reported as HR: hazard ratio. CI: confidence interval. *The assumed absolute effect of RT40 is based on Perry 2017 data. NMA: network meta‐analysis 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.
¹Contributing direct evidence was of low certainty (study design limitations and imprecision). ²Contributing direct evidence was of high or moderate certainty.

Summary of findings 3. Summary of findings on overall survival comparing treatments to standard radiotherapy

Estimates of effects and certainty assessment compared with standard radiotherapy (60 Gy/30 fractions) on overall survival in elderly people with newly diagnosed glioblastoma
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: radiotherapy with 40 Gy in 15 fractions (RT 40); chemoradiotherapy (CRT); chemoradiotherapy plus bevacizumab (BEV_CRT); radiotherapy plus bevacizumab (BEV_RT); temozolomide (TMZ); Comparison: standard radiotherapy (RT 60) Outcome: overall survival
All intervention options (7 RCTs; 1540 participants in total)*	Relative effect (network estimate) ** (95% CI)	Certainty of the evidence (GRADE)
RT 60 (5 RCTs; 713 participants)	Reference comparator	Reference comparator
RT 40 (4 RCTs; 930 participants)	HR 0.94 (0.72 to 1.23)	⊕⊕⊝⊝ Low¹
BEV_RT (1 RCT; 75 participants)	HR 1.01 (0.58 to 1.79)	⊕⊝⊝⊝ Very low^1,2
TMZ (3 RCTs, 538 participants)	HR 0.89 (0.71 to 1.11)	⊕⊝⊝⊝ Very low³
CRT (2 RCTs; 635 participants)	HR 0.63 (0.46 to 0.87)	⊕⊕⊝⊝ Low
BEV_CRT (1 RCT; 73 participants)	HR 0.52 (0.28 to 0.98)	Not graded
NMA‐'Summary of findings' table definitions Estimates are reported as HR: hazard ratio. CI: confidence interval. For assumed median survival times and absolute effect estimates, please refer to summary of findings Table for the main comparison. NMA: network meta‐analysis 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.
¹ Contributing direct evidence was of low certainty (study design limitations and imprecision). ² Downgraded for imprecision. ³ Contributing direct evidence of very low certainty (imprecision, study design limitations and inconsistency).

Summary of findings 4. Summary of findings on quality of life

The effect of different treatment comparisons for newly diagnosed glioblastoma in the elderly on quality of life
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: one treatment option Comparison: an alternative treatment option
Comparison	Narrative summary of evidence	No of studies (elderly participants)	Certainty of the evidence (GRADE)	Comments
RT vs supportive care only	Evidence on relative HRQoL is very uncertain.	1 study (59 participants at 30 day and 26 at 135 day time points, respectively)	⊕⊝⊝⊝ Low^1,2	The study reported that global assessments of deterioration over time also did not differ significantly between the two groups. The dropout rate was high and unbalanced so findings were difficult to interpret.
Hypofractionated RT vs standard RT (60Gy/30 fractions)	There may be little difference in HRQoL scores between hypofractionated and standard fractionation schedules in the 6 to 8 weeks following treatment.	2 studies (participants numbered 85 and 24, respectively, at the furthest time point)	⊕⊝⊝⊝ Low^1,2	One study compared a 25 Gy schedule with a 40 Gy schedule; the other compared a 40 Gy schedule with a standard 60 Gy schedule.
TMZ vs RT	No significant differences in global QoL scores at 3, 6, or 12 month measurements. However, there was a significant difference in discomfort from communication deficits, which were greatest for those receiving RT who died at between 6 and 12 months (P = 0·002).	1 study (306 participants)	⊕⊕⊕⊝ Moderate³	Evidence was not downgraded for attrition because data were reported for 82% of participants in each group for this outcome.
CRT vs RT	Authors reported that quality of life was similar in the two treatment groups. See Comments.	1 study (562 participants; number with QOL data was unclear)	Not graded	Investigators noted that attrition impacted the quantity of data. They conducted analyses using time to deterioration (with deterioration defined as a 10‐point decrease in the score on the function domain or a 10‐point increase in the score on the symptom domain) and plotted QoL scores over time. They reported that "There were no other clinically important differences between trial groups, which supports our observation that quality of life was similar in the two treatment groups."
BEV_CRT vs CRT	This was reported for the overall trial but not for elderly subgroup specifically.	1 study (73 participants)	Not graded	The authors reported significantly delayed deterioration in HRQoL scores in favour of BEV_CRT across five main HRQoL domains (global health, communication, social functioning, motor function, physical functioning). When progression of disease was removed as a deterioration event, the time to clinically significant deterioration or death remained statistically significant for communication, social functioning and global health.
BEV_RT vs RT	Investigators reported that "before progression, no differences were detected for individual scales in a generalized linear mixed model, except for less favourable values in arm A (BEV) for global health (P=0.048) and pain (P=0.027)".	1 study (75 participants)	Not graded	In the publication, global health was reported in a forest plot along with individual HRQoL items, such as cognitive functioning, emotional functioning and pain, measured with EORTC QLQ‐C30/BN20 scales. No other data were provided or obtained.
RIN_CRT vs control/CRT	This was reported for overall trial but not for elderly subgroup specifically.	1 study (174 participants)	Not graded	Investigators reported no significant differences between patients in their trial arms in any of the HRQoL measures.
TTF_AC vs CRT	This was reported for overall trial but not for elderly subgroup specifically.⁴	1 study (134 participants)	Not graded	There was no significant difference in HRQoL reported between the trial arms, except for itchy skin which was more prominent in the TTFields arm at 3,6 and 9 months (P = 0.005, P = 0008, P = 0.04). There was no significant difference at 12 months.
IRI_BEV_RT vs CRT	This was reported for overall trial but not for elderly subgroup specifically.	1 study (34 participants)	Not graded	There was no significant difference between the treatment arms.
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.
¹Sparse data from single studies [‐1] ² Serious risk of bias from attrition [‐1] ³ Downgraded because data were presented graphically and effects could not be estimated. ⁴ In this trial (Stupp 2017a) TTF_AC was compared with adjuvant chemotherapy (TMZ) only, after both arms had received concomitant CRT. Abbreviations: BEV_CRT: chemoradiotherapy plus bevacizumab; BEV_RT: radiotherapy plus bevacizumab; CI: confidence interval; CRT: chemoradiotherapy; HR: hazard ratio; IRI: irinotecan; RT: radiotherapy; TMZ: temozolomide; TTF_AC: tumour treating fields with adjuvant chemotherapy (after concomitant CRT) (TTF_AC); RIN_CRT: rindopepimut after CRT.

Summary of findings 5. Summary of findings on progression‐free survival

The effect of different treatment comparisons for newly diagnosed glioblastoma in the elderly on progression‐free survival
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: one treatment option Comparison: an alternative treatment option
Progression‐free survival	Relative effect (95% CI)	No of studies (elderly participants)	Certainty of the evidence (GRADE)	Comments
RT vs supportive care only	HR 0.28 (0.17 to 0.46)	1 study (81 participants)	⊕⊕⊕⊝ Moderate¹	Median time to progression was 3.5 months in the RT arm vs 1.3 months in the supportive care arm of this study (Keime‐Guibert 2007).
Hypofractionated RT vs standard RT (60Gy/30 fractions)	HR not reported	1 study (64 participants)	Not graded	This study (Roa 2015) reported that median progression‐free survival showed no statistically significant difference between arms (4.2 v 4.2 months in arms 1 and 2, respectively; P = 0.716).
TMZ vs RT	HR 1.15 (0.92 to 1.44)	1 study (373 participants)	⊕⊕⊝⊝ Low^2,3	No additional comments.
CRT vs RT	HR 0.50 (0.41 to 0.61)	1 study (562 participants)	⊕⊕⊕⊕ High	No additional comments.
BEV_CRT vs CRT	HR 0.78 (0.46 to 1.32)	1 study (73 participants)	⊕⊕⊝⊝ Low^1,2	In this study (AVAglio 2014), BEV_CRT did not increase overall survival either relative to CRT alone for elderly patients.
BEV_RT vs RT	HR 0.46 (0.27 to 0.78)	1 study (75 participants)	⊕⊕⊕⊝ Moderate¹	Despite delaying disease progression in this study (ARTE 2018), BEV_RT did not increase overall survival.
RIN_CRT vs CRT	Not reported separately for elderly subgroup.
TTF_AC vs CRT	Not reported separately for elderly subgroup.
IRI_BEV_RT vs CRT	Not reported separately for elderly subgroup.
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.
¹Sparse data from single studies [‐1] ² Serious risk of bias from attrition [‐1] ³ Serious imprecision Abbreviations:BEV_CRT: chemoradiotherapy plus bevacizumab; BEV_RT: radiotherapy plus bevacizumab; CI: Confidence interval; HR: hazard ratio;IRI: irinotecan; RT: radiotherapy; CRT: chemoradiotherapy; TMZ: temozolomide; TTF_AC: tumour treating fields with adjuvant chemotherapy (after concomitant CRT) (TTF_AC); RIN_CRT: rindopepimut after CRT.

Summary of findings 6. Summary of findings on severe adverse events

The effect of different treatments for newly diagnosed glioblastoma on severe adverse events
Patient or population: elderly people with newly diagnosed glioblastoma Settings: any Intervention: one treatment option Comparison: an alternative treatment option
Comparison	Narrative summary of evidence	No of studies (elderly participants)	Certainty of the evidence (GRADE)	Comments
RT vs supportive care only	Not reported.	NA	NA	None
Hypofractionated RT vs standard RT (60Gy/30 fractions)	This outcome was only reported in one small study and there were no instances of grade 3 or higher treatment‐related toxicity.	1 study (61)	Not graded	None
TMZ vs RT	TMZ probably increases the risk of thromboembolic and haematological adverse events	1 study (373)	⊕⊕⊕⊝ Moderate¹	Thrombocytopenia occurred in 24 TMZ vs 8 RT participants (RR 2.74, 95% CI 1.26 to 5.94) and lymphocytopenia occurred in 16 TMZ vs 2 RT participants (RR 7.30, 95% CI 1.70 to 31.31).
CRT vs RT	CRT probably increases the risk of haematological adverse events	1 study (562)	⊕⊕⊕⊝ Moderate¹	Neutropenia occurred in 22 CRT vs 2 RT participants (RR 10.30; 95% CI 2.45 to 43.34); throbocytopenia occurred in 30 CRT vs 1 RT participants (RR 28.56, 95% 3.92 to 207.86); lymphocytopenia occurred in 73 CRT vs 26 RT participants (RR 2.65, 95% CI 1.75 to 4.01); leucopenia occurred in 19 CRT vs 1 RT participant (RR 18.16, 96% CI 2.45 to 124.64); and anaemia occurred in 3 CRT vs 0 RT participants (RR 6.69, 95% CI 0.35 to 128.88).
BEV_CRT vs CRT	BEV_CRT probably increases the risk of grade 3+ thromboembolic events.	1 study (73)	⊕⊕⊕⊝ Moderate¹	Other adverse events data were not available for elderly subgroup specifically. For the larger study sample, Grade 3+ cerebral bleeding events (2.0% versus 0.9%) and wound healing events (3.3% vs 1.6%) were higher in the BEV plus CRT arm versus CRT alone. There were also higher rates of Grade 3+ thrombocytopenia (15% vs 9.8%) and infection rates (12.8% versus 7.8%) in the BEV_CRT arm.
BEV_RT vs RT	There was little or no difference in thromboembolic, haematological, and other severe adverse events reported in this one small study.	1 study (75)	⊕⊕⊝⊝ Low¹	An example of these non‐statistically significant findings for thrombocytopenia are that this SAE occurred in 8 BEV_RT vs 2 RT participants (RR 2.00, 95% CI 0.46 vs 8.73). Haematological SAEs occurred in 2 vs 0 participants, respectively (RR 2.55, 95% CI 0.13 vs 51.17).
RIN_CRT vs CRT	This was reported for overall trial but not for elderly subgroup specifically.	1 study (174 participants)	Not graded	The most common severe adverse events for the experimental (rindopepimut) versus control arm of the trial were: thrombocytopenia (9% vs 6%), fatigue (2% vs 5%), brain oedema (2% vs 3%), seizure (2% vs 2%) and headache (2% vs 3%). There was one death, secondary to pulmonary embolism, that was assessed as potentially related to the treatment in the experimental arm.
TTF_AC vs AC	This was reported for overall trial but not for elderly subgroup specifically.	1 study (134 participants)	Not graded	Overall, it was reported that there was no significant increase in rates of severe adverse events when TTF were added to adjuvant chemotherapy (48% vs 44%, P = 0.58).
IRI_BEV_RT vs CRT	This was reported for overall trial but not for elderly subgroup specifically.	1 study (34 participants)	Not graded	Overall, rates of severe adverse events were 72% in the experimental arm and 84% in the CRT arm. In the experimental arm, severe vascular events were most common (11.8%) and two cerebral haemorrhages occurred (one fatal). For patients in the CRT arm, severe haematological toxicity was most common (18.2%).
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.
¹ Downgraded ‐2 for sparse data from small single study and imprecision Abbreviations: BEV_CRT: chemoradiotherapy plus bevacizumab; BEV_RT: radiotherapy plus bevacizumab; CI: confidence interval; HR: hazard ratio; RT: radiotherapy; CRT: chemoradiotherapy; IRI: irinotecan; NA: not applicable; TMZ: temozolomide; TTF_AC: tumour treating fields with adjuvant chemotherapy (after concomitant CRT) (TTF_AC); RIN_CRT: rindopepimut after CRT.

Background

Description of the condition

Glioblastoma multiforme is a high‐grade, aggressive primary tumour of the central nervous system with a poor prognosis. The incidence of glioblastoma is increasing and this rise is most rapid in the elderly (Ferguson 2014). Use of the term 'the elderly' in relation to glioblastoma commonly refers to people 70 years and older (NCCN 2018). Age is an important consideration in the treatment of glioblastoma as it is a negative prognostic indicator (Lorimer 2017). A Surveillance Epidemiology and End Results (SEER) population analysis reported that for every year increase in patient age, there was a statistically significant decrease in survival (Thumma 2012). Median survival drops from about 12 to 18 months for younger people with glioblastoma, to three to six months for older age cohorts (Brodbelt 2015).

The molecular status of glioblastoma is also an important prognostic factor and several molecular subtypes of glioblastoma have been recognised (Lara‐Velazquez 2017). One of the most important molecular signatures is O⁶‐methylguanine‐DNA‐methyltransferase (MGMT) promoter methylation, which has been shown to confer predictive and prognostic benefit (Malmstrom 2012; Yin 2014). Treatment for glioblastoma is not curative and the natural history of the disease is that patients will relapse after treatment and it will ultimately be a fatal condition (Louis 2016). Retrospective studies have shown that older people are less likely to get aggressive, multi‐modality treatment (Iwamoto 2008; Lorimer 2017; Paszat 2001), but people with glioblastoma across all age groups who do get active treatment live longer (Brodbelt 2015). Direct healthcare costs for the management of malignant gliomas (malignant glioma encompasses anaplastic glioma, i.e. World Health Organization (WHO) grade 3 and 4) have been estimated at USD 32,764 per patient (2011 data; Raizer 2015).

Description of the intervention

The ‘standard of care' of treatment for patients aged under 70 years of age with glioblastoma consists of surgery followed by radiotherapy (60 Gy in 30 fractions) with concomitant and adjuvant temozolomide (TMZ) chemotherapy (NCCN 2018; Stupp 2005). This management plan is less often used in the elderly for the following reasons.

People over 70 years old were not included in the landmark trial (Stupp 2005), and a subsequent communication of the results of an exploratory subgroup analysis revealed that the survival benefit in this trial was not statistically significant for a subgroup of people aged 66 to 70 years (Laperriere 2013).
Shorter radiotherapy courses or chemotherapy alone can lead to better outcomes for the elderly than the standard course of radiotherapy. Patients rarely live long enough to develop late complications from radiation therapy, therefore larger fraction size may be justified to allow for a shortened course of treatment.
Both chemotherapy and radiotherapy treatment toxicities are often greater in the elderly (Lawrence 2011; Sijben 2008).
The shorter predicted survival time for older people with glioblastoma means that they might spend much of this time recovering from the six‐week course of radiotherapy.

Small prospective (Vuorinen 2003), and retrospective studies (Chaichana 2011a; Chaichana 2011b), have shown that, for people aged 65 and over with glioblastoma, maximal debulking (resection) is associated with better survival and a trend to longer time remaining independent versus biopsy alone. Therefore maximal resection, if feasible, is the recommended primary approach to glioblastoma in the elderly (NCCN 2018). Depending on a person's performance status, radiotherapy or chemotherapy, or both, can then be added. As it remains unclear which treatment is best for glioblastoma in the elderly, participation in clinical trials is strongly encouraged (NCCN 2018). There is little evidence to guide treatment of recurrent glioblastoma in the elderly and approaches are based on retrospective studies (Socha 2016).

Treatment with either radiotherapy or chemotherapy

A randomised trial of radiotherapy (50 Gy delivered over a period of five to six weeks) versus best supportive care showed that radiotherapy conferred a 12‐week survival benefit in older people with malignant glioma (Keime‐Guibert 2007). Another randomised trial found that radiotherapy (60 Gy over a period of six to seven weeks) was as effective as intensive ("dose‐dense") TMZ chemotherapy alone (Wick 2012). There is increasing interest in using hypofractionated radiotherapy (radiotherapy delivered over shorter period of time, e.g. 40 Gy in 15 fractions over three weeks and 34 Gy in 10 fractions over two weeks) for older people with glioblastoma, as these have been found to have similar survival benefits compared to the standard regimen of 60 Gy in 30 fractions over a period of six weeks (Malmstrom 2012; Roa 2004).

Combination treatment

A randomised trial has shown that adding TMZ to hypofractionated radiotherapy for older people with glioblastoma confers a survival advantage compared to hypofractionated radiotherapy alone (Minniti 2012; Perry 2017), but not necessarily for those people with MGMT unmethylated tumours.

How the intervention might work

Surgery is an important step in the treatment of glioblastoma. Also, there is evidence that surgery improves one‐ and two‐year survival rates compared to biopsy alone (Brown 2016). The extent of surgery can be divided into three main categories which have different definitions in the literature: ‘maximal' debulking or gross total resection (GTR), subtotal resection (STR), and biopsy. The role of maximal debulking surgery is to minimise the tumour volume that remains to optimise the impact of subsequent treatment modalities, which are likely to be more effective against small volume tumours (Lara‐Velazquez 2017).

Radiotherapy is delivered to the primary tumour or the surgical cavity with a margin to account for microscopic spread, patient movement, and set‐up error (Niyazi 2016). One of the most important mechanisms of action of radiation therapy is the promotion of double strand breaks in DNA which, if left unrepaired, will result in cell death (Baskar 2014). DNA damage is more likely to occur in rapidly dividing cells, such as glioblastoma tumour cells, rather than normal brain which has a slower rate of cellular turn over. This provides the therapeutic index between the tumour and normal surrounding tissue.

Systemic chemotherapy can enhance the therapeutic effect of radiotherapy but is also an effective treatment on its own. The most widely used chemotherapy agent for newly diagnosed glioblastoma is TMZ, which acts as a DNA alkylating agent (Zhang 2012). Those tumours with MGMT‐promoter methylation lack the MGMT enzyme which repairs the cytotoxic damage caused by TMZ, thereby making tumour cells more chemosensitive.

Why it is important to do this review

Previous research has demonstrated that increasing age has an important effect on overall survival and tolerability of treatment for patients with a diagnosis of glioblastoma (Thumma 2012). Increasing age, regardless of performance status, has an important influence on treatment decisions made by clinicians (Palmer 2018), however there is still a lack of consensus on the optimal treatment options for the elderly subgroup of patients with glioblastoma.

It is recognised that treating older people with glioblastoma presents unique challenges and that the standard approach is not always appropriate. There have been several randomised trials in recent years that have tested therapeutic strategies specifically for older people with glioblastoma (e.g. Malmstrom 2012; Perry 2017 Roa 2004; Wick 2012). Other trials including younger people have also performed subgroup analysis to test if therapeutic benefit is maintained in older people. Due to the variation in age thresholds to define the ‘elderly', performance status, treatment regimens, and molecular subtypes, it has been difficult to translate these individual studies into clinical practice. This is also because the focus of many intervention trials is on survival, which might not be the most important outcome to elderly people with glioblastoma; rather, the quality of the remainder of their life might be their most important consideration. As the median age of diagnosis is around 64 years of age (Ostrom 2015), a significant proportion of newly diagnosed patients fall into the 'elderly' category.

There is been some evidence to suggest that total direct costs of care associated with glioblastoma have been increasing over recent years with the increased use of costly systemic anti‐cancer treatments (Henaine 2016; Ray 2014). Selecting the appropriate management strategy for an elderly patient group is important from a quality of life perspective and also has significant resource implications (Raizer 2015). It has been estimated the average cost for a regimen of TMZ to treat a person with newly diagnosed glioblastoma is USD 46,693 (USD in 2018 converted from NZD 2005) (Hamilton 2005). It is therefore important to understand the costs and benefits to avoid implementing costly and potentially toxic treatment for little clinical benefit.

Currently there is no clear consensus on how to apply the available evidence to guide treatment of the individual person seen in clinic. A systematic review and network meta‐analysis (NMA) of randomised trials would help to inform the best approach to the treatment of older individuals with newly diagnosed glioblastoma and help to identify research gaps.

Objectives

Methods

Criteria for considering studies for this review

Types of studies

Randomised controlled trials (RCTs) for evidence on effectiveness and safety.
Full economic evaluations (cost‐effectiveness analyses, cost‐utility analyses, and cost‐benefit analyses) conducted alongside any study design and any model‐based economic evaluations for economic evidence.

Types of participants

Elderly people undergoing treatment for histologically confirmed newly diagnosed glioblastoma. For the purpose of this Cochrane Review, we defined ‘elderly' as 70 years and older; however, where investigators defined the ‘elderly' as over 65 years of age, we included these studies. We included studies of people of all ages that reported subgroup findings for elderly people (over 65 or 70 years of age) provided the participants in the subgroup numbered more than 20. We also included the mixed data if it was clear that 80% or more of participants in the study were over the age of 65 years. Similarly, where the study population included both grade 3 or 4 gliomas (anaplastic astrocytomas or glioblastoma), we tried to obtain separate data for participants with glioblastoma; if this was not possible, we considered including the study if more than half the study population had glioblastoma.

Types of interventions

Interventions evaluated alone or in combination with each other versus any of the other interventions included the following.

Radiotherapy (standard, hypofractionated, and other techniques).
Chemotherapy (temozolomide (TMZ) and other types).

We included all available regimens of radiotherapy and chemotherapy that were evaluated in randomised trials. If we identified interventions in the included studies of which we were not aware, we considered including them after we assessed their comparability with those interventions named above. We excluded phase 1 and 2 studies of novel interventions that have been shown to be detrimental and have not been developed further.

It was not possible to create separate networks according to the type of surgical procedure (gross total resection (GTR), subtotal resection (STR), and biopsy only). Within each network we assumed that any participants within the network could be randomised to any of the interventions e.g. an elderly person with histologically confirmed glioblastoma could be equally likely to be randomised to standard radiotherapy, chemotherapy, any combination of these or supportive care.

Types of outcome measures

Primary outcomes

Overall survival (time from randomisation to death from any cause).
Quality of life (QoL), as measured using a standardised questionnaire, e.g. the European Organisation for Research and Treatment of Cancer (EORTC) QLQ‐C30 or QLQ‐BN20 (specific for brain cancer), or the Functional Assessment of Cancer Therapy scale (FACT‐G (general) or FACT‐Br (specific for brain cancer)).

Secondary outcomes

Progression‐free survival (time from randomisation to disease progression or death from any cause).
Severe adverse events, according to standardised scales, e.g. Common Terminology Criteria for Adverse Events (CTCAE).
Cognitive impairment (objective or subjective), as measured by an overall cognitive function score, as a change‐over‐time score, or reported as individual cognitive function domains, e.g. verbal fluency, processing speed, memory, attention, and executive functioning, using a standardised measurement tool, e.g. Mini Mental State Exam (MMSE), EORTC, FACT.
Functional impairment or disability, as measured by an overall ability score and/or as a change of ability over time score using a standardised measurement tool, e.g. Karnofsky Performance Status Scale, Neurological Functions Score, EORTC, FACT; or as a categorical outcome as defined by investigators.
Fatigue, according to CTCAE, EORTC, or as defined by investigators.
Economic outcomes:
- resource use for health care;
- health state utilities;
- costs of health care;
- incremental cost‐effectiveness.

Search methods for identification of studies

Electronic searches

For studies on the effects of the interventions, we searched the following databases to 3 April 2019:

The Cochrane Central Register of Controlled Trials (CENTRAL; 2019, Issue 4), in the Cochrane Library;
MEDLINE via Ovid (1946 to March week 4 2019);
Embase via Ovid (1980 to 2019 week 13).

For economic evidence we searched the following databases:

MEDLINE via Ovid (January 2015 to March week 4 2019);
Embase via Ovid (January 2015 to 2019 week 13);
NHS Economic Evaluation Database (EED) to December 2014.

The NHS EED database was searched up to the end of December 2014 (when the last records were added to that database) and MEDLINE and Embase from 1 January 2015, as the NHS EED already included comprehensive searches of these databases prior to 2015. We also considered relevant grey literature (such as health technology assessments, reports, and working papers) for inclusion.

Please refer to Appendix 1 for the CENTRAL, MEDLINE, and Embase search strategies.

We did not apply language restrictions to any literature searches.

Searching other resources

We searched the following for ongoing trials.

ClinicalTrials.gov (clinicaltrials.gov/)
WHO International Clinical Trials Registry Platform (ICTRP) (apps.who.int/trialsearch/)

If ongoing trials that have not been published were identified through these searches, we approached the principal investigators to ask for an update on the trial status and any relevant unpublished data, if available.

We used the related articles feature of PubMed and handsearched the reference lists of included studies to identify newly published articles and additional studies of relevance. We also handsearched conference proceedings from 2014 to 2018 (five years) of the British Neuro‐Oncology Society, the Society of Neuro‐Oncology, the European Association of Neuro‐Oncology and the World Federation of Neuro‐Oncology Societies conferences for relevant ongoing or unpublished studies.

Data collection and analysis

Selection of studies

For the results of search 1 (trials of effects of interventions), the Information Specialist at the Cochrane Gynaecological, Neuro‐oncology and Orphan Cancer Group (CGNOC) downloaded all titles and abstracts retrieved by electronic searching to EndNote X8 and removed duplicates. Two review authors (TAL, CH, or ER) independently screened the remaining records and excluded studies that clearly did not meet the eligibility criteria. For potentially eligible records, copies of the full texts were obtained and three review authors (TAL,CH and ER) independently assessed them for eligibility. The respective review authors resolved any disagreements through discussion and, if necessary, consulted at least one other review author. We used Covidence to facilitate this study selection process and documented the reasons for exclusion of studies accordingly.

To inform the economic outcomes, full economic evaluations (cost‐effectiveness analyses, cost‐utility analyses, and cost‐benefit analyses), we considered cost analyses and comparative resource‐utilisation studies. Studies carried out alongside relevant RCTs and model‐based studies were considered for inclusion. Two review authors (TR and AK) independently screened for eligible studies.

Data extraction and management

Two review authors (TAL, CH, or ER) independently extracted data from included studies using a pre‐designed data extraction form (Higgins 2011). We extracted the following data.

Author contact details.
Country.
Setting.
Dates of participant accrual.
Funding source.
Inclusion and exclusion criteria.
Study design.
Study population and baseline characteristics:
- number of participants enrolled;
- number of participants analysed;
- age;
- gender.
Potential effect modifiers:
- molecular type of glioblastoma;
- performance status.
Intervention details:
- type of intervention, dose, timing, and other regimen details;
- Type of comparator.
Risk of bias assessment (see below).
Duration of follow‐up.
Primary outcome(s) of the study.
Review outcomes:
- For time‐to‐event outcomes (overall and progression‐free survival) we extracted the hazard ratio (HR) with its 95% confidence interval (CI) for time points as reported by the study authors. We noted the definition of and procedure used to identify progression. Where reported, we also extracted dichotomous data for these outcomes at author‐specified time points.
- For dichotomous outcomes (e.g. serious adverse events), we extracted the number of participants in each treatment arm that experienced the outcome of interest and the number of participants assessed.
- For continuous outcomes (e.g. QoL scores), we extracted the value and standard deviation of the outcome of interest and the number of participants assessed at the relevant time point in each group. We also extracted change‐from‐baseline score data where reported and noted the type of scale used.
- We extracted adjusted statistics where reported.
- Where possible, all data extracted were those relevant to an intention‐to‐treat analysis, in which participants were analysed in the groups to which they were assigned.
- We resolved differences between review authors by discussion or by appeal to a third review author when necessary.

Assessment of risk of bias in included studies

We assessed the risk of bias using Cochrane's ‘Risk of bias' tool and the criteria specified in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). This included assessment of the following.

Random sequence generation.
Allocation concealment.
Blinding of participants and healthcare providers.
Blinding of outcome assessors.
Incomplete outcome data (more than 20% missing data considered high risk).
Selective reporting of outcomes.
Other possible sources of bias, e.g. lack of a power calculation, baseline differences in group characteristics.

Two review authors (ER and CH) independently assessed risk of bias and resolved any differences in opinion by discussion or by consulting a third review author (TAL). We summarised judgements in ‘Risk of bias' tables along with the characteristics of the included studies and interpreted the results of meta‐analyses in light of the overall ‘Risk of bias' assessment. For more details about the ‘Risk of bias' assessment see Appendix 2.

We assessed economic evaluation studies for bias in two stages. The first stage involved assessing risk of bias from the sources of the effectiveness data. In economic evaluations carried out alongside clinical trials, we assessed these using the Cochrane ‘Risk of bias' tool, as described above. If the economic evaluation was model‐based, we used the ROBIS tool to assess bias in the effectiveness studies (Whiting 2016). The second stage involved assessing the risk of bias of the economic evidence (i.e. assessing the overall methodological quality). This was done using the Consolidated health Economic Evaluation Reporting Standards (CHEERS) checklist (Husereau 2013).

Measures of treatment effect

Effectiveness data

For time‐to‐event outcomes (e.g. overall survival), we extracted the hazard ratio (HR) with its 95% confidence interval (CI).
For continuous outcomes (e.g. QoL scores) we assumed that study authors would use different measurement scales, therefore, we planned to estimate the standardised mean difference (SMD) and its 95% CI using the pooled data. However, if the same measurement scale was used, we estimated the mean difference (MD) and its 95% CI. If studies did not report total values but, instead, reported change‐from‐baseline outcomes, we combined these change values with total measurement outcomes by using the (unstandardised) MD method in Review Manager 5 (RevMan 5) (RevMan 2014). We used subgroups to distinguish between MDs of change scores and MDs of final values, and pooled the subgroups in an overall analysis (Higgins 2011).
For dichotomous outcomes, we calculated the effect size as a risk ratio (RR) with its 95% CI.

Economic data

Two review authors (AK and TR) independently extracted data from relevant economic studies and summarised this information in tables. We extracted data on the following.

Type of evaluations.
Sources of effectiveness data.
Cost data.
Sources of cost data.
Sources of outcome valuations.
Analytical approach.

Unit of analysis issues

Two review authors (TAL and ER) assessed unit of analysis issues according to Higgins 2011, and resolved any differences in opinion by discussion. These included reports where there are multiple observations for the same outcome (e.g. repeated measurements with different scales or at different time points, recurring events). If meta‐analysis was not feasible or meaningful, we extracted data from all scales or time points and attempted to describe them narratively.

Multi‐arm trials

We included multi‐arm trials in this review. We treated multi‐arm studies as multiple independent comparisons in pairwise meta‐analyses and did not combine data from different arms. In the network meta‐analysis (NMA) we accounted for the correlation between the effect sizes derived from the same study (White 2015).

Dealing with missing data

We did not impute missing data. In the event of missing data, we wrote to study authors to request the data on primary outcomes and describe in the ‘Characteristics of included studies' tables how any missing data were obtained.

Assessment of heterogeneity

Assessment of clinical and methodological heterogeneity

We assessed clinical heterogeneity between studies by comparing characteristics of included participants, and interventions in each meta‐analysis of each comparison, by visual inspection of forest plots, by estimation of the percentage heterogeneity between trials which cannot be ascribed to sampling variation (Higgins 2003), by a formal statistical test of the significance of the heterogeneity (Deeks 2001), and, where possible, by subgroup analyses. If there was evidence of substantial heterogeneity, we investigated and reported the possible reasons for this.

Assessment of consistency across treatment comparisons

We examined the assumption of consistency by assessing the distribution of potential effect modifiers across the pairwise comparisons (Cipriani 2013; Jansen 2013; Salanti 2012). The assumption would hold if the following were true.

The common treatment used to compare different interventions indirectly was similar when it appeared in different trials.
All pairwise comparisons did not differ with respect to the distribution of effect modifiers.

Assessment of statistical heterogeneity and inconsistency

Assumptions when estimating the heterogeneity

In standard pairwise meta‐analyses, we estimated different heterogeneity variances for each pairwise comparison. In the NMA, we assumed a common estimate for the heterogeneity variance across the different comparisons (White 2015).

Measures and tests for heterogeneity

We assessed the presence of statistical heterogeneity within the pairwise comparisons using the I2 statistic, which is the percentage of variability that cannot be attributed to random error (Higgins 2003).

Assessment of statistical inconsistency

We were not able to assess statistically the global agreement between the various sources of evidence in a network of interventions (consistency). However, we were able to apply a local approach using a node‐splitting method (Dias 2010).

Assessment of reporting biases

In pairwise comparisons, if there were 10 or more studies included in meta‐analyses, we had planned to investigate reporting biases (such as publication bias) using funnel plots. However, in none of the analyses were 10 or more studies included.

Data synthesis

Methods for direct treatment comparisons

Initially we performed standard pairwise meta‐analyses for each comparison using the random‐effects model in Stata statistical software version 15.1 (STATA) and Review Manager software (RevMan 2014).

Methods for indirect and mixed comparisons

We conducted network meta‐analyses within a frequentist framework using multivariate meta‐analysis (White 2015), if we considered participants, comparisons, and outcomes to be sufficiently similar to ensure an answer that was clinically meaningful. We also used STATA commands for visualising and reporting NMA results (Chaimani 2015). If meta‐analysis was not possible but limited data were available, we attempted to synthesise narrative summaries according to guidance in the Cochrane Handbook.

We summarised characteristics and results of included economic evaluations using additional tables, supplemented by a narrative summary that compared and evaluated methods used and principal results between studies. Unit cost data were also tabulated, when available. We reported the currency and price year applicable to measures of costs in each original study alongside measures of costs, incremental costs, and incremental cost‐effectiveness by study. Where details of currency and price year were available in original studies, we converted measures of costs, incremental costs, and cost‐effectiveness to (latest year) international dollars value using implicit price deflators for gross domestic product (GDP) and GDP Purchasing Power Parities (EPPI Centre Cost Converter 2016). Details of the methodological characteristics of individual included health economics studies was summarised in Characteristics of included studies tables. All elements of the economics component of this review were conducted according to current guidance on the use of economics methods in the preparation and maintenance of Cochrane Reviews (Higgins 2011; Shemilt 2018; Wijnen 2016).

‘Summary of findings' tables and results reporting

Effectiveness summary of findings

We presented the primary outcomes in the 'Summary of findings' tables. Evidence for pairwise comparisons was assessed based on GRADEpro Guideline Development Tool (GDT) methods (GRADEpro 2015) (i.e. we assessed risk of bias, inconsistency, imprecision, indirectness and publication bias), whereas network evidence was assessed using the approach suggested by Puhan 2014 and advanced by Brignardello‐Petersen 2018. Narrative evidence summaries were prepared if data could not be synthesised and assessed according to the GRADE approach suggested by Murad 2017. The certainty of pairwise and network evidence for each outcome was rated as ‘high’, ‘moderate’, ‘low’ or ‘very low’ as defined according to the GRADE approach.

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.

To assess the network evidence, we assessed the certainty of the direct evidence (if any), the indirect evidence (if estimable) and the network evidence in this order. Direct evidence was assessed using the standard (pairwise) GRADE approach, but without assessing imprecision (i.e. we assessed risk of bias, inconsistency, indirectness and publication bias). Indirect evidence ratings, based on the certainty rating of the lower of the two arms forming the loop in the network diagram, were assessed when they contributed more than the direct evidence to the network estimates. The final step was to assess the certainty of the network effect estimate based on whether intransitivity was present (i.e. whether there were differences in study characteristics that may modify the effect in the direct comparisons that form the basis for the indirect estimate; Puhan 2014). The network estimate was assessed in the first instance as being equivalent to the higher of the direct and indirect estimates, and incoherence and imprecision were then considered, with downgrading by one level accordingly if serious. Where no direct evidence was available and when the treatments did not have a common comparator, we presented the network estimate but did not rate the certainty of the evidence. Where possible, we estimated the absolute effects of treatments relative to the effect of a given reference comparator based on an assumed risk, the source of which was stated. For median survival times, we based illustrative absolute effects on hazard ratios (HRs).

'Summary of findings' tables were designed following the approach suggested by Schunemann 2009 and by Yepes‐Nuñez 2019. In the 'Summary of findings' tables we provided justification for each assessment about the confidence in the estimates of effect (e.g. reasons for downgrading the certainty of the evidence), with confidence assessed as 95% credible intervals (CrI). Two review authors (TAL and ER) independently assessed the certainty of the evidence. We resolved any differences of opinion by discussion. We interpreted the graded evidence based on the Cochrane Effective Practice and Organisation of Care (EPOC) Group's guidance (Cochrane EPOC 2015) and, for time‐to‐event evidence, on suggestions in Barraclough 2011.

Relative treatment ranking

We computed ranking of probabilities for all included treatments and obtained a treatment hierarchy using the surface under the cumulative ranking curve (SUCRA). For primary outcomes, we assessed the robustness of these findings in sensitivity analysis.

Economic evaluation summary of findings

For the economic evaluation studies, we presented the following findings in a table.

Method of economic evaluation
Costs
Outcomes
Incremental cost‐effectiveness ratio

Subgroup analysis and investigation of heterogeneity

For pairwise comparisons we assessed heterogeneity using the I² statistic that measures the percentage of variability that cannot be attributed to random error (Higgins 2003). We considered clinical heterogeneity and risk of bias in the interpretation of any heterogeneity. The certainty of the evidence was downgraded for heterogeneity where I² ≥ 60%.

Due to sparse structure of the network, we assumed no substantial statistical heterogeneity and fitted a fixed‐effect model. However, we performed sensitivity analyses to examine the impact of our original assumptions by applying an alternative classification of radiotherapy with 50 Gy in one study (Keime‐Guibert 2007); removing one of the arms from three‐arm trial (hypofractionated radiotherapy Malmstrom 2012), and splitting the chemoradiotherapy node according to the radiation dose (hypofractionated radiotherapy and 60 Gy).

For primary outcomes, we had planned to assess findings by the different age thresholds used by investigators to define the elderly and by MGMT methylation status; however, data were insufficient for these subgroup analyses.

Sensitivity analysis

We performed sensitivity analyses to investigate assumptions that we made to facilitate a connected network, including:

pooling data from a study utilising a radiotherapy dose of 50Gy with studies utilising 60 Gy or 40 Gy dose schedules;
pooling data from study arms utilising combined chemoradiation, where studies utilised 60 Gy or 40 Gy radiotherapy dose schedules.

We based these assumptions on calculations of the equivalent doses (EQD2) and biologically effective doses (BED) of the different radiotherapy schedules utilised in included studies (Table 1). Note that data from trials using 34 Gy/10 fractions were pooled with those of 40 Gy/15 schedules in our NMA, as we considered the BEDs of these schedules to be sufficiently similar.

Table 1. Table of radiotherapy regimens used in included studies and biologically effective doses

Dose fractionation	EQD2 (Gy)	BED (Gy)
60 Gy/30 fractions	60	75
50 Gy/28 fractions	49	61
40 Gy/15 fractions	42	53
34 Gy/10 fractions	39	48
25 Gy/5 fractions	33	41

EQD2 and BED calculated for an alpha/beta of 8

EQD2 = equivalent dose; BED = biologically effective dose

Gy = Gray

We also conducted sensitivity analysis to investigate the impact of a single three‐arm study forming the only loop in the network and to justify the lack of assessment of inconsistency (see above).

Results

Description of studies

Results of the search

Intervention studies

Searches conducted on the 13 June 2018 and the 3 April 2019 led to the identification of 12 included studies (with 31 associated records) and two potentially eligible ongoing studies (with three associated records). We identified the following numbers of records through the first electronic database search.

MEDLINE: 1946 to May week 5 2018 – 930 records
Embase: 1980 to 2018 week 24 – 848 records
CENTRAL: Issue 5 2018 – 1571 records

The results of this initial search are summarised in Figure 1. Following de‐duplication across the databases, the combined total yield was 2493 records. The Information Specialist at the Cochrane Gynaecological, Neuro‐oncology and Orphan Cancer (CGNOC) Group ran these records through the Cochrane RCT 'Classifier’, which uses machine learning to identify records that are likely to be randomised controlled trials (RCTs). Following classification, 1462 records were identified as having more than a 10% likelihood of being RCTs, whilst 1031 references had less than a 10% likelihood of being RCTs. The Information Specialist then de‐duplicated the remaining 1462 records and sifted out the clearly irrelevant records (e.g. those that related to other types of cancers). Two study authors (CH, TAL) independently screened the remaining yield of 990 records. Out of these, 12 studies (ARTE 2018; AVAglio 2014; GLARIUS 2016; Green 1983; Keime‐Guibert 2007; Malmstrom 2012; Perry 2017; Roa 2004; Roa 2015; Stupp 2017a; Weller 2017; Wick 2012) with 31 associated records were finally included (Figure 1). Additionally, one ongoing study was identified (NCT01602588).

Figure 1

Study flow diagram (search date 13 June 2018)

The top‐up search on 3 April 2019 yielded 125 additional records to be screened on title and abstract. After de‐duplication and screening on title and abstract, six full‐text papers were retrieved. Three of these papers were additional publications related to two already included studies (AVAglio 2014; GLARIUS 2016), the other three were excluded with reasons. Additionally, searches of clinical trial registries and of relevant society conference proceedings from 2014 to 2018 identified one ongoing trial (NUTMEG 2018) and four potentially eligible records, respectively. The ongoing trial was added to the Ongoing studies section, including one conference abstract (NUTMEG 2018). The other three conference abstracts were classified as excluded studies. The results of the top‐up search are summarised in Figure 2.

Figure 2

Study flow diagram (search date 3 April 2019).

Economic studies

For economic studies, we identified the following numbers of records through electronic database searches conducted on 13 June 2018.

MEDLINE: 1946 to May week 5 2018 – 113 records
Embase: 1980 to 2018 week 24 – 151 records

Following de‐duplication across these databases, the total yield to be sifted was 101 records. The top‐up search conducted on the 3 April 2019 yielded an additional 22 records. Five titles and abstracts were identified for full‐text screening (Ghosh 2018, Jiang 2017; Moroney 2017; Roussakou 2017; Waschke 2018), one of which (Ghosh 2018) was included.

Included studies

We included 12 RCTs, six were conducted exclusively among elderly people (either defined as 65 years or older or 70 years or older) with newly diagnosed glioblastoma (ARTE 2018; Keime‐Guibert 2007; Perry 2017; Roa 2004; Roa 2015; Wick 2012). The other six RCTs included patients from a broader age range and reported some data separately for their elderly subgroup (AVAglio 2014; GLARIUS 2016; Green 1983; Malmstrom 2012; Stupp 2017a; Weller 2017), which we extracted for this review.

Numbers recruited and analysed

Altogether, approximately 1818 elderly participants involved in the included studies contributed data to the review. In seven studies, the elderly participants analysed numbered less than 100. In five studies (Green 1983; Malmstrom 2012; Perry 2017; Stupp 2017a; Wick 2012); the number analysed was more than 100, equalling to 107, 123, 562, 134, and 373 participants, respectively.

Location of studies

Six studies were conducted in the following individual countries: Canada (Roa 2004), France (Keime‐Guibert 2007), Germany (Wick 2012; GLARIUS 2016), the USA (Green 1983); Switzerland (ARTE 2018); the rest were multi‐country studies (AVAglio 2014; Malmstrom 2012; Perry 2017;Roa 2015; Stupp 2017a; Weller 2017).

Dates of recruitment

Accrual occurred before 1980 in one study (Green 1983) and between 1996 and 2001 in another (Roa 2004). In all other studies, accrual occurred from 2000 onwards.

Funding

Seven studies were funded by pharmaceutical companies (ARTE 2018: Roche Pharmaceuticals; AVAglio 2014: Hoffmann‐La Roche; GLARIUS 2016: Roche Pharmaceuticals; Malmstrom 2012: Merck; Perry 2017: Schering‐Plough/Merck; Weller 2017: Celldex Therapeutics; Wick 2012: Merck, Sharp & Dohme; two of these (Malmstrom 2012; Perry 2017) also received grants from national cancer research funds. One study (Stupp 2017a) received funding from a medical device company, Novocure Ltd). The rest were funded by research grants from national cancer research funds or charities.

Characteristics of study participants

Also see Table 2.

Table 2. Characteristics of included studies

Study ID	Elderly definition (years)	No. analysed	Male gender (%)	Performance status	MGMT methylated/ unmethylated/ unknown (%)	Treatment option
Study ID	Elderly definition (years)	No. analysed	Male gender (%)	Performance status	MGMT methylated/ unmethylated/ unknown (%)	1	2	3	4
ARTE 2018	≥ 65	75	64	KPS ≥ 60	21%/73%/5%*	RT₄₀	RT₄₀+ BEV	‐	‐
AVAglio 2014	≥ 70	73	63	WHO 0‐2	26%/59%/24%	RT₆₀+ TMZ + maintenance	RT₆₀+ TMZ + maintenance + BEV	‐	‐
GLARIUS 2016	≥ 65	34	67^c	KPS ≥ 70	100% unmethylated	RT₆₀+TMZ	RT₆₀+BEV+IRI	‐	‐
Green 1983	≥ 65	107	65	KPS ≥ 70	‐	RT₆₀+carmustine	RT₆₀+steroid	RT₆₀+procarbazine	RT₆₀+BCNU+steroid
Keime‐Guibert 2007	≥ 70	85	63	KPS ≥ 70	‐	RT₅₀	Supportive care	‐	‐
Malmstrom 2012	≥ 70^a	123	59	WHO 0‐2	45%/55%/NR^e	RT₆₀	RT_30‐34	TMZ	‐
Perry 2017	≥ 65	562	61	ECOG 0‐2	47%/53%/NR^f	RT₄₀	RT₄₀ + TMZ + maintenance TMZ	‐	‐
Roa 2004	≥ 60^b	95	47	KPS ≥ 50	‐	RT₆₀	RT₄₀
Roa 2015	≥ 65^c	61	58	KPS ≥ 50	‐	RT₄₀	RT₂₅	‐	‐
Stupp 2017a	≥ 65	134	68	KPS ≥ 70	37%/53%/9%	RT₆₀+TMZ + maintenance TMZ	RT₆₀+TMZ + maintenance TMZ + TTF	‐	‐
Weller 2017	≥ 65	96	63^d	ECOG 0‐2	34%/59%/7%	RT₆₀+TMZ	RT₆₀+ TMZ + maintenance TMZ + RIN	‐	‐
Wick 2012	≥ 65	373	47	KPS ≥ 60	20%/36%/44%	RT₆₀	TMZ	‐	‐

BEV: bevacizumab; ECOG: Eastern Cooperative Oncology Group; Gy: Grays; IRI: irinotecan; KPS: Karnofsky performance score; RIN: rindopepimut; RT: radiotherapy; TTF: tumour treating fields; TMZ: temozolomide; WHO: World Health Organization

^a The whole sample (n = 291) comprised participants ≥ 60 years. The median age was 70 years for all study groups, ranging between 60 and 88 years.

^bMedian age was approximately 72 with a standard deviation of 5 years

^c Data for the ≥ 65 year age group were reported in the Guedes de Castro 2017 substudy report.

^d Gender data specific to the elderly subgroup were not reported separately

^e For approximately 70% of participants with MGMT data available

^f For approximately 63% of total participants with MGMT data available

Age

Eight studies defined older patients using an age threshold of 65 years, two studies (AVAglio 2014; Keime‐Guibert 2007) used an age threshold of 70 years, and two studies recruited participants from 60 years of age (Malmstrom 2012; Roa 2004). One of the latter studies defined an older subgroup using a threshold of 70 (Malmstrom 2012), whereas the other did not define an older subgroup and presented all data together (Roa 2004). As the mean age of participants in the latter study was about 72 years with a standard deviation of about five years, the majority of participants in this study would have been over 65 years of age, but the exact proportion of the sample that this represents was unclear (see Risk of bias in included studies).

Gender

Most studies had participant gender ratios of about 3 to 2 in favour of male participants; however, in two studies, the proportion of men and women was roughly equal (Roa 2015; Wick 2012).

Performance status

Most studies required that participants had a certain performance status prior to enrolment and did not recruit participants who were not self‐caring. Thus, participants of six studies specified Karnofsky performance scores (KPS) of 60 or more (ARTE 2018; Wick 2012), or 70 or more (GLARIUS 2016; Green 1983; Keime‐Guibert 2007; Stupp 2017a). Two studies specified an Eastern Cooperative Oncology Group (ECOG) performance status of 0 to 2 (Perry 2017; Weller 2017), and two specified a World Health Organization (WHO) performance status of 0 to 2 (AVAglio 2014; Malmstrom 2012). Two studies, however, recruited participants with poorer performance status (minimum KPS of 50) (Roa 2004; Roa 2015), patients with a KPS of 50 require considerable assistance and frequent medical care (Table 3).

Table 3. Performance scores

Karnofsky Status	Karnofsky Grade	ECOG Grade	ECOG Status
Normal, no complaints	100	0	Fully active, able to carry on all pre‐disease performance without restriction
Able to carry on normal activities. Minor signs or symptoms of disease	90	1	Restricted in physically strenuous activity but ambulatory and able to carry out work of a light or sedentary nature, e.g. light house work, office work
Normal activity with effort	80	1	Restricted in physically strenuous activity but ambulatory and able to carry out work of a light or sedentary nature, e.g. light house work, office work
Care for self. Unable to carry on normal activity or to do active work	70	2	Ambulatory and capable of all self‐care but unable to carry out any work activities. Up and about more than 50% of waking hours
Requires occasional assistance, but able to care for most of his needs	60	2	Ambulatory and capable of all self‐care but unable to carry out any work activities. Up and about more than 50% of waking hours
Requires considerable assistance and frequent medical care	50	3	Capable of only limited self‐care, confined to bed or chair more than 50% of waking hours
Disabled. Requires special care and assistance	40	3	Capable of only limited self‐care, confined to bed or chair more than 50% of waking hours
Severly disabled. Hospitalisation indicated though death non imminent	30	4	Completely disabled. Cannot carry on any self‐care. Totally confined to bed or chair
Very sick. Hospitalisation necessary. Active supportive treatment necessary	20	4	Completely disabled. Cannot carry on any self‐care. Totally confined to bed or chair
Moribund	10	4	Completely disabled. Cannot carry on any self‐care. Totally confined to bed or chair
Dead	0	5	Dead

As published in Am J Clin. Oncol: Oken 1982

O6‐methylguanine DNA methyltransferase (MGMT)‐methylation status

Eight studies reported the MGMT‐methylation status of their participants (ARTE 2018; AVAglio 2014; GLARIUS 2016; Malmstrom 2012; Perry 2017; Stupp 2017a; Weller 2017;Wick 2012). In the overall samples, the MGMT‐methlyated status was represented by at least 21% of participants with these test results: in ARTE 2018 26%, in AVAglio 2014 47%, in Malmstrom 2012 46.6%, in Perry 2017 37%, in Stupp 2017a 34%, in Weller 2017, and 20% in Wick 2012. In GLARIUS 2016, all participants had MGMT‐unmethylated glioblastomas. For two studies in which the elderly were a subgroup (Malmstrom 2012; Weller 2017), MGMT‐methylation status was reported for the broader sample and might not necessarily have reflected the MGMT‐methylation status of the elderly subgroup relevant to this review.

Interventions and comparisons

Most studies (10) randomised participants to two treatment arms but one trial (Malmstrom 2012) had three treatment arms and one had four treatment arms (Green 1983). The majority of treatments offered to patients were either radiotherapy alone, systemic anti‐cancer treatment (SACT) alone, or a combination of both. One trial (Keime‐Guibert 2007) had a standard management arm of supportive care, and one RCT used a medical device, known as tumour treating fields (TTF), in combination with radiotherapy and TMZ (Stupp 2017a) in its experimental arm. All studies randomised participants after diagnosis and before radiotherapy and/or chemotherapy with the exception of Stupp 2017a and Weller 2017, both of which randomised participants after chemoradiotherapy and before commencement on adjuvant TMZ.

In the trials that included elderly patients only (ARTE 2018; Keime‐Guibert 2007; Perry 2017; Roa 2004; Roa 2015; Wick 2012), the reference treatment arms were radiotherapy alone (60 Gy in 30 fractions (Roa 2004; Wick 2012), or 40 Gy in 15 fractions (ARTE 2018; Perry 2017; Roa 2015)), or supportive care (Keime‐Guibert 2007). The experimental arms in these trials were hypofractionated radiotherapy treatment alone (40 Gy in 15 fractions (Roa 2004), or 25 Gy in five fractions (Roa 2015), radiotherapy (40 Gy in 15 fractions combined with a systematic anti‐cancer treatment TMZ (Perry 2017) or bevacizumab (ARTE 2018), radiotherapy (50 Gy in 28 fractions with supportive care (Keime‐Guibert 2007), or TMZ alone (Wick 2012).

In the six trials (AVAglio 2014; GLARIUS 2016; Green 1983; Malmstrom 2012; Stupp 2017a; Weller 2017) that included patients of all ages, with the elderly as a subgroup, the radiotherapy fractionation in the reference treatment was exclusively 60 Gy in 30 fractions. In the reference arms of the trials, this was used alone (Malmstrom 2012), in combination with concomitant and adjuvant TMZ +/‐ placebo (AVAglio 2014; GLARIUS 2016; Stupp 2017a; Weller 2017), or in combination with intravenous carmustine (BCNU) (Green 1983). The experimental arms in these trials were mostly 60 Gy of radiotherapy in 30 fractions in combination with additional or alternative SACTs (AVAglio 2014; GLARIUS 2016; Green 1983; Weller 2017). Malmstrom 2012 was the only trial that included an experimental treatment arm with TMZ alone and hypofractionated radiotherapy alone (34 Gy in six fractions over two weeks). Stupp 2017a was the only trial to include a medical device (TTF) and they used this device in their experimental arm in combination with adjuvant TMZ following chemoradiotherapy (60 Gy of radiotherapy plus TMZ).

Radiotherapy fractionation and delivery

All of the included RCTs, except one (Green 1983) used megavoltage (MV) photon radiotherapy to the tumour or tumour bed with a 2 cm to 3 cm margin. Green 1983 used whole brain radiotherapy (WBRT), which is likely a reflection of the more limited technological capabilities to deliver conformal radiotherapy in the 1970s when this trial was open to recruitment. A comparison of the biologically effective dose (BED) and E2D2 of radiotherapy fractionation schedules used across all trials is outlined in Table 1.

Systemic anti‐cancer treatment

TMZ was the most frequently used SACT. When combined with 60 Gy in 30 fractions, it was used as per the "Stupp" regimen (Stupp 2005). This comprises 75 mg/m2 of TMZ given concomitantly with six weeks of radiotherapy, followed by adjuvant treatment delivered over five days each month at a dose of 150 mg/m2 to 200 mg/m2. In the original Stupp regimen (Stupp 2005), adjuvant treatment was continued for a total of six cycles. AVAglio 2014; GLARIUS 2016; Stupp 2017a all followed this regimen. Weller 2017 specified that adjuvant TMZ could be continued for six to 12 cycles or longer, and Perry 2017 specified that up to 12 adjuvant cycles of TMZ could be delivered. When TMZ was used alone (Wick 2012, Malmstrom 2012), this was given orally using a week‐on/week‐off schedule of 100 mg/m2/day for up to six months of treatment (Wick 2012) or 200 mg/m2 on days one to five of every 28 days for up to six cycles (Malmstrom 2012).

Bevacizumab was used in the experimental arm of three trials (ARTE 2018, AVAglio 2014, GLARIUS 2016) and was delivered intravenously at a dose of 10 mg/kg every two weeks in all of these trials. Irinotecan was used in combination with bevacizumab and radiotherapy in the GLARIUS 2016 trial and delivered intravenously at a dose of 125 mg/m2 every two weeks.

Green 1983 combined WBRT with intravenous BCNU (80 mg/m2/day on three successive days every eight weeks) in their reference arm. In the first experimental arm of this trial, BCNU was replaced by high‐dose oral methylprednisolone (400 mg/m2/day in three divided doses for seven days) in four weekly cycles, and their second experimental arm combined BCNU and high‐dose methylprednisolone. The last experimental arm in the trial by Green 1983, combined WBRT with procarbazine which was given orally at a total dose of 150 mg/m2/day in three or four divided doses for 28 consecutive days every eight weeks.

Finally, Weller 2017 used rindopepimut (500 ug) admixed with 150 ug granulocyte‐macrophage colony‐stimulating factor (GM‐CSF), given via monthly intradermal injection in their experimental arm and 100 ug keyhole limpet haemocyanin in their control arm, both in combination with standard oral TMZ (150 mg/m2 to 200 mg/m2 for 5 of 28 days) for six to 12 months or longer. In this trial, all patients had completed standard chemoradiation with 60 Gy of radiotherapy and concomitant TMZ prior to commencing treatment in either the experimental or control treatment arm.

Other treatments

Carmustine wafers

Although not randomised between arms, Stupp 2017a was the only trial that specified that treatment with implanted carmustine wafers was permitted for patients on either arm of their trial. The proportion of patients receiving carmustine wafers in each arm was not reported.

Medical devices

Stupp 2017a was the only trial to use a medical device. Tumour treating fields (TTF) comprise an external medical device that is worn by the patient. It consists of four transducer arrays which are connected to a portable device. These arrays are applied to the patients’ shaved scalp via nine electrodes and emit low‐intensity, intermediate frequency (200kHz) alternating electric fields to the brain. The patient carries the device in a backpack and is encouraged to wear the device for at least 18 hours per day. Determining the layouts of the transducer is performed using a TTF mapping software system. Patients and their families are trained on how to use the device and how to trouble shoot problems with the device by nursing staff and a device technician. The patients must replace the transducer arrays twice weekly and the treatment is delivered on an outpatient basis. Participants in the Stupp 2017a trial were randomised after the completion of chemoradiation and, therefore, the TTF treatment was given in the adjuvant setting only in combination with TMZ, and not given concomitantly with radiotherapy. TTF treatment was to be initiated at least four weeks but not more than seven weeks from the last day of radiotherapy.

Supportive care

Keime‐Guibert 2007 was the only trial to include supportive care as a management option. This was used alone in the reference arm and in combination with radiotherapy (50 Gy in 28 fractions) in the experimental arm. Supportive care was defined as any mixture of treatment with corticosteroids and anti‐epileptics, as well as physical and psychological support and management by a palliative care team. There was no information on the timing of when referral to the palliative care team was made.

Outcomes and follow‐up

Table 4 outlines both the primary and secondary outcomes from each of the included trials that were of interest for the purposes of this review, along with the evaluation tools used to assess each outcome. All of the included trials, except Green 1983, reported overall survival outcomes using time‐to‐event analysis and provided median overall survival for either all participants (ARTE 2018; Keime‐Guibert 2007; Malmstrom 2012; Perry 2017; Roa 2004; Wick 2012), or for a subgroup of patients if the trial was not restricted to recruiting elderly patients (AVAglio 2014; GLARIUS 2016; Roa 2015; Stupp 2017a; Weller 2017). Although reporting median survival, several trials did not provide a hazard ratio to show the difference between survival for elderly patients in different treatment arms (ARTE 2018; GLARIUS 2016; Roa 2015). Median survival data, where reported, are tabulated in Table 5.

Table 4. Table of outcomes reported in included studies

	Trial	OS	QoL	PFS	Severe adverse events	Cognitive impairments	Functional impairment	Fatigue
1	Green 1983	For 65+ age subgroup, number of deaths (overall, no time point specified) and death rate (number of deaths per 10 patient‐months) reported.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.
2	Roa 2004 (Elderly patients only in this trial, defined as age 60 years or over; mean age was 72 years with SD 5 years)	Median OS reported (ITT analysis), including HR and a KM curve. Percentage of patients alive at 6 months also reported (Table 1 and Figure 1 of main manuscript).	Low rates of FACT‐Br version 3 completion (45% overall) by patients precluded meaningful analysis. Protocol specified FACT‐Br completion at baseline, 3 weeks after starting RT, at the end of RT, and at 3‐month intervals thereafter. Table 2 of main manuscript.	Not done.	Not reported.	Not reported.	Difference in averaged KPS scores and change in KPS over time between the two groups (0‐6 months from start of RT). KPS at baseline, 3 weeks, 6 weeks, first and second follow up. Table 1, Table 2 and Figure 2 of main manuscript.	Not reported.
3	Keime‐Guibert 2007 (Elderly patients only ‐ aged 70 years or over).	Median OS reported (ITT analysis), including HR and KM curve.	QLQ‐C30 and QLQ‐BN20 and completion rate. Changes in mean score at baseline, day 30, day 60, day 90 and day 135.	Median PFS.	Tolerance of treatment reported but not clear which toxicity scoring used.	QLQ‐C30 includes cognitive. MDRS score at baseline, 60 days and 135 days. MMSE score. NPI (Neuropysch inventory).	KPS decline over time.	QLQ‐C30 includes fatigue.
4	Malmstrom 2012 (Patients in this trial all aged over 60 years old but reported data for 70+ years subgroup)	Median OS and 1 year survival percentage. As this study had three arms, survival analyses were done using three pairwise comparisons. A KM curve was presented for the overall patient population and a subgroup of patients aged 60‐70 years and aged over 70. HR reported for this outcome (Table 2). OS for patients with mMGMT and umMGMT disease also reported. The two radiotherapy arms combined for this analysis. In the supplementary appendix, OS and KM reported comparing the TMZ arm and the hypofractionated RT arm. These outcomes were reported for patients overall, a subgroup of patients aged 60‐70 and a subgroup of patients aged over 70. HR reported for these outcomes. 1 year OS probabilities also reported.	EORTC QLQ‐30v3 and QLQ‐BN20 at baseline, 6 weeks, 3 months and 6 months. Mean change of score from baseline reported for each domain (Figure 4). No data reported for 6 month time‐point due to low completion rate.	Not reported.	WHO grading for all except nausea and vomiting (CTCAE v2), but only reported for overall sample and not elderly subgroup.	Reported as a category of a QoL assessment but only for overall sample and not elderly subgroup.	Not reported.	WHO grading system. Reported fatigue G2‐5, but only for overall sample and not elderly subgroup.
5	Wick 2012 (All elderly patients – aged over 65 years)	Median OS reported (ITT analysis) with HR and KM curve. OS differences between patients with mMGMT and uMGMT disease also reported (with HR and KM curve). OS survival (PP analysis) also reported but no HR or KM curve for these outcomes. Survival % at 6 months and 1 year. Also report proportions for death, or disease progression or death for each arm.	EORTC QLQ‐30 and BN20 used for HRQoL assessment. Results available for 82% of the patients (in appendix).	Median event free survival (rather than PFS) and also mMGMT and uMGMT. %EFS at 6 months and 1 year.	CTCAEv3 used to collect adverse event data.	MMSE performed at baseline, monthly during treatment and then every 3 months. Median score (with 95% CI) reported overall, before treatment and after primary treatment.	KPS at baseline.	Fatigue/asthenia G3‐4 reported as an adverse event.
6	AVAglio 2014 (comprises several reports) Chinot 2014 main report presents findings for whole sample, elderly subgroup reported in supplementary appendix. Saran 2016 provides information on safety outcomes. Taphoorn 2015 reported HR QoL outcomes for sample as a whole.	OS for 60‐69 and 70+ subgroups (found in supplementary appendix 2). No KM curves for these subgroups.	Taphoorn 2015 reports the results for QLQ‐C30 and QLQ‐BN20 for the whole sample (age >18), not for the elderly subgroup.	PFS for 60‐69 and 70+ subgroups (number and HR).	Mason 2014 is a separate publication about arterial adverse events and gives event rates for the over 65 subgroup.	Not for elderly subgroup.	Not for elderly subgroup.	Not for elderly subgroup.
7	Roa 2015 (comprises several reports) (Guedes de Castro 2017 reports on outcomes only for those patients aged over 65s and Fidarova 2015 presents quality of life results as a conference abstract).	Median OS reported (ITT analysis). Outcomes for elderly and frail, and elderly and not frail subgroups. No HR reported for OS outcomes. KM curves included in Guedes de Castro 2017 Figures 1‐4.	The mean scores from a combination of EORTC QLQC30 and QLQ‐BN20 reported at baseline, 4 weeks and 8 weeks. Categorical scales were transformed to linear scalers from 0‐100. Difference between mean scores between treatment groups reported in Guedes de Castro 2017 Table 3.	Median PFS reported (ITT). No KM or HR for elderly only patients.	Adverse events were recorded weekly during RT, 4 weeks after RT and every 3 months until tumour progression using CTCAE v3. AEs only reported in the case of a grade 3 or greater toxicity.	MMSE at baseline.	KPS at baseline.	Fatigue mentioned in Fidarova 2015 abstract for overall trial population.
8	GLARIUS 2016 (comprises several reports) Herrlinger 2016 reports the main findings for the whole sample, aged from 18+ years. Kebir 2016 is an abstract which reports specifically on the differences in OS between younger and older patients in the trial.	Median OS for 65+ subgroup in both arms reported (modified ITT analysis). No HR or KM curve available for the elderly cohort. yes	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.
9	Perry 2017 (All elderly patients – aged 65 years old or over).	Median OS reported (ITT). KM curve and HR reported for this outcome. Median OS also reported for subgroups of patients aged 65‐70 years old, 71‐75 and 76+. Also reported OS rate at 12, 18 and 24 months for all patients and for patients with umMGMT and mMGMT disease.	QLQC30 and BN20. Up to 18 months post treatment. Baseline/1 week/3 week reported. Time to deterioration in QoL reported.	Median PFS.	CTCAE v3 used for adverse event reporting.	MMSE at baseline.	ECOG at baseline.	Not separately reported.
10	Weller 2017 (Trial not restricted to elderly patients only. Includes patients aged 18 years +).	Report OS (events per patient) for patients aged 65+ subgroup. The subgroup if further divided into those who have minimal residual disease (MRD) and significant residual disease (SRD). There is a HR given for patients aged 65+ in the MRD subgroup and in the SRD subgroup separately. There is no HR reported for the 65+ group overall. No KM curves for the elderly subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.
11	Stupp 2017a (Trial not restricted to elderly patients only. Includes patients aged 18 years +). Taphoorn 2018 is separate publication which reports on HRQoL outcomes.	Median OS for 65+ subgroup. HR and KM curve reported for this outcome in the 65+ subgroup. The proportion of patients in each arm of the trial who were alive at the end of the study also reported.	HR QoL was measured using EORTC QLQ‐C30 and BN20 questionnaires at baseline and every 3 months for up to 12 months. Mean QoL scores and mean change from baseline reported. Outcomes not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.
12	ARTE 2018 (All elderly patients in the trial – aged 65 years or over)	Median OS (ITT and PP) and 1 year survival rate. Also reported OS depending on molecular panel subtype. No HR reported for median OS differences. KM curves are shown.	Reported median deterioration free survival from baseline. Individual functional and symptom scores from EORTC QLQ‐C30/BN20 before tumour progression analysed in a generalised linear model that controlled for time treatment interactions.	Median PFS. ITT and PP. And %PFS at 6 months. Also had independent central review for some patients. (66 with MRI available). Also looked at PFS per molecular panel subtype.	Yes, reported in supplementary material. G3‐5 fatigue, seizures, headaches, other neuro, neuropysch, haematological, arterial hypertension, thromboembolic, all infections, cutaneous, GI.	MMSE at baseline and serial measurements (0,7,19 weeks).	KPS at baseline.	Fatigue reported as a category of QoL and as an adverse events.

CTCAE: Common Terminology Criteria for Adverse Events; EORTC: European Organisation for Research and Treatment of Cancer (EORTC) QLQ‐C30 or QLQ‐BN20 (specific for brain cancer); FACT‐Br: Functional Assessment of Cancer Therapy scale [specific for brain cancer]); HR: hazard ratio; HRQoL: Health‐related quality of life; ITT: Intention to treat; mITT: modified ITT; KM curve: Kaplan Meier curve; KPS: Karnofsky performance status; MMSE: mini mental state examination; mMGMT: Methylated MGMT; umMGMT: unmethylated MGMT; PFS: progression free survival; PP: per protocol; OS: overall survival.

Table 5. Median survival associated with treatment options evaluated in included studies

Study ID	Age (years) and performance status	Treatment arm	Overall Survival (OS)			Progression Free Survival (PFS)
Study ID	Age (years) and performance status	Treatment arm	Median OS (months)	lower 95% CI (months)	upper 95% CI (months)	Median PFS (months)	lower 95% CI (months)	upper 95% CI (months)
Green 1983	Age ≥ 65 KPS ≥ 70%	4 arms; see Table 1	‐	‐	‐	‐	‐	‐
Keime‐Guibert 2007	Age ≥ 70 KPS ≥ 70%	Supportive care	3.9	3.1	5.0	1.3	1.0	1.8
Keime‐Guibert 2007	Age ≥ 70 KPS ≥ 70%	RT (50 Gy/28 fractions/5‐6 weeks) + supportive care	6.8	5.9	8.1	3.5	2.5	5.2
Roa 2004	Age ≥ 60 KPS ≥ 50%	RT (40 Gy/15 fractions/3 weeks)	5.6	‐	‐	‐	‐	‐
Roa 2004	Age ≥ 60 KPS ≥ 50%	RT (60 Gy/30 fractions/6 weeks)	5.1	‐	‐	‐	‐	‐
Roa 2015^a (elderly and/or frail)	Age ≥ 65 KPS 50‐70%	RT (25 Gy/5 fractions/1 week)	6.8	4.5	9.1	4.3	2.6	5.9
Roa 2015^a (elderly and/or frail)	Age ≥ 65 KPS 50‐70%	RT (40 Gy/15 fractions)	6.2	4.7	7.7	3.2	0.1	6.3
Roa 2015 (elderly and frail)	Age ≥ 65 KPS ≥ 50%	RT (25 Gy/5 fractions/1 week)	7.5	5.3	9.7	‐	‐	‐
Roa 2015 (elderly and frail)	Age ≥ 65 KPS ≥ 50%	RT (40 Gy/15 fractions)	6.7	4.5	8.9	‐	‐	‐
Roa 2015^a(elderly and non‐frail)	Age ≥ 65 KPS ≥ 80%	RT (25 Gy/5 fractions/1 week)	8.0	5.9	10.0	‐	‐	‐
Roa 2015^a(elderly and non‐frail)	Age ≥ 65 KPS ≥ 80%	RT (40 Gy/15 fractions/3 weeks)	8.0	5.3	10.3	‐	‐	‐
Malmstrom 2012	Age ≥ 70 WHO 0‐2	RT (60 Gy/30 fractions/6 weeks)	5.2	4.0	6.3	‐	‐	‐
		RT (34 Gy/10 fractions/2 weeks)	7.0	5.2	8.8	‐	‐	‐
		TMZ	9.0	6.2	11.8
Malmstrom 2012 (unmethylated)	Age ≥ 70 WHO 0‐2	RT (any schedule)	7.0	5.8	8.3	‐	‐	‐
Malmstrom 2012 (unmethylated)	Age ≥ 70 WHO 0‐2	TMZ	6.8	5.9	7.7	‐	‐	‐
Malmstrom 2012 (methylated)	Age ≥ 70 WHO 0‐2	RT (any schedule)	8.2	6.6	9.9	‐	‐	‐
Malmstrom 2012 (methylated)	Age ≥ 70 WHO 0‐2	TMZ	9.7	8.0	11.4	‐	‐	‐
Wick 2012^b	Age ≥ 65 KPS ≥ 60%	RT (60Gy/30 fractions/6 weeks)	9.6	8.2	10.8	4.7	4.2	5.2
Wick 2012^b	Age ≥ 65 KPS ≥ 60%	TMZ	8.6	7.3	10.2	3.3	3.2	4.1
Wick 2012 (unmethylated)	Age ≥ 65 KPS ≥ 60%	RT (60Gy/30 fractions/6 weeks)	10.4	8.0	11.6	4.6	3.7	6.3
Wick 2012 (unmethylated)	Age ≥ 65 KPS ≥ 60%	TMZ	7.0	5.7	8.7	3.3	3.0	3.5
Wick 2012 (methylated)	Age ≥ 65 KPS ≥ 60%	RT (60Gy/30 fractions/6 weeks)	9.6	6.4	not reached	4.6	4.2	5.0
Wick 2012 (methylated)	Age ≥ 65 KPS ≥ 60%	TMZ	not reached	10.1	not reached	8.4	5.5	11.7
Perry 2017	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks)	7.6	7.0	8.4	3.9	3.5	4.3
Perry 2017	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks) + TMZ + maintenance TMZ	9.3	8.3	10.3	5.3	4.6	6.2
Perry 2017^c (unmethylated)	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks)	7.9	‐	‐	‐	‐	‐
Perry 2017^c (unmethylated)	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/ 3 weeks) + TMZ + maintenance TMZ	10	‐	‐	‐	‐	‐
Perry 2017^c (methylated)	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks)	7.7	‐	‐	‐	‐	‐
Perry 2017^c (methylated)	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks) + TMZ+ maintenance TMZ	13.5	‐	‐	‐	‐	‐
AVAglio 2014	Age ≥ 70 WHO 0‐2	RT (60 Gy/30 fractions) + TMZ + maintenance TMZ	‐	‐	‐	‐	‐	‐
AVAglio 2014	Age ≥ 70 WHO 0‐2	RT (60 Gy/30 fractions) + TMZ + BEV + maintenance	‐	‐	‐	‐	‐	‐
Stupp 2017a^d	Age ≥ 65 KPS ≥ 70%	RT (±60 Gy/30 fractions/6 weeks) + TMZ + maintenance TMZ	13.7	7.6	24.8	‐	‐	‐
Stupp 2017a^d	Age ≥ 65 KPS ≥ 70%	RT (±60 Gy/30 fractions/6 weeks) + TMZ + maintenance TMZ + TTFields	17.4	9.0	31.5	‐	‐	‐
ARTE 2018	Age ≥ 65 KPS ≥ 60%	RT (40 Gy/15 fractions/3 weeks)	12.2	9.2	15.2	4.8	3.0	6.6
ARTE 2018	Age ≥ 65 KPS ≥ 60%	RT (40 Gy/15 fractions/3 weeks) + BEV	12.1	10.2	14.0	7.6	6.2	9.0
GLARIUS 2016^e (unmethylated)	Age ≥ 65 KPS ≥ 70%	RT (60 Gy/30 fractions/6 weeks) + TMZ + maintenance	17.5	‐	‐	‐	‐	‐
GLARIUS 2016^e (unmethylated)	Age ≥ 65 KPS ≥ 70%	RT (60 Gy/30 fractions/6 weeks) + BEV + IRI + maintenance	13.4	‐	‐	‐	‐	‐
Weller 2017	Age ≥ 65 KPS ≥ 60%	RT (60 Gy/30 fractions/6 weeks) + TMZ + maintenance	‐	‐	‐	‐	‐	‐
Weller 2017	Age ≥ 65 KPS ≥ 60%	RT (60 Gy/30 fractions/6 weeks) + TMZ + maintenance + RIN	‐	‐	‐	‐	‐	‐

BEV: bevacizumab; ECOG: Eastern Cooperative Oncology Group; Gy: Grays; IRI: irinotecan; KPS: Karnofsky performance score; OS: overall survival; PFS: progression free survival; RIN: rindopepimut; RT: radiotherapy; TTF: tumour treating fields; TMZ: temozolomide; WHO: World Health Organization

^aData from the Guedes de Castro et al 2017 substudy report on patients ≥ 65 years only

^bThis study reported event free survival (EFS), not PFS. Findings by MGMT methylation status were reported by authors in the later (2017) publication.

^cFrom Perry 2012 abstract

^dMGMT promotor methylation status was not reported separately for the elderly subgroup. Median survival data were reported as time from randomisation not diagnosis. Randomisation in this trial occurred after concomitant chemoradiotherapy.

^eFrom substudy data reported in a conference proceeding abstract by Kebir et al, 2016.

The proportion of patients alive at six months (Roa 2004, Wick 2012), 12 months (ARTE 2018; Malmstrom 2012; Perry 2017; Wick 2012), 18 months (Perry 2017) and 24 months (Perry 2017) was also used to report survival outcomes. Stupp 2017a and Green 1983 reported the proportion of patients who had died by the end of the study period.

The second main outcome of interest was health‐related quality of life (HRQoL). The most common tools used to collect HRQoL data were patient‐completed questionnaires, specifically the European Organisation for the Research and Treatment of Cancer (EORTC) QLQ‐C30 and QLQ‐BN20 questionnaires. The results of HRQoL outcomes for elderly patients using these questionnaires were reported for five trials (ARTE 2018; Keime‐Guibert 2007; Malmstrom 2012; Perry 2017; Wick 2012). Roa 2004 was the only trial to use the Functional Assessment of Cancer Therapy‐Brain (FACT‐Br) questionnaire but the results of using this tool were not reported due to a low completion rate. Several of the studies that included both younger and elderly patients did record HRQoL, but did not report results of these assessments for elderly subgroups separately (AVAglio 2014, GLARIUS 2016, Green 1983, Stupp 2017a, Weller 2017).

Follow‐up times were varied and were often not documented in the trial publications. For those trials that did specify their follow‐up time, AVAglio 2014 had follow‐up for at least 17 months (with the end of study at 64 months after opening) and Stupp 2017a specified a median follow‐up of 40 months (interquartile range, 34 to 66 months) with a minimum follow‐up of 24 months. Only two patients in the over‐65 age group were alive at 60 months of follow‐up. Wick 2012 had a minimum follow‐up of 12 months (median 25.2 months (range 20.0 to not reached)). Roa 2015 specified that all patients were followed up until death.

Excluded studies

In selecting studies for evaluation of treatment effectiveness, excluded studies numbered 145 records. Studies were excluded mainly for the following reasons.

Ineligible study design, e.g. non‐randomised trial; editorial: Bent 2009; Blumenthal 2018; Boisen 2018; Boxerman 2013; Catterall 1980; Chamberlain 2005; Chong 2018; Cohen 2005; Corn 1994; Das 2017; Dherijha 2018; Espana 1978; Halperin 1993; Jeremic 1999; Koc 2008; Lamers 2008; Lorimer 2016; McCarthy 2017; Napolitano 1999; Pinzi 2017; Reyes‐Botero 2018; Soffietti 2017; Solth 2018; Stupp 2002; Vellayappan 2017 (25 studies)

Ineligible study population, e.g. not elderly participants; not newly diagnosed glioblastoma: Ali 2018; Armstrong 2013; Athanassiou 2005; Balana 2016; Bampoe 2000; Batchelor 2013; Beije 2015; Bhandari 2013; Bhandari 2017; Bleehen 1981; Bleehen 1991; Blumenthal 2015; Bogdahn 2011; Boiardi 1992; Bower 1997; Brandes 2016; Brisman 1976; Brown 2016; Buckner 2001; Buckner 2006; Carpentier 2017; Castro 1997; Chang 1983; Chauffer 2014; Cianfriglia 1980; Clarke 2009; Combs 2008; Curran 1992; Deutsch 1989; Dinapoli 1993; Du 2018; Duncan 1986; Elinzano 2018; Eljamel 2008; Elliott 1997; Eyre 1983; Farkkila 1994; Field 2015; Field 2017; Fischer 1985; Fulton 1984; Gaber 2013; Gilbert 2013; Glinski 1993; Grossman 2003; Halperin 1996; Harada 1996; Hatlevoll 1985; Henriksson 2006; Hiesiger 1995; Hildebrand 1994; Hitchon 1999; Hofland 2014; Imbesi 2006; Iwadate 1993; Karacetin 2011; Kim 2011; Knerich 1990; Kocher 2008; Kochii 2000; Kong 2017; Lanzetta 2003; Lee 2015; Lenartz 2000; Levin 1979; Levin 2000; Levin 2006; Lissoni 1993; Ludgate 1988; Mallick 2018; Mao 2015; Marshall 2006; Montemor 2008; MRC 1983; Nabors 2015; Nelson 1988; Payne 1982; Peszynski 1988; Phillips 2003; Prados 2001; Reagan 1976; Shapiro 1976; Shapiro 1989; Shapiro 1992; Sharma 2003; Simpson 1976; Sneed 1998; Socha 2016; Solero 1979; Solomon 2013; Souhami 2004; Stadler 1984; Stupp 2005; Stupp 2009; Stupp 2014; Stupp 2015; Szczepanek 2013; Takakura 1986; Taphoorn 2005; Urtasun 1982; Ushio 1985; Wakabayashi 2018; Wang 2008; Weller 2003; Werner‐Wasik 1996; Westphal 2003; Westphal 2006; Westphal 2015; Wick 2009; Wick 2016; Yang 2018; Zhu 2017 (112 studies).

Other reasons were insufficient information (three studies: Felzmann 2013; Felzmann 2014; Muragaki 2017) and a different study objective (five studies:Stragliotto 2013; Stummer 2006; Stummer 2011; Stummer 2017; Westphal 2013).

Risk of bias in included studies

All included studies were RCTs and the trial quality was generally high, with most studies assessed as having a low risk of bias overall Figure 3. For the individual study risk of bias explanations, please refer to the Characteristics of included studies tables.

Figure 3

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.

Allocation

The method of randomisation was described for all of the studies; therefore, all studies were at low risk of bias for the sequence generation criterion. Regarding concealment allocation at the participant selection stage, sixr studies were assessed as being at low risk of bias for this criterion (Green 1983; Keime‐Guibert 2007; Roa 2004; Stupp 2017a; Weller 2017; Wick 2012). However, allocation concealment was not clearly described in the other studies, which were assessed as having unclear risk of bias for this criterion.

Blinding

Most studies were open‐label studies, therefore, were potentially at a high risk of bias for blinding, and most did not describe assessor blinding. However, two studies was double‐blinded (AVAglio 2014; Weller 2017) and, therefore, were assessed as having a low risk of bias for this criterion.

Incomplete outcome data

For the primary outcome, most included studies had good follow‐up with low dropout rates. One study (Wick 2012) that compared radiotherapy with temozolomide had relatively higher dropout rates in the radiotherapy arm (14% versus 5%), which might have influenced the findings. This study was assessed as having an unclear risk of bias for this domain.” Weller 2017 had unclear risk for this domain.

For studies that measured quality of life in an elderly population (Perry 2017; Roa 2004; Roa 2015), attrition was a problem that had a major impact on the quality of these findings. We therefore considered these studies to be at high risk of attrition bias for quality of life findings.

Selective reporting

All studies reported pre‐specified outcomes and were considered to be at a low risk of bias for this criterion.

Other potential sources of bias

In the context of the review evidence, Stupp 2017a represented a high risk of bias due to the timing of randomisation in this trial. Randomisation was performed for a select group of patients who had completed concomitant chemoradiotherapy without progressive disease as those who died during chemoradiotherapy or who had severe early toxicities would have dropped out by the time of randomisation. Consequently, we decided not to include this trial in the quantitative synthesis due to concern over networks transitivity. .

Similarly, Weller 2017 randomised participants after concomitant chemoradiotherapy and its findings would have downgraded for indirectness; however, this trial contributed no data to meta‐analyses. All patients in the GLARIUS 2016 trial had MGMT unmethylated glioblastoma , which is associated with a shorter survival time than MGMT‐methylated tumours. Whilst this could bias the findings of review meta‐analyses, this trial contributed no data to pooled analyses. REview authors had no other serious risk of bias concerns, although in some studies the risk of bias due to protocol deviations was assessed as unclear.

Quality of economic studies

The quality of the trial on which Ghosh 2018 is based has been discussed in the previous section (Roa 2015). The study was found to have a low risk of bias (with the exception of the blinding, which is open‐label). The Consolidated health Economic Evaluation Reporting Standards (CHEERs) checklist (Husereau 2013) and the Consensus on Health Economic Criteria (CHEC (checklist (Evers 2005) were applied to the study to assess the quality of economic evaluation as recommended but the current guidelines (Shemilt 2018). The results can be found in Table 6 and Table 7. The results of the CHEERS reporting checklist show that a number of parameters are not reported (e.g. sources of costs, time horizon, perspective). The results of the CHEC checklist show that there a number of issues with the methodological quality of the study, including inappropriate costing and analysis methods.

Table 6. CHEC list* for included economic studies

CHEC ITEM	Ghosh et al (2018)
Is the study population clearly described?	Y
Are competing alternatives clearly described?	Y
Is a well‐defined research question posed in answerable form?	Y
Is the economic study design appropriate to the stated objective?	Y
Is the chosen time horizon appropriate to include relevant costs and consequences?	N
Is the actual perspective chosen appropriate?	N
Are all important and relevant costs for each alternative identified	N
Are all costs measured appropriately in physical units?	Y
Are costs valued appropriately?	N
Are all important and relevant outcomes for each alternative identified?	N
Are all outcomes measured appropriately?	N
Are outcomes valued appropriately?	N
Is an incremental analysis of costs and outcomes of alternatives performed?	Y
Are all future costs and outcomes discounted appropriately?	N
Are all important variables, whose values are uncertain, appropriately subjected to sensitivity analysis?	N
Do the conclusions follow from the data reported?	N
Does the study discuss the generalizability of the results to other settings and patient/client groups?	N
Does the article indicate that there is no potential conflict of interest of study researcher(s) and funder(s)?	Y
Are ethical and distributional issues discussed appropriately?	N

*Evers 2005

Table 7. CHEERS checklist* of included studies

CHEERS Quality Checklist
Section of paper	Component	Reported on Page Number
Ghosh et al (2018)
Title and abstract	Identify the study as an economic evaluation or use more specific terms such as ‘‘cost‐effectiveness analysis’’, and describe the interventions compared.	114
Title and abstract	Provide a structured summary of objectives, perspective, setting, methods (including study design and inputs), results (including base case and uncertainty analyses), and conclusions.	114
Methods	Describe characteristics of the base case population and subgroups analysed, including why they were chosen.	Supplementary material
	State relevant aspects of the system(s) in which the decision(s) need(s) to be made.	Not reported
	Describe the perspective of the study and relate this to the costs being evaluated.	Not reported
	Describe the interventions or strategies being compared and state why they were chosen.	114‐115
	State the time horizon(s) over which costs and consequences are being evaluated and say why appropriate.	Partially, stated but not justified
	Report the choice of discount rate(s) used for costs and outcomes and say why appropriate.	Not reported
	Describe what outcomes were used as the measure(s) of benefit in the evaluation and their relevance for the type of analysis performed.	115
	Describe fully the design features of the single effectiveness study and why the single study was a sufficient source of clinical effectiveness data.	115‐116
	Describe approaches used to estimate resource use associated with the alternative interventions. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs.	Not reported
	Report the dates of the estimated resource quantities and unit costs. Describe methods for adjusting estimated unit costs to the year of reported costs if necessary. Describe methods for converting costs into a common currency base and the exchange rate.	115
	Describe and give reasons for the specific type of decision‐analytical model used. Providing a figure to show model structure is strongly recommended.	N/A
	Describe all structural or other assumptions underpinning the decision‐analytical model.	N/A
	Describe all analytical methods supporting the evaluation. This could include methods for dealing with skewed, missing, or censored data; extrapolation methods; methods for pooling data; approaches to validate or make adjustments (such as half cycle corrections) to a model; and methods for handling population heterogeneity and uncertainty.	Partial, uncertainty discussed.
Results	Report the values, ranges, references, and, if used, probability distributions for all parameters. Report reasons or sources for distributions used to represent uncertainty where appropriate. Providing a table to show the input values is strongly recommended.	Not reported
	For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost‐effectiveness ratios.	116‐118 & supplementary material
	Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact of methodological assumptions (such as discount rate, study perspective).	116
	If applicable, report differences in costs, outcomes, or cost‐effectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.	N/A
Discussion	Summarise key study findings and describe how they support the conclusions reached. Discuss limitations and the generalisability of the findings and how the findings fit with current knowledge.	118‐119
Other	Describe how the study was funded and the role of the funder in the identification, design, conduct, and reporting of the analysis. Describe other non‐monetary sources of support.	119
Other	Describe any potential for conflict of interest of study contributors in accordance with journal policy. In the absence of a journal policy, we recommend authors comply with International Committee of Medical Journal Editors recommendations.	119

*Evers 2005

Effects of interventions

Results of network meta‐analysis (NMA)

Network meta‐analysis could only be performed for the primary outcome of overall survival.

Overall Survival

Seven trials contributed data to this time‐to‐event outcome and, across all studies included in the NMA, the following treatments were used either as the experimental or the comparison treatment (Figure 4).

Figure 4

Four trial arms of hypofractionated radiotherapy (40 Gy/15 fractions) (RT 40)
Five trial arms of standard radiotherapy (60 Gy/30 fractions), including one trial utilising a 50 Gy/28 fractions (RT 60)
One trial arm of supportive care only (SuppCare)
Three trial arms of temozolomide (TMZ)
Two trial arms of chemoradiotherapy, including one trial utilising 40 Gy in 15 fractions and one utilising 60 Gy in 30 fractions (CRT)
One trial arm of bevacizumab plus chemoradiotherapy (BEV CRT)
One trial arm of bevacizumab plus radiotherapy (BEV RT)

The forest plot is presented in Figure 5 and effect estimates and certainty ratings for the overall survival network can be found in Table 8. Evidence derived from the network only (i.e. where there were no common comparators) was not graded. Pooled network estimates suggested that, compared with supportive care only, any of the treatments except for bevacizumab plus radiotherapy lead to better overall survival. However, only three treatments (RT40, RT60 and TMZ) could be compared either directly or indirectly with supportive care through a common comparator. The graded evidence related to the three comparisons with direct and/or indirect evidence can be interpreted as follows.

Figure 5

Forest plot of all treatment comparisons for overall survival

Table 8. Table of estimate effects and certainty ratings for overall survival

Comparison	Direct Evidence		Indirect Evidence		Network Evidence
Comparison	HR (95% CI)	Certainty	HR (95% CI)	Certainty	HR (95% CI)	Certainty
RT60* vs Supp Care	0.47 (0.29 to 0.76)	Moderate¹	Not estimable²	–	0.47 (0.29 to 0.76)	Moderate
RT40 vs Supp Care	–	–	0.44 (0.25 to 0.77)	Low³	0.44 (0.25 to 0.77)	Low
CRT vs Supp Care	–	–	–	–	0.30 (0.17 to 0.53)	Not graded⁴
TMZ vs Supp Care	–	–	0.42 (0.25 to 0.71)	Low³	0.42 (0.25 to 0.71)	Low
BEV_CRT vs Supp Care	–	–	–	–	0.25 (0.11 to 0.54)	Not graded⁴
BEV_RT vs Supp Care	–	–	–	–	0.48 (0.23 to 1.00)	Not graded⁴

CRT vs RT40	0.67 (0.56 to 0.80)	High	–	–	0.67 (0.56 to 0.80)	High
BEV_CRT vs RT40	–	–	0.56 (0.31 to 0.99)	Moderate⁵	0.56 (0.31 to 0.99)	Moderate
TMZ vs RT40**	0.72 (0.50 to 1.05)	Low⁶	–	–	0.95 (0.71 to 1.26)	Low
BEV_RT vs RT40	1.08 (0.65 to 1.78)	Low⁶	Not estimable²	–	1.08 (0.66 to 1.78)	Low

RT40 vs RT60	0.74 (0.55 to 1.01)	Low⁶	Not estimable²	–	0.94 (0.72 to 1.23)	Low
BEV_RT vs RT60	–	–	1.01 (0.58 to 1.79)	Very low⁷	1.01 (0.58 to 1.79)	Very low
BEV_CRT vs RT60	–	–	–	–	0.52 (0.28 to 0.98)	Not graded⁴
CRT vs RT60	–	–	0.63 (0.46 to 0.87)	Low⁸	0.63 (0.46 to 0.87)	Low
TMZ vs RT60	0.86 (0.68 to 1.09)	Very low⁹	–	–	0.89 (0.71 to 1.11)	Very low

BEV_RT vs CRT	–	–	1.61 (0.95 to 2.74)	Low¹⁰	1.61 (0.95 to 2.74)	Low
BEV_CRT vs CRT	0.83 (0.48 to 1.43)	Low⁶	Not estimable²	–	0.83 (0.48 to 1.44)	Low
TMZ vs CRT	–	–	1.42 (1.01 to 1.98)	Low¹⁰	1.42 (1.01 to 1.98)	Low

BEV_RT vs TMZ	–	–	1.14 (0.64 to 2.02)	Very low^10,11	1.14 (0.64 to 2.02)	Very low
BEV_CRT vs TMZ	–	–	–	–	0.59 (0.31 to 1.12)	Not graded⁴

BEV_CRT vs BEV_RT	–	–	–	–	0.52 (0.24 to 1.10)	Not graded⁴
¹ Evidence derived from a single small study ² Could not be estimated because the intervention was not connected via a loop in the evidence network ³ Contributing direct evidence was of moderate or low certainty ⁴ There was no direct evidence for this comparison, which did not connect via a common comparator, therefore the certainty of evidence was not graded. ⁵ Contributing direct evidence was of high or moderate certainty ⁶ Downgraded for study design limitations and imprecision ⁷ Contributing direct evidence was of low certainty; network estimate imprecise ⁸ Contributing direct evidence was of high or low certainty ⁹ Downgraded for imprecision, study design limitations and inconsistency ¹⁰ Contributing direct evidence was of high or low certainty ¹¹Downgraded for imprecision

*RT50 (Keime‐Guibert 2007) coded as RT60.

**RT34 (Malmstrom 2012) coded as RT40.

Abbreviations: BEV_CRT = bevacizumab plus chemoradiotherapy; CI = confidence interval; ; CRT = chemoradiotherapy; RT40 = radiotherapy (40Gy in 15 fractions); RT60 = radiotherapy (60Gy in 30 fractions); Supp Care = supportive care; TMZ = temozolomide; TTF_AC = tumour treating fields with adjuvant chemotherapy (after concomitant chemotherapy)

RT60 probably improves overall survival time compared with supportive care only (HR 0.47, 95% CI 0.29 to 0.76; moderate‐certainty evidence)
RT40 may improve overall survival time compared with supportive care only (HR 0.44, 95% CI 0.25 to 0.77; low‐certainty evidence)
TMZ may improve overall survival time compared with supportive care only (HR 0.42, 95% CI 0.25 to 0.71; low‐certainty evidence)
Effect estimates of other treatment options compared with supportive care were not graded for the reasons given in the summary of findings Table for the main comparison.

Similarly, pooled network estimates for four treatments (CRT, BEV_CRT, TMZ, and BEV‐RT ) could be compared with hypofractionated radiotherapy (RT40) through a common comparator, and graded evidence can be interpreted as follows.

CRT improves overall survival time compared with RT40 (HR 0.67, 95%CI 0.56 to 0.80; high‐certainty evidence). On average, this equates to a 33% lower risk of death over the course of the disease and a 49% increase in survival time.
BEV_CRT probably improves overall survival time compared with RT40 (HR 0.56, 95% CI 0.31 to 0.99; moderate‐certainty evidence)
There may be little or no difference in overall survival time between TMZ and RT40 (HR 0.95, 95% CI 0.71 to 1.26; low‐certainty evidence)
There may be little or no difference in overall survival time between BEV_RT and RT40 (HR 1.08, 95% CI 0.66 to 1.78; low‐certainty evidence)

Based on these findings, the average absolute effects on risk of death and median survival time of treatments relative to supportive care have been illustrated in summary of findings Table 2.

Pooled network estimates for four treatments (RT40, BEV_RT, TMZ and CRT) could be compared with 'standard' radiotherapy (60 Gy in 30 fractions) through a common comparator. The graded evidence related to these comparisons can be interpreted as follows.

There may be little or no difference in overall survival time between RT40 and RT60 (HR 0.94, 95% CI 0.72 to 1.23; low‐certainty evidence)
CRT may improve survival time compared with RT60 (HR 0.63, 95% CI 0.46 to 0.87; low‐certainty evidence)
The evidence on effects of TMZ and BEV_RT compared with RT60 was graded very low‐certainty.

Based on these findings, the average absolute effects on risk of death and median survival time of treatments relative to RT40 have been illustrated in summary of findings Table 3.

Interpretation of other graded network estimates are as follows.

BEV_RT may be associated with shorter overall survival time compared with CRT, however, the effect estimate includes the possibility of little or no difference (HR 1.61, 95% CI 0.95 to 2.74; low‐certainty evidence)
There may be little or no difference in overall survival time between BEV_CRT and CRT (HR 0.83, 95% CI 0.48 to 1.44; low‐certainty evidence)
TMZ may be associated with shorter overall survival time compared with CRT (HR 1.42, 95% CI 1.01 to 1.98; low‐certainty evidence)
The evidence on effects of BEV_RT compared with TMZ was graded very low‐certainty.

Ranking the treatments according to effectiveness

Table 9 gives an overview of mean SUCRA ranking of treatments according to relative effects on overall survival. BEV with CRT was ranked as the best treatment and supportive care only as the worst treatment. The second best treatment was CRT and the third best treatment was TMZ. These rankings should be interpreted with caution as they do not take into account the certainty of the evidence.

Table 9. Overview of SUCRA rankings

Treatment option	Main NMA model	Sensitivity analysis A	Sensitivity analysis B	Sensitivity analysis C	Sensitivity analysis D
BEV_CRT	1.4	1.4	1.4	–	–
CRT	1.8	1.8	1.9	1.1*	1.1*
TMZ	3.8	3.8	3.7	2.8	2.8
RT40	4.3	4.3	4.2	3.3	3.3
BEV_RT	4.7	4.7	4.8	3.8	3.8
RT60	5.0	5.1	5.0	4.0	4.0
Supp_Care	7.0	7.0	7.0	6.0	6.0

*CRT40

BEV_CRT = chemoradiotherapy plus bevacizumab; CRT = chemoradiotherapy; RT40 = radiotherapy (40Gy in 15 fractions); RT60 = radiotherapy (60Gy in 30 fractions); Supp_Care = supportive care; TMZ = temozolomide; TTF_AC = tumour treating fields with adjuvant chemotherapy (after concomitant chemotherapy)

Sensitivity analysis A: Keime‐Guibert 2007 study’s RT50 arm is coded as RT40; sensitivity analysis B: network without a loop due to exclusion of the Malmstrom 2012 TMZ vs RT40 arm to avoid duplication of data; sensitivity analysis C.1 and C.2: disconnected networks due to non‐pooling of CRT40 and CRT60 arms from different studies; sensitivity analysis D: disconnected network due to non‐pooling of CRT40 and CRT60 arms from different studies, and Keime‐Guibert 2007 study arm RT50 coded as RT40.

Sensitivity Analysis

Four sensitivity analyses were conducted and ranking of treatments relative to each other remained consistent with the main network findings (Table 9).

Overall survival data from included studies not contributing data to the NMA

There were five trials (GLARIUS 2016, Green 1983, Roa 2015, Weller 2017 and Stupp 2017a) that did not contribute overall survival data to the NMA. .

For the elderly subgroup of patients aged 65 and over, Green 1983 reported the number of deaths (103/107) and the death rate (number of deaths per 10 patient months). The death rate in the elderly subgroup was significantly higher (P < 0.00001) than in other age groups, however there was no evaluation of how death rates compared between the treatment groups.

For the Roa 2015 trial, a separate publication reported survival data per protocol and by intention‐to‐treat (ITT) for the 61 elderly patients (65 and over) who participated (Guedes de Castro 2017). The median overall survival difference was not statistically different in patients receiving 25 Gy in 5 fractions of radiotherapy compared to those receiving 40 Gy in 15 fractions of radiotherapy (6.8 months; 95% CI, 4.5‐9.1 months compared to 6.2 months; 95% CI, 4.7‐7.7 months, respectively, P = 0.936, no hazard ratio provided). The ITT analysis was conducted separately for 'elderly and not frail' patients and 'elderly and frail' patients and there was no significant difference in overall survival detected between treatment arms for either comparison.

For the GLARIUS 2016 trial, a separate abstract reported overall survival data for the modified intention‐to‐treat (ITT population for patients aged 65 and over (n = 34) compared with younger patients. In the RT + BEV/IRI arm, younger patients survived significantly longer compared with those aged 65+ (median overall survival of 17.5 months for patients aged under 65 versus 13.4 months for patients aged 65+, P < 0.001). For patients treated with CRT, no significant difference was found between age groups (median overall survival for younger patients was 20.0 months compared to 17.3 months for patients aged 65+, P = 0.567). Whilst the median overall survival for patients aged 65+ years was reported as 13.4 months and 17.3 months in the BEV/IRI and TMZ arms, respectively, there was no direct comparison stating the level of significance for overall survival between treatment arms performed for this age group.

Weller 2017 reported the number of deaths for patients with maximally resected disease (MRD) receiving rindopepimut and TMZ (31/46) versus TMZ only (36/50), with a corresponding HR of 1.21 (95% CI, 0.71 to 2.06, P = 0.48). This was also reported for the group of patients with significant residual disease (SRD) (HR 0.68, 95% CI, 0.39‐1.19, P = 0.18). Thus, there were no clear differences in overall survival between treatment arms for either participant population.

Lastly, in Stupp 2017a, tumour treating fields plus adjuvant temozolomide (TTF_AC) was compared with adjuvant temozolomide only among patients receiving CRT. In the subgroup of patients 65 years and older, the estimated HR was 0.51 (95% CI, 0.33 to 0.77) in favour of TTF_AC, with 11% of participants in the TTF_AC (10/89) and 4% (2/45) in the CRT only group alive by the end of the study. This trial was not incorporated into NMA because participants were randomised after they had received radiotherapy and concomitant chemotherapy (i.e. participants received treatment prior to the study interventions), while other studies in the NMA randomised participants before they had received radiotherapy or chemotherapy (i.e. were participants were treatment‐naive).

Other Outcomes

Evidence from pairwise comparisons of trial data pertaining to elderly participants is reported by treatment comparison below.

Radiotherapy versus supportive care

One study (Keime‐Guibert 2007) with 81 participants contributed data to this comparison.

Health‐related quality of life (HRQoL)

Evidence related to HRQoL at 30, 60, 90 and 135 day time points after diagnosis. The data suggested slightly better HRQoL scores among people receiving supportive care at the first three time points, and slightly better HRQoL scores for the radiotherapy arm at the 135 day time point (Analysis 1.1). This study also reported cognition (Analysis 1.2) and fatigue scores (Analysis 1.3) for these time points. At the furthest time point (135 days), cognition scores favoured the supportive care arm, whereas there was no clear difference in fatigue scores between the study arms at any time point. As evidence was derived from a single small study with high, unequal attrition (low response rate to questionnaires and more deaths occurring in the supportive care group), we assessed the HRQoL findings as low‐certainty evidence.

Progression‐free survival

Evidence suggested that radiotherapy probably improves progression‐free survival compared with supportive care only (HR 0.28, 95% CI 0.17 to 0.46; moderate‐certainty evidence, with downgrading as evidence was derived from a small single study).

Severe adverse events

Not reported.

Hypofractionated radiotherapy versus standard radiotherapy (60 Gy)

One included study compared 40 Gy/15 fraction schedule with a 60 Gy/30 fraction schedule (Roa 2004) and another compared a 25 Gy/5 fraction schedule with a 40 Gy/15 fraction schedule (Roa 2015), therefore data were not pooled.

Health‐related quality of life (HRQoL)

Data were reported at four weeks and eight weeks after treatment in Roa 2015 and these suggested that there may be little or no difference in HRQoL between 25 Gy/5 fraction and 40 Gy/15 fraction schedules at either time point after the one‐week and three‐week treatment schedules, respectively (Analysis 2.1; Analysis 2.2). In Roa 2004, HRQoL data were measured using the KPS at three and six weeks after treatment as medians with interquartile ranges (IQRs) and, similarly suggested little or no difference in effect on HRQoL. Subsequent follow‐up of participants in Roa 2004 also suggested little difference in average HRQoL scores, however, attrition increased with time. We downgraded the certainty of this narrative evidence of little or no difference in HRQoL to low (sparse data [‐1] and attrition bias [‐1]).

Severe adverse events

There were no instances of grade 3 or higher treatment‐related toxicity in Roa 2004 (Analysis 2.3), and this outcome was not reported in Roa 2015.

Progression‐free survival

This was not reported in Roa 2004. Roa 2015 provided a KM curve but not a hazard ratio for (HR) progression‐free survival. Alongside the KPM curve, it was reported that median progression‐free survival showed no statistically significant difference between arms (4.2 versus 4.2 months in arms 1 and 2, respectively; P = 0.716). We did not grade this evidence.

Chemotherapy (TMZ) versus radiotherapy

Two studies contributed overall survival data for this comparison (Malmstrom 2012; Wick 2012) but only one of them (Wick 2012) reported additional outcomes separately for the elderly population of interest.

Health‐related quality of life

Wick 2012 reported no clinically meaningful or significant differences in overall QoL scores at 3, 6, or 12 month measurements or other individual QoL items (emotional function, social function, nausea and vomiting, fatigue, loss of appetite, future uncertainty), except for discomfort from communication deficits, which were greatest for patients in the radiotherapy group who died at between six and 12 months (P = 0.002). They were presented graphically over time in a supplementary appendix without raw data.

Progression‐free survival

Evidence from Wick 2012 suggested that there may be little or no difference in event‐free survival (where events were progression or death) between TMZ and standard radiotherapy (373 participants; HR 1.15, 95% CI 0.92 to 1.44; Analysis 3.1; low‐certainty evidence, downgraded for study design limitations and imprecision).

Severe adverse events

Evidence derived from Wick 2012 (373 participants) suggested that, compared with standard radiotherapy.

TMZ increases the risk of thromboembolic events (Analysis 3.2) and increases the risk of severe (grade 3+) neutropenia, lymphopenia, and thrombocytopenia (Analysis 3.3; Analysis 3.4; Analysis 3.5; moderate‐certainty evidence, downgraded ‐1 for imprecision); however the confidence intervals are imprecise and the actual effect may differ from the point estimate in these analyses.
There may be little or no difference in the risk of serious infection, fatigue, nausea and vomiting, weight loss, neurological symptoms, seizures, elevated liver enzymes, and cutaneous adverse events (Analysis 3.6; Analysis 3.7; Analysis 3.8; Analysis 3.9; Analysis 3.10; Analysis 3.11; Analysis 3.12; Analysis 3.13; all low‐certainty evidence, downgraded ‐2 for serious imprecision).

Chemoradiotherapy (TMZ plus radiotherapy) versus radiotherapy

Evidence from one study (Perry 2017) contributed data to outcomes other than overall survival for an elderly population. The radiotherapy schedule used in this study was 40 Gy in 15 fractions.

Health‐related quality of life

This was briefly reported in Perry 2017 and investigators noted that attrition impacted the quantity of data. They conducted analyses using time to deterioration (with deterioration defined as a 10‐point decrease in the score on the function domain or a 10‐point increase in the score on the symptom domain) and plotted QoL scores over time. They reported that quote: "only nausea and vomiting and constipation were associated with significant differences in time to deterioration, which was shorter in the CRT group than in the radiotherapy alone group. They reported that ".There were no other clinically important differences between trial groups, which supports our observation that quality of life was similar in the two treatment groups." We did not grade this evidence.

Progression‐free survival

The evidence suggested that chemoradiotherapy delays disease progression compared with radiotherapy only (562 participants; HR 0.50, 95% CI 0.41 to 0.61; Analysis 4.1; high‐certainty evidence).

Severe adverse events

Evidence suggested that chemoradiotherapy probably increases the risk of grade 3+ neutropenia, thrombocytopenia, and leucopenia, however the confidence intervals are imprecise and the actual effect may differ from the point estimate in these analyses. (Analysis 4.2; Analysis 4.3; Analysis 4.4; Analysis 4.5; moderate‐certainty evidence, downgraded for imprecision).

Other evidence suggested that there is probably little or no difference in grade 3+ anaemia (Analysis 4.6) and other grade 3+ treatment‐related toxicity (Analysis 4.7) (both moderate‐certainty evidence, downgraded due to imprecision).

Bevacizumab plus CRT (TMZ plus radiotherapy) versus CRT

One study with 73 participants (AVAglio 2014) contributed data to this pairwise comparison.

Health‐related quality of life

This was reported for overall trial but not for elderly subgroup specifically (for overall findings, see below).

Progression‐free survival

Evidence suggested that adding bevacizumab to chemoradiotherapy may make little or no difference to disease progression (HR 0.78, 95% CI 0.46 to 1.32; low‐certainty evidence, downgraded due to imprecision and study design limitations; Analysis 5.1).

Severe adverse events

Evidence suggested that adding bevacizumab to chemoradiotherapy probably increases the risk of grade 3+ thromboembolic events compared with CRT alone (RR 16.63, 95% CI 1.00 to 275.42; moderate‐certainty evidence, downgraded due to imprecision; Analysis 5.2). No other adverse events were reported for the elderly only. Serious adverse events reported for the overall sample including younger participants can be found below.

Bevacizumab plus radiotherapy versus radiotherapy

One study with 75 participants (ARTE 2018) contributed data to this pairwise comparison.

Health‐related quality of life

In the publication, global health was reported in a forest plot along with individual HRQoL items, such as cognitive functioning, emotional functioning and pain, measured with EORTC QLQ‐C30/BN20 scales. Investigators reported that quote: "before progression, no differences were detected for individual scales in a generalized linear mixed model, except for less favourable values in arm A for global health (P=0.048) and pain (P=0.027)". No other data were provided or obtained and we did not grade this evidence.

Progression‐free survival

Evidence suggested that adding bevacizumab to radiotherapy probably delays disease progression (HR 0.46, 95% CI 0.27 to 0.78; moderate‐certainty evidence, downgraded due to study design limitations; Analysis 6.1).

Severe adverse events

Low‐certainty evidence suggested that there may be little or no difference in various grade 3+ adverse events reported in this study, including thromboembolic events (Analysis 6.3); haematological events (Analysis 6.2); infections (Analysis 6.4); fatigue (Analysis 6.5); seizures (Analysis 6.6); headache (Analysis 6.7); neuropsychiatric events (Analysis 6.8); neurological events (Analysis 6.9); hypertension (Analysis 6.10); cutaneous adverse events (Analysis 6.11); and gastrointestinal events (Analysis 6.12).

Other comparisons did not report progression‐free survival for the elderly subgroups of participants.

Health‐related quality of life (HRQoL) and adverse event data not specific to elderly patients

Several of the included studies (AVAglio 2014; GLARIUS 2016; Green 1983; Malmstrom 2012; Roa 2004; Stupp 2017a; Weller 2017) reported HRQoL or adverse event data for the overall trial population but not separately for the elderly subgroup of patients. Although they are not specific to the elderly population, these overall findings may give some indication of the degree of toxicity of the treatments and any detriment to patients' quality of life for the elderly subgroup too. We have therefore presented the main results below with corresponding P values when available. This evidence is not rated for certainty.

Health‐related quality of life (HRQoL)

Malmstrom 2012 measured HRQoL at baseline, six weeks and three months using EORTC QLQ‐C30 and BN‐20 questionnaires. Patients in the TMZ arm generally reported better quality of life than in either of the radiotherapy arms (60 Gy and 34 Gy), but ratings for global health status were similar.
AVAglio 2014 reported HRQoL outcomes in a separate publication (Taphoorn 2015). The addition of BEV to CRT delayed deterioration (reported as deterioration‐free survival (DFS)) across five pre‐selected HRQoL scale measures (global health, physical functioning, social functioning, motor dysfunction and communication deficit). Deterioration was defined as a clinically significant deterioration in HRQoL (worsening of 10 or more points on the respective HRQoL scale), progressive disease, or death. It was suggested that the delayed disease progression in the BEV_CRT arm (reported in the main publication) may have influenced the DFS result. When progressive disease was excluded as an event, participants treated with BEV_CRT had a statistically significantly delayed deterioration in HRQoL domains of communication, social functioning and global health but not for motor dysfunction or physical functioning.
GLARIUS 2016, which compared treatment with CRT versus treatment with RT60 in combination with concomitant and adjuvant bevacizumab and adjuvant irinotecan, reported HRQoL using QLQ C30 and BN20 questionnaires measured at baseline and every three months until death or end of study. There was no significant difference between the treatment arms.
Weller 2017 compared adding rindopepimut or control to adjuvant TMZ after CRT and reported no significant differences between patients in their trial arms in any of the HRQoL measures.
Stupp 2017a reported HRQoL in a separate publication (Taphoorn 2018). There was no significant difference in HRQoL reported between the trial arms, except for itchy skin which was more prominent in the TTF arm at 3, 6 and 9 months (P = 0.005, P = 0008, P = 0.04). There was no significant difference at 12 months.

Severe adverse events (Common Terminology Criteria for Adverse Events (CTCAE) grade 3+)

Green 1983 reported the percentage of patients in each of the four trial groups that experienced specific toxicities. The proportion of patients in the procarbazine arm who suffered a grade 3+ dermatologic or allergic reaction (25%) and nausea and/or vomiting (12.5%) were both significantly higher than for the other three arms of the trial. Infection rates were highest for the BCNU and methylprednisolone arm (34.3%) and rates of uncontrolled diabetes and skeletomuscular complications were highest for the methylprednisolone arm (3.5% and 7.8% ,respectively).
Malmstrom 2012 reported toxicity using WHO grading for adverse events, except for nausea and vomiting for which they used the CTCAE version 2.0. In the overall population, episodes of grade 3+ haematological toxicities (neutropenia, pancytopenia and thrombocytopenia) were only seen in the TMZ arm. The incidence of grade 3+ seizures and fatigue was more common in the radiotherapy arms compared to the TMZ arm. Infection rates were similar across all arms. There were two patients who had fatal infections, one in the TMZ group and one in the 60 Gy RT group. There was one death in the TMZ group attributed to bleeding due to grade 2 thrombocytopenia.
AVAglio 2014 used CTCAE v3.0 to measure severe adverse events. Grade 3+ cerebral bleeding events (2.0% versus 0.9%) and wound healing events (3.3% versus 1.6%) were higher in the BEV plus CRT arm versus CRT alone. There were also higher rates of G3+ thrombocytopenia (15% versus 9.8%) and infection rates (12.8% versus 7.8%) in the BEV arm.
GLARIUS 2016 used CTCAE v3.0 to measure severe adverse events. Rates of severe adverse events were 72% for the bevacizumab (BEV)‐irinotecan (IRI) plus RT 60 arm, and 84% in the CRT arm. For the BEV/IRI/RT60 arm, severe vascular events were most common (11.8%) and two cerebral haemorrhages occurred (one fatal). For patients in the CRT arm, severe haematological toxicity was most common (18.2%).
In Weller 2017, the most common severe adverse events for the experimental (rindopepimut) versus placebo arm of the trial were: thrombocytopenia (9% versus 6%), fatigue (2% versus 5%), brain oedema (2% versus 3%), seizure (2% versus 2%) and headache (2% versus 3%). There was one death, secondary to pulmonary embolism, that was assessed as potentially related to the treatment in the experimental arm.
Stupp 2017a reported that there was no significant increase in rates of severe adverse events when TTF were added to adjuvant chemotherapy (48% versus 44%, P = 0.58). There was a numerically higher incidence of some adverse events in the TTF group but the authors report that this was a reflection of the longer duration of TMZ treatment in this group due to delayed occurrence of progression and that the difference disappeared when adverse event incidence was normalised to duration of treatment. There was a higher incidence of skin toxicity (grade 3 in 2%) for the TTF arm compared to the control arm.

Economic evidence

The economic evaluation that was identified (Ghosh 2018) was a cost‐effectiveness analysis (CEA) and cost‐utility analysis (CUA). The associated study compared the use of a short course of radiotherapy in elderly patients based on the trial by Roa 2015. The trial reports clinical outcomes were expressed as overall survival and progression‐free survival for the CEA and as quality adjusted life years (QALYs) for the CUA (Table 10).

Table 10. Data extraction table for economic studies

Author(s), Year & Title

Type of Evaluations

Sources of effectiveness data

Sources of Cost Data

Sources of outcome valuations

Analytical approach

Ghosh 2018

Improved cost‐effectiveness of short‐course radiotherapy in elderly and/or frail patients with glioblastoma

Cost‐effectiveness analysis

Cost‐utility analysis

Effectiveness data was taken from a phase III randomised trial.

Direct unit medical costs were collected from each country participating in the trial (from 2015).

Outcomes presented were overall survival (OS) and progression free survival (PFS). The outcomes were also presented as QALYs. Preference values were taken from European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ‐C30). The scores were mapped onto the EQ‐5D questionnaire from which utility values were derived. The values were taken from three sources; Kontodimopoulos 2009, Kim 2012 and McKenzie 2009. The QALY calculation based on the assumption that the patient on treatment is to live for 4 months.

Restricted mean overall survival (RMOS)

Incremental cost‐effectiveness ratio (ICER).

Direct unit medical costs (i.e. costs which result from the utilisation of the medical intervention) were collected from the associated trial (Roa 2015). The costs were broken down for each country participating in the trial. These costs were shown in an additional table. No indirect costs (i.e. costs associated with losses as a consequence of illness, such as production or leisure time lost to patients and their families) were included. The direct costs included the costs of the dexamethasone, magnetic resonance imaging (MRI) scans, computed tomography (CT) scans and radiotherapy (Table 11). The authors presented costs as USD 2015, but did not describe the methods for converting the costs from the various participating countries. The mean total cost of the 25 Gy arm was $2,475 and the mean total cost for the 40 Gy arm was $2,868. The authors report that confidence intervals were undefined for the difference in cost due to the negative ‐cost difference, although the scientific rationale for this statement is unclear, as there is no reason why a CI could not be estimated when the point estimate for the difference in cost between the short course and commonly used RTs is negative.

Table 11. Unit costs table for Included studies

Component	Study	Country	Estimated costs of resources used	Source	Currency	Average number of Resources Used Arm 1 / Arm 2		Source
Dexamethasone 4 mg tablet	Ghosh 2018	Belarus	0.27	Not reported	US Dollars (conversion not reported)	44	16	Trial Dataset
		Brazil (Porto Alegre)	0.06	‐ Not reported	US Dollars (conversion not reported)	44	60	Trial Dataset
		Georgia	0.3	Not reported	US Dollars (conversion not reported)	18	4	Trial Dataset
		India	0.01	Not reported	US Dollars (conversion not reported)	32	24	Trial Dataset
		Poland	1.36	Not reported	US Dollars (conversion not reported)	12	8	Trial Dataset
CT	Ghosh 2018	Belarus	269	Not reported	US Dollars (conversion not reported)	4	3	Trial Dataset
		Brazil (Porto Alegre)	119	Not reported	US Dollars (conversion not reported)	4	3	Trial Dataset
		Georgia	130	Not reported	US Dollars (conversion not reported)	1	1	Trial Dataset
		India	25	Not reported	US Dollars (conversion not reported)	1	1	Trial Dataset
		Poland	57	Not reported	US Dollars (conversion not reported)	4	1.5	Trial Dataset
MRI	Ghosh 2018	Belarus	330	Not reported	US Dollars (conversion not reported)	5	‐	Trial Dataset
		Brazil (Porto Alegre)	293	Not reported	US Dollars (conversion not reported)	4	‐	Trial Dataset
		Georgia	230	Not reported	US Dollars (conversion not reported)	1	3	Trial Dataset
		India	58.3	Not reported	US Dollars (conversion not reported)	3	1	Trial Dataset
		Poland	130	Not reported	US Dollars (conversion not reported)	3	3	Trial Dataset
RT (5 fractions)	Ghosh 2018	Belarus	600	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Brazil (Porto Alegre)	1046	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Georgia	2900	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		India	25	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Poland	3472	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
RT (15 fractions)	Ghosh 2018	Belarus	1800	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Brazil (Porto Alegre)	1985	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Georgia	3800	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		India	42	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Poland	3472	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset

The results of the cost‐effectiveness were expressed as Incremental Cost Effectiveness Ratios (ICERs). The reported ICERs in USD were ‐$3,062 for the restricted mean overall survival per life‐year gained and ‐$17,693 USD for the restricted mean progression‐free survival. The presentation of negative ICERs is not advised, as negative data points have no meaningful ordering (O'Brien 2002). For overall survival, the study reports a net benefit with 25 Gy of ‐$46,907 at a societal willingness to pay level $50,000, a net benefit of ‐$93,438 at the $100,000 threshold, and a net benefit of ‐$159,970 at the $200,000 threshold. For progression‐free survival, net benefit is reported as ‐$1,933 at the $50,000 threshold, ‐$4,241 at the $100,000, and ‐$8,680 at the $200,000 threshold. Given the clinical outcome data presented in the paper, it is unclear how these numbers are calculated as they do not make intuitive sense.

The utility values for the CUA were derived from three different mapping algorithms from the EORTC QLQ‐C30 questionnaire (Kontodimopoulos 2009;Kim 2012; McKenzie 2009). Two reviews have identified limitations in using the EORTC QLQ‐C30 mapping algorithms and the three algorithms used in this study performed poorly in validation tests, so their outputs should be used cautiously (Doble 2016; Woodcock 2018). The authors held the assumption that the participants would survive for four months with either treatment. The QALY valued for the 40 Gy treatment was therefore: QALY overall = 0:333 times the utility obtained at baseline. However, the QALY calculated using the mapping algorithm was calculated as: QALY = Utility at month 1 times 0:083 + Utility at month 4 times 0:333. This would result in a 25 Gy‐treated QALY being calculated for five months, whereas the 40 Gy treated individual was being calculated for four months, which was against the authors stated assumption. This means that the gains calculated for the QALY ICERs may be due to this potential calculation error rather than the effect of the intervention itself.

Stochastic sensitivity analysis was carried out in the form of bootstrapping to assess sampling uncertainty. The authors did not carry out a deterministic sensitivity analysis, as variation in cost and survival effect size were analysed using the bootstrap procedure and cost‐effectiveness acceptability curves. This is not in line with current UK guidelines (NICE 2012) who recommend the use of deterministic and probabilistic sensitivity analysis to assess parameter uncertainty. The International Society of Pharmacoeconomics and Outcomes Research (ISPOR) guidelines also recommend the examination of parameter uncertainty (Husereau 2013).

The authors conclude that since their ICER values are less than the threshold, they can conclude that the 25 Gy radiotherapy is cost‐effective. The review authors cannot replicate the results of this economic evaluation from the data presented and the study results should be considered with extreme caution. As this was the only study identified and has potential quality issues, this demonstrates a paucity of economic evidence regarding the management of newly diagnosed glioblastoma in the elderly.

Quality of Economic studies

The quality of the trial on which Ghosh 2018 is based (Roa 2015) has been discussed in Risk of bias in included studies. The study was found to have a low risk of bias (with the exception of the blinding, which was open‐label). The CHEERs checklist (Husereau 2013) and the CHEC checklist (Evers 2005) were applied to the study to assess the quality of economic evaluation as recommended by the current guidelines (Shemilt 2018). The results from these can be found in Table 6 and Table 7. The results of the CHEERS reporting checklist show that there a number of parameters are not reported (e.g. sources of costs, time horizon, perspective). The results of the CHEC checklist show that there a number of issues with the methodological quality of the study, including inappropriate costing and analysis methods.

Discussion

Summary of main results

The review included 12 studies involving approximately 1818 elderly participants and several different treatment comparisons. Seven treatment options could be connected in a network meta‐analysis (NMA) for the outcome of overall survival. Other treatments and outcomes were evaluated in pairwise comparisons where data were available.

Overall survival

We found high‐certainty evidence that chemoradiotherapy results in a 33% lower risk of death on average over the course of the disease after diagnosis (20% to 44% lower), or about a 50% increase in survival time compared to treatment with hypofractionated radiotherapy (RT) (40 Gy) alone. Other evidence was assessed as moderate to very low certainty, with most evidence graded as low or very low certainty, meaning that the effect estimates may be substantially different from those estimated in our network meta‐analysis (NMA). However, all treatments evaluated led to a clear increase in survival time relative to supportive care only, except for the bevacizumab plus radiotherapy option.

In terms of treatment rankings, bevacizumab (BEV) added to chemoradiotherapy (CRT) ranked as the best treatment, CRT ranked second best, temozolomide (TMZ) ranked third best, hypofractionated radiotherapy fourth, and supportive care only ranked last. These ranking should be interpreted with caution as they do not take into account the certainty of the evidence, notably that there may be little or no difference in overall survival when BEV is added to CRT. Therefore, the higher position of BEV_CRT in the ranking is not supported by evidence of a clear survival benefit over CRT. On sensitivity analysis, when CRT40 (with 40 Gy RT) and CRT60 (with 60 Gy RT) interventions were considered separately, CRT (with RT40 Gy) ranked first in the larger of the two networks thus created.

Summary of findings with illustrative effects can be found in the summary of findings Table for the main comparison, summary of findings Table 2 and summary of findings Table 3. We were unable to conduct subgroup analyses by age threshold (65+ or 70+ year threshold), or methylguanine‐DNA‐methyltransferase (MGMT) methylation status; few included studies reported the latter and where these data were present, they were usually reported for the sample overall and not for the elderly subgroup.

Quality of life

Moderate‐certainty narrative evidence suggests that overall, there may be little difference in quality of life between TMZ and radiotherapy, except for discomfort from communication deficits, which were more common with radiotherapy. Data on quality of life for other treatment comparisons were sparse and negatively impacted by attrition, with the limited available evidence derived from elderly participants suggesting little or no difference in quality of life with radiotherapy versus supportive care only, and short course versus longer/standard radiotherapy courses (summary of findings Table 4). Narrative evidence from a single study of chemoradiotherapy versus radiotherapy only suggesting little difference in quality of life was not graded.

Progression‐free survival

High‐certainty evidence shows that chemoradiotherapy delays disease progression compared with hypofractionated radiotherapy only. Moderate‐certainty evidence suggests that radiotherapy with 60 Gy probably delays disease progression compared with supportive care only and that bevacizumab with radiotherapy probably delays disease progression compared with hypofractionated radiotherapy alone. Evidence for other treatment comparisons is of low or very low certainty.

Severe adverse events

Moderate‐certainty evidence suggests that severe haematological toxicities and thromboembolism are more common with TMZ than with radiotherapy and the risk probably increases with the addition of BEV to CRT.

Economic evaluations

The review identified a single economic evaluation in the target population. This study had a number of issues relating to methodological quality so the results should be interpreted very cautiously. This demonstrates a current lack of economic evidence evaluating different strategies of managing of newly diagnosed glioblastoma in the elderly. Other economic evaluations that were identified did not consider the over 65's with newly diagnosed glioblastoma as a relevant population or subgroup.

Overall completeness and applicability of evidence

The only outcome that could be assessed in a network was that of overall survival. Thus, it is not known how the treatment options compare with regard to the other important review outcomes, such as quality of life or severe adverse events. This is a serious limitation of the evidence gathered in this review as a treatment ranked as best for overall survival, for example, could be worst for quality of life. More research on quality of life among patients receiving treatment for glioblastoma is necessary to elucidate these other relative effects. However, attrition is a notable problem for investigators gathering these sorts of data.

In general, the review evidence is applicable to elderly patients with a Karnofsky Performance Score of more than 70 per cent, i.e. those patients capable of self‐care (see Table 3). We found little evidence to inform guidance on the most appropriate treatments for people with KPS less than 70 percent. Two studies that evaluated different hypofractionated radiotherapy regimen included patients at the frailer end of the spectrum, using a KPS of 50 percent (Roa 2004; Roa 2015). Evidence from Roa 2015 suggested little or no difference in the median survival between a 25 Gy/5 fraction regimen and the 40Gy/15 fraction regimen among elderly and frail patients; however, the effectiveness of the 25Gy/5 fraction regimen could not be evaluated against other possible treatment options in the network due to insufficient data. Malmstrom 2012 also permitted entry to their trial of patients (7/291 patients) with a WHO performance status of 3 if this was specifically due to neurological status. However, this was a minority of patients and the outcomes for these patients were not reported separately.

Data were also relatively scarce for certain treatments, particularly newer treatment options (e.g. those employing bevacizumab) and some treatment options lacked overall survival data in a comparable form (e.g. radiotherapy given as 25Gy in five fractions); therefore, such treatments could not be ranked at all against other treatments. Crucially, we were unable to compare chemoradiotherapy utilising 60 Gy in 30 fractions (standard) with hypofractionated regimens, either directly or indirectly, due to limitations of the network connections. Therefore, we could not ascertain a network effect estimate for standard versus a hypofractionated chemoradiotherapy regimen, nor rank these different regimen.

One study (Roa 2015) compared two hypofractionated chemoradiotherapy regimen (25 Gy in five fractions versus 40 Gy in 15 fractions). Unfortunately, we were unable to compare and rank the 25 Gy radiotherapy regimen because the overall survival data in this study were reported as median survival times with P‐values, rather than as time‐to‐event data (HRs and 95% CIs). The difference in median survival was not statistically significant (P=0.936). We rated the resulting evidence as low‐certainty. We were unable to obtain the relevant time‐to‐event data from the investigators for this review.

In addition to the limitations of the evidence with respect to performance status mentioned above, there were very few data on health‐related quality of life (HRQOL) and severe adverse effects reported specifically for elderly patients. In studies where the elderly were a subgroup of a sample with a broader age range, these outcomes were frequently reported by investigators for the overall sample. When we found no specific data on these outcomes for the elderly, we reported the main findings for the broader age group at the end of the results section. However the applicability of these findings to an elderly population is not known and the actual relative effects may be quite different.

Quality of the evidence

The network evidence that was rated as high quality/certainty was:

chemoradiotherapy increases time to death and delays disease progression compared with hypofractionated radiotherapy.

Evidence of moderate quality/certainty, meaning that our actual effect may differ somewhat from our point estimate (or may change with further research), included the following:

chemoradiotherapy increases the risk of grade 3+ neutropenia, thrombocytopenia, and leucopenia compared with hypofractionated radiotherapy alone;
bevacizumab plus chemoradiotherapy increases time to death compared with hypofractionated radiotherapy alone;
standard radiotherapy increases time to death and delays disease progression compared with supportive care only;
bevacizumab plus radiotherapy delays disease progression compared with hypofractionated radiotherapy alone;
temozolomide increases the risk of grade 3+ thromboembolic events, neutropenia, lymphopenia, and thrombocytopenia compared with standard radiotherapy; and
bevacizumab plus chemoradiotherapy increases the risk of thromboembolic events compared with chemoradiotherapy alone.

Other evidence was low or very low quality/certainty and the effect estimates (if any) are likely to change with further research. Data on quality of life were sparse and the certainty of the evidence tended to be of very low quality/certainty or unrateable.

SUCRA sSurface under the cumulative ranking curve) ranking does not take into account the certainty of the evidence and a high‐ranked treatment may be based on low‐certainty evidence (Mbuagbaw 2017). In our main NMA, the evidence on bevacizumab added to chemoradiotherapy compared with chemotherapy alone was less robust than the evidence on chemoradiotherapy alone; however, it ranked higher than the latter. When compared with chemoradiotherapy, low‐certainty evidence suggested that the addition of bevacizumab to chemoradiotherapy did not improve overall survival, highlighting that rankings should be interpreted with caution and that more research evidence may be needed to improve the certainty of the rankings.

Potential biases in the review process

There were some important differences between trial inclusion criteria, treatments and outcome reporting that could not be accounted for in this review process and could have contributed to potential bias.

Definitions of 'the elderly'

Eight trials defined the elderly subgroup as 65+ years old (ARTE 2018; GLARIUS 2016; Green 1983; Perry 2017; Roa 2015; Stupp 2017a; Weller 2017; Wick 2012), and only three studies contributed data for the elderly according to the review focus of 70+ years old (AVAglio 2014; Keime‐Guibert 2007; Malmstrom 2012) (Table 2). People aged between 65 and 70 years were shown to have a longer median survival compared with those of 70+ years (6 months versus 3.2 months, respectively) in a large UK audit (Brodbelt 2015). Our decision to pool data for these studies was pragmatic and taken at the protocol stage, because we knew that data specifically for the 70+ age group would be sparse.

We also included one study (Roa 2004) that defined elderly patients as aged 60+ years because the review protocol dictated that the overall results from trials including younger patients could be included if the proportion of patients in the trial aged over 65 years exceeded 80%. In Roa 2004, the mean participant age was 72.4 years for patients treated with 60 Gy and 71.0 years for patients treated with 40 Gy. The standard deviation for these groups was 5.4 and 5.5 years, respectively. Therefore the majority of participants in this trial were likely to be aged over 65 years; whether the proportion exceeded 80% as per our inclusion criteria is unclear. After attempting to contact the authors of Roa 2004, with no further information on the proportion of patients included aged over 65 years obtained, we decided to include this study based on this rationale. The median survival of 5.1 months in the RT60 arm of this trial was similar to the median survival reported by Malmstrom 2012 for patients aged 70+ who received 60 Gy (5.2 months), and less than the median survival reported for patients aged 70+ in Keime‐Guibert 2007 (29.1 weeks), who received 50 Gy, and for patients aged 65+ in Wick 2012 (9.6 months) who received 60 Gy.

As younger participants survive longer than older participants, the effect of including studies with 65+ year old participants might have over‐estimated the beneficial effects of treatments for the 70+ year old age group.

Radiotherapy treatment

Green 1983 was the oldest study included in the review and was published 21 years before the next included study (Roa 2004). As whole brain radiotherapy was used (WBRT) was used in Green 1983, the radiotherapy volume treated was much larger than the treatment volumes specified for the other trials that included radiotherapy. The larger treatment volume and likely sub‐optimal planning and treatment delivery techniques would be considered unacceptable by modern standard and is likely to have affected the tolerability of the treatment and the rate of adverse events, especially for elderly patients. Whilst we included this trial, it did not contribute survival data to the NMA or data for pairwise comparisons of other review outcomes, therefore any potential bias introduced by including this study would be minimal.

Timing of randomisation

Most trials randomised patients in the period following surgical resection when they were radiotherapy‐ and chemotherapy‐naive. Two trials (Stupp 2017a; Weller 2017) performed randomisation after patients had completed concomitant CRT and both specified that patients must have received at least 90% of the planned radiotherapy dose (60 Gy). This will have selected for a group of patients with a better prognosis than those in trials using the earlier time point of randomisation, by excluding those patients who were unable to tolerate treatment due to adverse events or who died or progressed prior to finishing radiotherapy. Whilst Stupp 2017a reported relevant time‐to‐event data for the elderly subgroup, we did not include these data in the NMA due to the risk of intransitivity.

Survival times

Most of the included studies (including Stupp 2017a and Weller 2017, which randomised participants after concomitant CRT) calculated overall survival from the time of randomisation, but for several studies the starting point for overall survival analysis was not described (ARTE 2018; AVAglio 2014; GLARIUS 2016) and for one trial (Wick 2012), overall survival was measured from the date of surgery.

Tumour response assessment

For those trials that reported a response rate or progression‐free survival, several (ARTE 2018; AVAglio 2014; Keime‐Guibert 2007; Perry 2017; Stupp 2017a; Weller 2017; Wick 2012) used repeated magnetic resonance imaging (MRI) for the basis of their assessments. Roa 2015 did not specify the modality of imaging and in Keime‐Guibert 2007, imaging was permitted with MRI or computed tomography (CT). ARTE 2018 and Weller 2017 used the response assessment in neuro‐oncology (RANO) criteria (Wen 2010), GLARIUS 2016; Stupp 2017a; Wick 2012 specified the MacDonald criteria (MacDonald 1990), and AVAglio 2014; Keime‐Guibert 2007; Perry 2017 described their response criteria, which were based on specific MRI appearances and/or steroid use and symptoms. Roa 2015 did not specify the criteria used. Although there are similarities between these response criteria, the differences may mean that the response rate or progression‐free survival. results across these trials are not comparable.

Extent of surgical resection

Extent of surgical resection influences prognosis (Pessina 2018). Most trials permitted inclusion of patients who had undergone biopsy or partial or complete surgical resection. In the Keime‐Guibert 2007 trial, which was one of the older included studies, at least half the participants had biopsy only. Including this trial in the NMA might, therefore, have led to the effect on overall survival estimated for other treatments to be slightly over‐estimated.

In the trial by Weller 2017, the primary analysis was conducted on patients who had maximal surgical resection (MRD) only. Whilst in the context of NMA, this could have favourably biased the effect of the study intervention, Weller 2017 did not contribute data to the NMA, therefore did not bias the NMA findings. The extent of resection was not described in ARTE 2018, and we did not evaluate the extent of possible bias from including this trial; however, the direct evidence derived from this trial in the NMA was rated low certainty.

Molecular subtypes

Two trials included patients with disease of particular molecular subtypes only, which may not have been comparable with patients included in other trials. Weller 2017 specified that only patients with glioblastoma with confirmed epidermal growth factor receptor (EGFR) vIII expression were eligible for trial inclusion. The rationale was that rindopepimut was most likely to be active in disease expressing this mutation. This explains the relatively small number of patients enrolled (n = 745) compared to the number assessed for eligibility (n = 4652). In the GLARIUS 2016 trial, only patients with unmethylated MGMT were eligible for trial inclusion. As MGMT methylation has a known prognostic effect on survival of patients with glioblastoma , it is important to consider the results of this trial and how they compare to results from other trials in this context.

Analysis and grading

For the main NMA for overall survival, comparing all treatments with supportive care only, we made the following assumptions:

that it was reasonable to pool data on chemoradiotherapy (combined TMZ and radiotherapy), irrespective of the radiotherapy dose schedule used (40 Gy/15 fractions and 60 Gy/30 fractions in Perry 2017 and AVAglio 2014, respectively);
that data on a 50 Gy/28 fraction radiotherapy schedule from the one study (Keime‐Guibert 2007) employing this unconventional dose schedule could be pooled with those of 60 Gy/30 fraction.

The rationale for the first assumption was that 'standard' chemoradiotherapy might include either radiotherapy schedule combined with TMZ. In addition, pairwise analysis suggested that there was no clear difference in overall survival between 40 Gy/15 and the 60 Gy/30 schedules when employed without chemotherapy. However, the decision to pool these data was also influenced by the fact that pooling these data would facilitate a connected network. To evaluate the extent to which the first assumption impacted the findings of the review, we conducted a sensitivity analysis by not pooling these data, which resulted in a network with a disconnected (separate) comparison (Figure 6C). Findings of the relative effects of the interventions were similar to the main findings and made no difference to the treatment ranking of chemoradiotherapy. In grading this evidence, we therefore did not downgrade the evidence on chemoradiotherapy for intransitivity (differences in study characteristics that may modify treatment effect) because intransitivity did not appear to have a serious impact on the estimates of effect.

Figure 6

Network diagrams for four sensitivity analyses

With respect to the second assumption, we performed a sensitivity analysis by pooling the Keime‐Guibert 2007 data (RT50 Gy) with the 40 Gy/15 fraction node (Figure 6A) and found that this also made no difference to the treatment rankings. Similarly, we did not downgrade the evidence for intransitivity.

The only loop in the network was derived from a single study with three arms (Malmstrom 2012), which led to duplication of data at the hypofractionated RT node in the main analysis. To evaluate the impact of this on the findings, we performed a sensitivity analysis by removing one of the study arms (TMZ versus hypofractionated RT) (Figure 6B). This made little difference to effect estimates and treatment rankings, therefore we did not downgrade for intransitivity (differences in study characteristics that may modify treatment effect).

Where studies evaluated the radiotherapy schedules only (i.e. without chemotherapy), we did not pool data for 40 Gy/15 and 60 Gy/30 schedules. However, we conducted an exploratory analysis by collapsing these nodes to evaluate the extent to which so doing would have impacted on the review findings.

Because most treatments were not part of a loop in the main network analysis, and the only loop came from a single study such that the results for the nodes in the loop correlated with one another, it was not possible to calculate indirect estimates of effect. This meant that we had to adopt a modified grading approach as it was not possible to assess incoherence (differences between the direct and indirect estimates of effect).

Calculation of absolute risk of death

As an assumed baseline risk for supportive care, we used survival data from Keime‐Guibert 2007. Whilst this is an older study and had a lower proportion of patients having had maximal surgical resection, it was the only one to evaluate supportive care only. Noting that in the an epidemiological glioblastoma study (Brodbelt 2015), the death rate for the 70+ age group was approximately 50% at three months from diagnosis (with a median survival of 3.2 months reported), we felt that the Keime‐Guibert 2007 data were a reasonable baseline against which to illustrate the potential effects of the different treatments. Relative to the Keime‐Guibert 2007 data, it could be argued then, for the relevant comparisons, that the estimates of more recent interventions relate to maximal surgical resection plus the experimental option (e.g. CRT) and not just the experimental option alone.

Agreements and disagreements with other studies or reviews

Our review evidence on hypofractionated (40 Gy) versus standard radiotherapy, suggesting little or no difference in overall survival between these radiotherapy doses, contrasts with evidence from Bleehen 1991, a trial conducted among mostly younger patients, which implied that higher radiotherapy doses were more effective. Bleehen 1991 compared postoperative treatment with 60 Gy RT in 30 fractions with 45 Gy RT in 20 fractions for patients with grade 3 or grade 4 glioma aged 18+ years, reporting that the higher dose significantly improved overall survival in this study. Although 61% (n = 272) of patients in Bleehen 1991 had a diagnosis of glioblastoma and 32% (n = 140) were aged over 60, no subgroup analyses for an elderly cohort were reported and therefore it is impossible to know if the same survival advantage would have been seen for the group of older (65+) patients in this trial.

Reyes‐Botero 2018 was a single‐arm, non‐randomised phase II trial (n = 66) which treated patients aged 70+, and with a : Karnofsky performance score (KPS) of under 70 with TMZ 130‐150 mg/m2 per day for five days every four weeks concomitantly with BEV 10 mg/kg every two weeks. Median overall survival of 23.9 weeks (95% confidence interval (CI), 19 to 27.6 weeks) was less than observed for patients of the same age treated with TMZ alone using the same schedule (9.0 months, 95% CI, 6.2 to 11.8 months) in a trial included in this NMA (Malmstrom 2012), however the patients in Malmstrom 2012 had better performance status. The adverse events which occurred from using TMZ and BEV were reported by the Reyes‐Botero 2018 authors to be tolerable; however, there were three deaths from pulmonary embolism, intestinal perforation and cerebral haemorrhage, which were recorded as probably being attributed to treatment. These are in keeping with the characteristic adverse events accompanying treatment with BEV described in trials included in this review (ARTE 2018; AVAglio 2014; GLARIUS 2016).

We know from service audits, surveys and guidelines (Brodbelt 2015; NCCN 2018; NICE 2018; Palmer 2018), that best supportive care is often the most popular treatment option for patients with poorer performance status. Although most studies in our review included relatively fit patients (KPS over 60 to 70), two trials (Roa 2004; Roa 2015) included unfit (KPS as low as 50) elderly patients. Findings from these two trials suggest that less intensive and less toxic treatment options, such as 40 Gy in 15 fractions or 25 Gy in five fractions, may be appropriate in selected cases. Interestingly, the median survival for unfit, elderly patients in Roa 2015 (reported in Guedes de Castro 2017) was superior in both treatment arms (40 Gy in 15 fractions (6.2 months; 95% CI, 4.7 to 7.7 months), 25 Gy in five fractions (6.8 months; 95%, 4.5 to 9.1 months)), compared with survival outcomes previously reported for fitter elderly patients receiving best supportive care alone (median overall survival 3.9 months, Keime‐Guibert 2007).

Gállego 2011 was a single‐arm phase II trial of TMZ (150 to mg/m2/day 200 mg/m2/day for five days every four weeks until progression) in elderly (age >70 years) and frail (KPS less than 70) patients. Median overall survival was six months and, in the small subgroup of patients known to have MGMT methylated disease, median survival was 31 weeks. Overall, quality of life and cognition improved on treatment and approximately one third of patients showed an improvement in KPS of at least 10 points. Although not a direct comparison, this survival time is also longer than reported for fitter elderly patients receiving supportive care alone (Keime‐Guibert 2007). In line with conclusions from Roa 2004 and Roa 2015, this suggests that active treatment in carefully selected patients of poorer performance status can be well‐tolerated without significant deterioration in quality of life.

Cost‐effectiveness evaluations of glioblastoma treatments specific to elderly populations are needed. A Canadian observational study that reported a median overall survival of six months among its elderly glioblastoma cohort, found that at least one in five of these over 65‐year‐olds spent more than 20% of their remaining time as inpatients (Moroney 2017); therefore, understanding the potential resource implications of this group is important. There are existing economic evaluation studies in non‐elderly populations, but we are unsure of their quality and applicability. Specifically, we are uncertain about the applicability of some recent economic evaluations on BEV and tumour treating fields (TTF) that have been conducted in non‐elderly populations with newly diagnosed glioblastoma (Bernard Arnoux 2016; Gazauskas 2019; Kovic 2015).

Figure 1

Study flow diagram (search date 13 June 2018)

Figure 2

Study flow diagram (search date 3 April 2019).

Figure 3

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.

Figure 4

Figure 5

Forest plot of all treatment comparisons for overall survival

Figure 6

Network diagrams for four sensitivity analyses

Analysis 1.1

Comparison 1 Radiotherapy (50 Gy) versus supportive care, Outcome 1 HRQOL.

Analysis 1.2

Comparison 1 Radiotherapy (50 Gy) versus supportive care, Outcome 2 Cognition.

Analysis 1.3

Comparison 1 Radiotherapy (50 Gy) versus supportive care, Outcome 3 Fatigue.

Analysis 1.4

Comparison 1 Radiotherapy (50 Gy) versus supportive care, Outcome 4 Progression‐free survival.

Analysis 2.1

Comparison 2 Short‐course radiotherapy versus standard radiotherapy, Outcome 1 HRQOL at 4 weeks.

Analysis 2.2

Comparison 2 Short‐course radiotherapy versus standard radiotherapy, Outcome 2 HRQOL at 8 weeks.

Analysis 2.3

Comparison 2 Short‐course radiotherapy versus standard radiotherapy, Outcome 3 Treatment toxicity G3+.

Analysis 3.1

Comparison 3 Chemotherapy versus radiotherapy, Outcome 1 Progression‐free survival.

Analysis 3.2

Comparison 3 Chemotherapy versus radiotherapy, Outcome 2 Thromboembolic event G3+.

Analysis 3.3

Comparison 3 Chemotherapy versus radiotherapy, Outcome 3 Neutropenia G3+.

Analysis 3.4

Comparison 3 Chemotherapy versus radiotherapy, Outcome 4 Lymphopenia G3+.

Analysis 3.5

Comparison 3 Chemotherapy versus radiotherapy, Outcome 5 Thrombocytopenia G3+.

Analysis 3.6

Comparison 3 Chemotherapy versus radiotherapy, Outcome 6 Infection G3+.

Analysis 3.7

Comparison 3 Chemotherapy versus radiotherapy, Outcome 7 Fatigue/asthenia G3+.

Analysis 3.8

Comparison 3 Chemotherapy versus radiotherapy, Outcome 8 Nausea/vomiting G3+.

Analysis 3.9

Comparison 3 Chemotherapy versus radiotherapy, Outcome 9 Weight loss G3+.

Analysis 3.10

Comparison 3 Chemotherapy versus radiotherapy, Outcome 10 Neurological symptoms G3+.

Analysis 3.11

Comparison 3 Chemotherapy versus radiotherapy, Outcome 11 Seizures G3+.

Analysis 3.12

Comparison 3 Chemotherapy versus radiotherapy, Outcome 12 Elevated liver enzymes G3+.

Analysis 3.13

Comparison 3 Chemotherapy versus radiotherapy, Outcome 13 Cutaneous adverse event G3+.

Analysis 4.1

Comparison 4 Chemoradiotherapy versus radiotherapy, Outcome 1 Progression‐free survival.

Analysis 4.2

Comparison 4 Chemoradiotherapy versus radiotherapy, Outcome 2 Neutropenia G3+.

Analysis 4.3

Comparison 4 Chemoradiotherapy versus radiotherapy, Outcome 3 Thrombocytopenia G3+.

Analysis 4.4

Comparison 4 Chemoradiotherapy versus radiotherapy, Outcome 4 Lymphopenia G3+.

Analysis 4.5

Comparison 4 Chemoradiotherapy versus radiotherapy, Outcome 5 Leucopenia G3+.

Analysis 4.6

Comparison 4 Chemoradiotherapy versus radiotherapy, Outcome 6 Anaemia G3+.

Analysis 4.7

Comparison 4 Chemoradiotherapy versus radiotherapy, Outcome 7 Treatment toxicity G3+.

Analysis 5.1

Comparison 5 Other + chemoradiotherapy versus chemoradiotherapy, Outcome 1 Progression‐free survival.

Analysis 5.2

Comparison 5 Other + chemoradiotherapy versus chemoradiotherapy, Outcome 2 Thromboembolic events G3+.

Analysis 6.1

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 1 Progression‐free survival.

Analysis 6.2

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 2 Thromboembolic events G3+.

Analysis 6.3

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 3 Haematological events G3+.

Analysis 6.4

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 4 Infections G3+.

Analysis 6.5

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 5 Fatigue G3+.

Analysis 6.6

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 6 Seizures G3+.

Analysis 6.7

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 7 Headaches G3+.

Analysis 6.8

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 8 Neuropsychiatric events G3+.

Analysis 6.9

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 9 Neurological events G3+.

Analysis 6.10

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 10 Hypertension G3+.

Analysis 6.11

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 11 Cutaneous adverse events G3+.

Analysis 6.12

Comparison 6 Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy), Outcome 12 Gastrointestinal events G3+.

Summary of findings for the main comparison. Summary of findings on overall survival comparing treatments to supportive care only

Estimates of effects, certainty assessment and rankings of different treatment options compared with supportive care only on overall survival in elderly people with glioblastoma
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: radiotherapy with 40 Gy in 15 fractions (RT40); radiotherapy with 60 Gy in 30 fractions (RT60); chemoradiotherapy (CRT); chemoradiotherapy plus bevacizumab (BEV_CRT); radiotherapy plus bevacizumab (BEV_RT); temozolomide (TMZ) Comparison: supportive care only Outcome: overall survival
All intervention options (7 RCTs; 1540 participants in total)*	Relative effect (network estimate) ** (95% CI)	Illustrative absolute effects for death at 6 months (95% CIs)	Certainty of the evidence (GRADE)	Ranking¥
Supportive care only*** (1 RCT, 81 participants)	Reference comparator	76 per 100***	Reference comparator	7.0 (worst)
RT60 (5 RCTs; 713 participants)	HR 0.47 (0.29 to 0.76)	49 per 100 (34 to 66)	⊕⊕⊕⊝ Moderate¹	5.0
BEV_RT (1 RCT; 75 participants)	HR 0.48 (0.23 to 1.00)	50 per 100 (28 to 76)	Not graded²	4.7
RT40 (4 RCTs; 930 participants)	HR 0.44 (0.25 to 0.77)	47 per 100 (30 to 67)	⊕⊕⊝⊝ Low³	4.3
TMZ (3 RCTs, 538 participants)	HR 0.42 (0.25 to 0.71)	45 per 100 (30 to 64)	⊕⊕⊝⊝ Low³	3.8
CRT (2 RCTs; 635 participants)	HR 0.30 (0.17 to 0.53)	35 per 100 (22 to 53)	Not graded²	1.8
BEV_CRT (1 RCT; 73 participants)	HR 0.25 (0.11 to 0.54)	30 per 100 (15 to 54)	Not graded²	1.4
NMA‐'Summary of findings' table definitions Estimates are reported as HR: hazard ratio, CI: confidence interval,NMA: network meta‐analysis. * This refers to the number of studies in the network evaluating the given intervention and the number of participants involved in these studies. All NMA effect estimates in this 'Summary of findings' table are derived 100% from indirect evidence, except for the comparison of RT60 versus supportive care, which was directly compared in one study. Where there was no common comparator for the comparison we did not grade the certainty of the evidence. * The assumed effect of supportive care is based on Keime‐Guibert 2007 data, which may be slightly overestimated because a high proportion of patients underwent biopsy only. ¥These rankings do not take into account the certainty of the evidence and should be interpreted with caution. The estimates of ungraded evidence are very uncertain.
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.
¹ Downgraded ‐1 as the evidence was derived from a single small study. ² There was no common comparator for the comparison (i.e. the intervention was not connected in a loop in the evidence network), therefore we did not grade the certainty of the evidence. ³ Contributing direct evidence was of moderate or low certainty. Abbreviations BEV_CRT; chemoradiotherapy plus bevacizumab; CI: confidence interval; CRT; chemoradiotherapy; HR: hazard ratio: RT40; radiotherapy (40 Gy in 15 fractions); RT60; radiotherapy (60 Gy in 30 fractions); SC; supportive care; TMZ; temozolomide; TTF_AC; tumour treating fields plus adjuvant chemotherapy) after concomitant CRT)

Summary of findings for the main comparison. Summary of findings on overall survival comparing treatments to supportive care only

Summary of findings 2. Summary of findings on overall survival comparing treatments to hypofractionated radiotherapy

Estimates of effects and certainty assessments compared with hypofractionated radiotherapy (40 Gy/15 fractions) on overall survival in elderly people with newly diagnosed glioblastoma
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: chemoradiotherapy (CRT); chemoradiotherapy plus bevacizumab (BEV_CRT); radiotherapy plus bevacizumab (BEV_RT); temozolomide (TMZ) Comparison: hypofractionated radiotherapy (RT 40) Outcome: overall survival
All intervention options (7 RCTs; 1540 participants in total)*	Relative effect (network estimate) ** (95% CI)	Illustrative absolute effects for death at 12 months (95% CIs)	Certainty of the evidence (GRADE)
RT 40 (4 RCTs; 930 participants)	Reference comparator	78 per 100***	Reference comparator
BEV_RT (1 RCT; 75 participants)	HR 1.08 (0.66 to 1.78)	81 per 100 (63 to 93)	⊕⊕⊝⊝ Low¹
TMZ (3 RCTs, 538 participants)	HR 0.95 (0.71 to 1.26)	76 per 100 (66 to 85)	⊕⊕⊝⊝ Low¹
CRT (2 RCTs; 635 participants)	HR 0.67 (0.56 to 0.80)	64 per 100 (57 to 70)	⊕⊕⊕⊕ High
BEV_CRT (1 RCT; 73 participants)	HR 0.56 (0.31 to 0.99)	57 per 100 (37 to 78)	⊕⊕⊕⊝ Moderate²
For the comparison with standard radiotherapy (60 Gy/30 fractions) see summary of findings Table 3.
NMA‐'Summary of findings' table definitions Estimates are reported as HR: hazard ratio. CI: confidence interval. *The assumed absolute effect of RT40 is based on Perry 2017 data. NMA: network meta‐analysis 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.
¹Contributing direct evidence was of low certainty (study design limitations and imprecision). ²Contributing direct evidence was of high or moderate certainty.

Summary of findings 2. Summary of findings on overall survival comparing treatments to hypofractionated radiotherapy

Summary of findings 3. Summary of findings on overall survival comparing treatments to standard radiotherapy

Estimates of effects and certainty assessment compared with standard radiotherapy (60 Gy/30 fractions) on overall survival in elderly people with newly diagnosed glioblastoma
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: radiotherapy with 40 Gy in 15 fractions (RT 40); chemoradiotherapy (CRT); chemoradiotherapy plus bevacizumab (BEV_CRT); radiotherapy plus bevacizumab (BEV_RT); temozolomide (TMZ); Comparison: standard radiotherapy (RT 60) Outcome: overall survival
All intervention options (7 RCTs; 1540 participants in total)*	Relative effect (network estimate) ** (95% CI)	Certainty of the evidence (GRADE)
RT 60 (5 RCTs; 713 participants)	Reference comparator	Reference comparator
RT 40 (4 RCTs; 930 participants)	HR 0.94 (0.72 to 1.23)	⊕⊕⊝⊝ Low¹
BEV_RT (1 RCT; 75 participants)	HR 1.01 (0.58 to 1.79)	⊕⊝⊝⊝ Very low^1,2
TMZ (3 RCTs, 538 participants)	HR 0.89 (0.71 to 1.11)	⊕⊝⊝⊝ Very low³
CRT (2 RCTs; 635 participants)	HR 0.63 (0.46 to 0.87)	⊕⊕⊝⊝ Low
BEV_CRT (1 RCT; 73 participants)	HR 0.52 (0.28 to 0.98)	Not graded
NMA‐'Summary of findings' table definitions Estimates are reported as HR: hazard ratio. CI: confidence interval. For assumed median survival times and absolute effect estimates, please refer to summary of findings Table for the main comparison. NMA: network meta‐analysis 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.
¹ Contributing direct evidence was of low certainty (study design limitations and imprecision). ² Downgraded for imprecision. ³ Contributing direct evidence of very low certainty (imprecision, study design limitations and inconsistency).

Summary of findings 3. Summary of findings on overall survival comparing treatments to standard radiotherapy

Summary of findings 4. Summary of findings on quality of life

The effect of different treatment comparisons for newly diagnosed glioblastoma in the elderly on quality of life
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: one treatment option Comparison: an alternative treatment option
Comparison	Narrative summary of evidence	No of studies (elderly participants)	Certainty of the evidence (GRADE)	Comments
RT vs supportive care only	Evidence on relative HRQoL is very uncertain.	1 study (59 participants at 30 day and 26 at 135 day time points, respectively)	⊕⊝⊝⊝ Low^1,2	The study reported that global assessments of deterioration over time also did not differ significantly between the two groups. The dropout rate was high and unbalanced so findings were difficult to interpret.
Hypofractionated RT vs standard RT (60Gy/30 fractions)	There may be little difference in HRQoL scores between hypofractionated and standard fractionation schedules in the 6 to 8 weeks following treatment.	2 studies (participants numbered 85 and 24, respectively, at the furthest time point)	⊕⊝⊝⊝ Low^1,2	One study compared a 25 Gy schedule with a 40 Gy schedule; the other compared a 40 Gy schedule with a standard 60 Gy schedule.
TMZ vs RT	No significant differences in global QoL scores at 3, 6, or 12 month measurements. However, there was a significant difference in discomfort from communication deficits, which were greatest for those receiving RT who died at between 6 and 12 months (P = 0·002).	1 study (306 participants)	⊕⊕⊕⊝ Moderate³	Evidence was not downgraded for attrition because data were reported for 82% of participants in each group for this outcome.
CRT vs RT	Authors reported that quality of life was similar in the two treatment groups. See Comments.	1 study (562 participants; number with QOL data was unclear)	Not graded	Investigators noted that attrition impacted the quantity of data. They conducted analyses using time to deterioration (with deterioration defined as a 10‐point decrease in the score on the function domain or a 10‐point increase in the score on the symptom domain) and plotted QoL scores over time. They reported that "There were no other clinically important differences between trial groups, which supports our observation that quality of life was similar in the two treatment groups."
BEV_CRT vs CRT	This was reported for the overall trial but not for elderly subgroup specifically.	1 study (73 participants)	Not graded	The authors reported significantly delayed deterioration in HRQoL scores in favour of BEV_CRT across five main HRQoL domains (global health, communication, social functioning, motor function, physical functioning). When progression of disease was removed as a deterioration event, the time to clinically significant deterioration or death remained statistically significant for communication, social functioning and global health.
BEV_RT vs RT	Investigators reported that "before progression, no differences were detected for individual scales in a generalized linear mixed model, except for less favourable values in arm A (BEV) for global health (P=0.048) and pain (P=0.027)".	1 study (75 participants)	Not graded	In the publication, global health was reported in a forest plot along with individual HRQoL items, such as cognitive functioning, emotional functioning and pain, measured with EORTC QLQ‐C30/BN20 scales. No other data were provided or obtained.
RIN_CRT vs control/CRT	This was reported for overall trial but not for elderly subgroup specifically.	1 study (174 participants)	Not graded	Investigators reported no significant differences between patients in their trial arms in any of the HRQoL measures.
TTF_AC vs CRT	This was reported for overall trial but not for elderly subgroup specifically.⁴	1 study (134 participants)	Not graded	There was no significant difference in HRQoL reported between the trial arms, except for itchy skin which was more prominent in the TTFields arm at 3,6 and 9 months (P = 0.005, P = 0008, P = 0.04). There was no significant difference at 12 months.
IRI_BEV_RT vs CRT	This was reported for overall trial but not for elderly subgroup specifically.	1 study (34 participants)	Not graded	There was no significant difference between the treatment arms.
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.
¹Sparse data from single studies [‐1] ² Serious risk of bias from attrition [‐1] ³ Downgraded because data were presented graphically and effects could not be estimated. ⁴ In this trial (Stupp 2017a) TTF_AC was compared with adjuvant chemotherapy (TMZ) only, after both arms had received concomitant CRT. Abbreviations: BEV_CRT: chemoradiotherapy plus bevacizumab; BEV_RT: radiotherapy plus bevacizumab; CI: confidence interval; CRT: chemoradiotherapy; HR: hazard ratio; IRI: irinotecan; RT: radiotherapy; TMZ: temozolomide; TTF_AC: tumour treating fields with adjuvant chemotherapy (after concomitant CRT) (TTF_AC); RIN_CRT: rindopepimut after CRT.

Summary of findings 4. Summary of findings on quality of life

Summary of findings 5. Summary of findings on progression‐free survival

The effect of different treatment comparisons for newly diagnosed glioblastoma in the elderly on progression‐free survival
Patient or population: elderly people with newly diagnosed glioblastoma Interventions: one treatment option Comparison: an alternative treatment option
Progression‐free survival	Relative effect (95% CI)	No of studies (elderly participants)	Certainty of the evidence (GRADE)	Comments
RT vs supportive care only	HR 0.28 (0.17 to 0.46)	1 study (81 participants)	⊕⊕⊕⊝ Moderate¹	Median time to progression was 3.5 months in the RT arm vs 1.3 months in the supportive care arm of this study (Keime‐Guibert 2007).
Hypofractionated RT vs standard RT (60Gy/30 fractions)	HR not reported	1 study (64 participants)	Not graded	This study (Roa 2015) reported that median progression‐free survival showed no statistically significant difference between arms (4.2 v 4.2 months in arms 1 and 2, respectively; P = 0.716).
TMZ vs RT	HR 1.15 (0.92 to 1.44)	1 study (373 participants)	⊕⊕⊝⊝ Low^2,3	No additional comments.
CRT vs RT	HR 0.50 (0.41 to 0.61)	1 study (562 participants)	⊕⊕⊕⊕ High	No additional comments.
BEV_CRT vs CRT	HR 0.78 (0.46 to 1.32)	1 study (73 participants)	⊕⊕⊝⊝ Low^1,2	In this study (AVAglio 2014), BEV_CRT did not increase overall survival either relative to CRT alone for elderly patients.
BEV_RT vs RT	HR 0.46 (0.27 to 0.78)	1 study (75 participants)	⊕⊕⊕⊝ Moderate¹	Despite delaying disease progression in this study (ARTE 2018), BEV_RT did not increase overall survival.
RIN_CRT vs CRT	Not reported separately for elderly subgroup.
TTF_AC vs CRT	Not reported separately for elderly subgroup.
IRI_BEV_RT vs CRT	Not reported separately for elderly subgroup.
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.
¹Sparse data from single studies [‐1] ² Serious risk of bias from attrition [‐1] ³ Serious imprecision Abbreviations:BEV_CRT: chemoradiotherapy plus bevacizumab; BEV_RT: radiotherapy plus bevacizumab; CI: Confidence interval; HR: hazard ratio;IRI: irinotecan; RT: radiotherapy; CRT: chemoradiotherapy; TMZ: temozolomide; TTF_AC: tumour treating fields with adjuvant chemotherapy (after concomitant CRT) (TTF_AC); RIN_CRT: rindopepimut after CRT.

Summary of findings 5. Summary of findings on progression‐free survival

Summary of findings 6. Summary of findings on severe adverse events

The effect of different treatments for newly diagnosed glioblastoma on severe adverse events
Patient or population: elderly people with newly diagnosed glioblastoma Settings: any Intervention: one treatment option Comparison: an alternative treatment option
Comparison	Narrative summary of evidence	No of studies (elderly participants)	Certainty of the evidence (GRADE)	Comments
RT vs supportive care only	Not reported.	NA	NA	None
Hypofractionated RT vs standard RT (60Gy/30 fractions)	This outcome was only reported in one small study and there were no instances of grade 3 or higher treatment‐related toxicity.	1 study (61)	Not graded	None
TMZ vs RT	TMZ probably increases the risk of thromboembolic and haematological adverse events	1 study (373)	⊕⊕⊕⊝ Moderate¹	Thrombocytopenia occurred in 24 TMZ vs 8 RT participants (RR 2.74, 95% CI 1.26 to 5.94) and lymphocytopenia occurred in 16 TMZ vs 2 RT participants (RR 7.30, 95% CI 1.70 to 31.31).
CRT vs RT	CRT probably increases the risk of haematological adverse events	1 study (562)	⊕⊕⊕⊝ Moderate¹	Neutropenia occurred in 22 CRT vs 2 RT participants (RR 10.30; 95% CI 2.45 to 43.34); throbocytopenia occurred in 30 CRT vs 1 RT participants (RR 28.56, 95% 3.92 to 207.86); lymphocytopenia occurred in 73 CRT vs 26 RT participants (RR 2.65, 95% CI 1.75 to 4.01); leucopenia occurred in 19 CRT vs 1 RT participant (RR 18.16, 96% CI 2.45 to 124.64); and anaemia occurred in 3 CRT vs 0 RT participants (RR 6.69, 95% CI 0.35 to 128.88).
BEV_CRT vs CRT	BEV_CRT probably increases the risk of grade 3+ thromboembolic events.	1 study (73)	⊕⊕⊕⊝ Moderate¹	Other adverse events data were not available for elderly subgroup specifically. For the larger study sample, Grade 3+ cerebral bleeding events (2.0% versus 0.9%) and wound healing events (3.3% vs 1.6%) were higher in the BEV plus CRT arm versus CRT alone. There were also higher rates of Grade 3+ thrombocytopenia (15% vs 9.8%) and infection rates (12.8% versus 7.8%) in the BEV_CRT arm.
BEV_RT vs RT	There was little or no difference in thromboembolic, haematological, and other severe adverse events reported in this one small study.	1 study (75)	⊕⊕⊝⊝ Low¹	An example of these non‐statistically significant findings for thrombocytopenia are that this SAE occurred in 8 BEV_RT vs 2 RT participants (RR 2.00, 95% CI 0.46 vs 8.73). Haematological SAEs occurred in 2 vs 0 participants, respectively (RR 2.55, 95% CI 0.13 vs 51.17).
RIN_CRT vs CRT	This was reported for overall trial but not for elderly subgroup specifically.	1 study (174 participants)	Not graded	The most common severe adverse events for the experimental (rindopepimut) versus control arm of the trial were: thrombocytopenia (9% vs 6%), fatigue (2% vs 5%), brain oedema (2% vs 3%), seizure (2% vs 2%) and headache (2% vs 3%). There was one death, secondary to pulmonary embolism, that was assessed as potentially related to the treatment in the experimental arm.
TTF_AC vs AC	This was reported for overall trial but not for elderly subgroup specifically.	1 study (134 participants)	Not graded	Overall, it was reported that there was no significant increase in rates of severe adverse events when TTF were added to adjuvant chemotherapy (48% vs 44%, P = 0.58).
IRI_BEV_RT vs CRT	This was reported for overall trial but not for elderly subgroup specifically.	1 study (34 participants)	Not graded	Overall, rates of severe adverse events were 72% in the experimental arm and 84% in the CRT arm. In the experimental arm, severe vascular events were most common (11.8%) and two cerebral haemorrhages occurred (one fatal). For patients in the CRT arm, severe haematological toxicity was most common (18.2%).
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.
¹ Downgraded ‐2 for sparse data from small single study and imprecision Abbreviations: BEV_CRT: chemoradiotherapy plus bevacizumab; BEV_RT: radiotherapy plus bevacizumab; CI: confidence interval; HR: hazard ratio; RT: radiotherapy; CRT: chemoradiotherapy; IRI: irinotecan; NA: not applicable; TMZ: temozolomide; TTF_AC: tumour treating fields with adjuvant chemotherapy (after concomitant CRT) (TTF_AC); RIN_CRT: rindopepimut after CRT.

Summary of findings 6. Summary of findings on severe adverse events

Table 1. Table of radiotherapy regimens used in included studies and biologically effective doses

Dose fractionation	EQD2 (Gy)	BED (Gy)
60 Gy/30 fractions	60	75
50 Gy/28 fractions	49	61
40 Gy/15 fractions	42	53
34 Gy/10 fractions	39	48
25 Gy/5 fractions	33	41
EQD2 and BED calculated for an alpha/beta of 8 EQD2 = equivalent dose; BED = biologically effective dose Gy = Gray

Table 1. Table of radiotherapy regimens used in included studies and biologically effective doses

Table 2. Characteristics of included studies

Study ID	Elderly definition (years)	No. analysed	Male gender (%)	Performance status	MGMT methylated/ unmethylated/ unknown (%)	Treatment option
Study ID	Elderly definition (years)	No. analysed	Male gender (%)	Performance status	MGMT methylated/ unmethylated/ unknown (%)	1	2	3	4
ARTE 2018	≥ 65	75	64	KPS ≥ 60	21%/73%/5%*	RT₄₀	RT₄₀+ BEV	‐	‐
AVAglio 2014	≥ 70	73	63	WHO 0‐2	26%/59%/24%	RT₆₀+ TMZ + maintenance	RT₆₀+ TMZ + maintenance + BEV	‐	‐
GLARIUS 2016	≥ 65	34	67^c	KPS ≥ 70	100% unmethylated	RT₆₀+TMZ	RT₆₀+BEV+IRI	‐	‐
Green 1983	≥ 65	107	65	KPS ≥ 70	‐	RT₆₀+carmustine	RT₆₀+steroid	RT₆₀+procarbazine	RT₆₀+BCNU+steroid
Keime‐Guibert 2007	≥ 70	85	63	KPS ≥ 70	‐	RT₅₀	Supportive care	‐	‐
Malmstrom 2012	≥ 70^a	123	59	WHO 0‐2	45%/55%/NR^e	RT₆₀	RT_30‐34	TMZ	‐
Perry 2017	≥ 65	562	61	ECOG 0‐2	47%/53%/NR^f	RT₄₀	RT₄₀ + TMZ + maintenance TMZ	‐	‐
Roa 2004	≥ 60^b	95	47	KPS ≥ 50	‐	RT₆₀	RT₄₀
Roa 2015	≥ 65^c	61	58	KPS ≥ 50	‐	RT₄₀	RT₂₅	‐	‐
Stupp 2017a	≥ 65	134	68	KPS ≥ 70	37%/53%/9%	RT₆₀+TMZ + maintenance TMZ	RT₆₀+TMZ + maintenance TMZ + TTF	‐	‐
Weller 2017	≥ 65	96	63^d	ECOG 0‐2	34%/59%/7%	RT₆₀+TMZ	RT₆₀+ TMZ + maintenance TMZ + RIN	‐	‐
Wick 2012	≥ 65	373	47	KPS ≥ 60	20%/36%/44%	RT₆₀	TMZ	‐	‐
BEV: bevacizumab; ECOG: Eastern Cooperative Oncology Group; Gy: Grays; IRI: irinotecan; KPS: Karnofsky performance score; RIN: rindopepimut; RT: radiotherapy; TTF: tumour treating fields; TMZ: temozolomide; WHO: World Health Organization ^a The whole sample (n = 291) comprised participants ≥ 60 years. The median age was 70 years for all study groups, ranging between 60 and 88 years. ^bMedian age was approximately 72 with a standard deviation of 5 years ^c Data for the ≥ 65 year age group were reported in the Guedes de Castro 2017 substudy report. ^d Gender data specific to the elderly subgroup were not reported separately ^e For approximately 70% of participants with MGMT data available ^f For approximately 63% of total participants with MGMT data available

Table 2. Characteristics of included studies

Table 3. Performance scores

Karnofsky Status	Karnofsky Grade	ECOG Grade	ECOG Status
Normal, no complaints	100	0	Fully active, able to carry on all pre‐disease performance without restriction
Able to carry on normal activities. Minor signs or symptoms of disease	90	1	Restricted in physically strenuous activity but ambulatory and able to carry out work of a light or sedentary nature, e.g. light house work, office work
Normal activity with effort	80	1	Restricted in physically strenuous activity but ambulatory and able to carry out work of a light or sedentary nature, e.g. light house work, office work
Care for self. Unable to carry on normal activity or to do active work	70	2	Ambulatory and capable of all self‐care but unable to carry out any work activities. Up and about more than 50% of waking hours
Requires occasional assistance, but able to care for most of his needs	60	2	Ambulatory and capable of all self‐care but unable to carry out any work activities. Up and about more than 50% of waking hours
Requires considerable assistance and frequent medical care	50	3	Capable of only limited self‐care, confined to bed or chair more than 50% of waking hours
Disabled. Requires special care and assistance	40	3	Capable of only limited self‐care, confined to bed or chair more than 50% of waking hours
Severly disabled. Hospitalisation indicated though death non imminent	30	4	Completely disabled. Cannot carry on any self‐care. Totally confined to bed or chair
Very sick. Hospitalisation necessary. Active supportive treatment necessary	20	4	Completely disabled. Cannot carry on any self‐care. Totally confined to bed or chair
Moribund	10	4	Completely disabled. Cannot carry on any self‐care. Totally confined to bed or chair
Dead	0	5	Dead
As published in Am J Clin. Oncol: Oken 1982

Table 3. Performance scores

Table 4. Table of outcomes reported in included studies

	Trial	OS	QoL	PFS	Severe adverse events	Cognitive impairments	Functional impairment	Fatigue
1	Green 1983	For 65+ age subgroup, number of deaths (overall, no time point specified) and death rate (number of deaths per 10 patient‐months) reported.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.
2	Roa 2004 (Elderly patients only in this trial, defined as age 60 years or over; mean age was 72 years with SD 5 years)	Median OS reported (ITT analysis), including HR and a KM curve. Percentage of patients alive at 6 months also reported (Table 1 and Figure 1 of main manuscript).	Low rates of FACT‐Br version 3 completion (45% overall) by patients precluded meaningful analysis. Protocol specified FACT‐Br completion at baseline, 3 weeks after starting RT, at the end of RT, and at 3‐month intervals thereafter. Table 2 of main manuscript.	Not done.	Not reported.	Not reported.	Difference in averaged KPS scores and change in KPS over time between the two groups (0‐6 months from start of RT). KPS at baseline, 3 weeks, 6 weeks, first and second follow up. Table 1, Table 2 and Figure 2 of main manuscript.	Not reported.
3	Keime‐Guibert 2007 (Elderly patients only ‐ aged 70 years or over).	Median OS reported (ITT analysis), including HR and KM curve.	QLQ‐C30 and QLQ‐BN20 and completion rate. Changes in mean score at baseline, day 30, day 60, day 90 and day 135.	Median PFS.	Tolerance of treatment reported but not clear which toxicity scoring used.	QLQ‐C30 includes cognitive. MDRS score at baseline, 60 days and 135 days. MMSE score. NPI (Neuropysch inventory).	KPS decline over time.	QLQ‐C30 includes fatigue.
4	Malmstrom 2012 (Patients in this trial all aged over 60 years old but reported data for 70+ years subgroup)	Median OS and 1 year survival percentage. As this study had three arms, survival analyses were done using three pairwise comparisons. A KM curve was presented for the overall patient population and a subgroup of patients aged 60‐70 years and aged over 70. HR reported for this outcome (Table 2). OS for patients with mMGMT and umMGMT disease also reported. The two radiotherapy arms combined for this analysis. In the supplementary appendix, OS and KM reported comparing the TMZ arm and the hypofractionated RT arm. These outcomes were reported for patients overall, a subgroup of patients aged 60‐70 and a subgroup of patients aged over 70. HR reported for these outcomes. 1 year OS probabilities also reported.	EORTC QLQ‐30v3 and QLQ‐BN20 at baseline, 6 weeks, 3 months and 6 months. Mean change of score from baseline reported for each domain (Figure 4). No data reported for 6 month time‐point due to low completion rate.	Not reported.	WHO grading for all except nausea and vomiting (CTCAE v2), but only reported for overall sample and not elderly subgroup.	Reported as a category of a QoL assessment but only for overall sample and not elderly subgroup.	Not reported.	WHO grading system. Reported fatigue G2‐5, but only for overall sample and not elderly subgroup.
5	Wick 2012 (All elderly patients – aged over 65 years)	Median OS reported (ITT analysis) with HR and KM curve. OS differences between patients with mMGMT and uMGMT disease also reported (with HR and KM curve). OS survival (PP analysis) also reported but no HR or KM curve for these outcomes. Survival % at 6 months and 1 year. Also report proportions for death, or disease progression or death for each arm.	EORTC QLQ‐30 and BN20 used for HRQoL assessment. Results available for 82% of the patients (in appendix).	Median event free survival (rather than PFS) and also mMGMT and uMGMT. %EFS at 6 months and 1 year.	CTCAEv3 used to collect adverse event data.	MMSE performed at baseline, monthly during treatment and then every 3 months. Median score (with 95% CI) reported overall, before treatment and after primary treatment.	KPS at baseline.	Fatigue/asthenia G3‐4 reported as an adverse event.
6	AVAglio 2014 (comprises several reports) Chinot 2014 main report presents findings for whole sample, elderly subgroup reported in supplementary appendix. Saran 2016 provides information on safety outcomes. Taphoorn 2015 reported HR QoL outcomes for sample as a whole.	OS for 60‐69 and 70+ subgroups (found in supplementary appendix 2). No KM curves for these subgroups.	Taphoorn 2015 reports the results for QLQ‐C30 and QLQ‐BN20 for the whole sample (age >18), not for the elderly subgroup.	PFS for 60‐69 and 70+ subgroups (number and HR).	Mason 2014 is a separate publication about arterial adverse events and gives event rates for the over 65 subgroup.	Not for elderly subgroup.	Not for elderly subgroup.	Not for elderly subgroup.
7	Roa 2015 (comprises several reports) (Guedes de Castro 2017 reports on outcomes only for those patients aged over 65s and Fidarova 2015 presents quality of life results as a conference abstract).	Median OS reported (ITT analysis). Outcomes for elderly and frail, and elderly and not frail subgroups. No HR reported for OS outcomes. KM curves included in Guedes de Castro 2017 Figures 1‐4.	The mean scores from a combination of EORTC QLQC30 and QLQ‐BN20 reported at baseline, 4 weeks and 8 weeks. Categorical scales were transformed to linear scalers from 0‐100. Difference between mean scores between treatment groups reported in Guedes de Castro 2017 Table 3.	Median PFS reported (ITT). No KM or HR for elderly only patients.	Adverse events were recorded weekly during RT, 4 weeks after RT and every 3 months until tumour progression using CTCAE v3. AEs only reported in the case of a grade 3 or greater toxicity.	MMSE at baseline.	KPS at baseline.	Fatigue mentioned in Fidarova 2015 abstract for overall trial population.
8	GLARIUS 2016 (comprises several reports) Herrlinger 2016 reports the main findings for the whole sample, aged from 18+ years. Kebir 2016 is an abstract which reports specifically on the differences in OS between younger and older patients in the trial.	Median OS for 65+ subgroup in both arms reported (modified ITT analysis). No HR or KM curve available for the elderly cohort. yes	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.
9	Perry 2017 (All elderly patients – aged 65 years old or over).	Median OS reported (ITT). KM curve and HR reported for this outcome. Median OS also reported for subgroups of patients aged 65‐70 years old, 71‐75 and 76+. Also reported OS rate at 12, 18 and 24 months for all patients and for patients with umMGMT and mMGMT disease.	QLQC30 and BN20. Up to 18 months post treatment. Baseline/1 week/3 week reported. Time to deterioration in QoL reported.	Median PFS.	CTCAE v3 used for adverse event reporting.	MMSE at baseline.	ECOG at baseline.	Not separately reported.
10	Weller 2017 (Trial not restricted to elderly patients only. Includes patients aged 18 years +).	Report OS (events per patient) for patients aged 65+ subgroup. The subgroup if further divided into those who have minimal residual disease (MRD) and significant residual disease (SRD). There is a HR given for patients aged 65+ in the MRD subgroup and in the SRD subgroup separately. There is no HR reported for the 65+ group overall. No KM curves for the elderly subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.	Not for 65+ subgroup.
11	Stupp 2017a (Trial not restricted to elderly patients only. Includes patients aged 18 years +). Taphoorn 2018 is separate publication which reports on HRQoL outcomes.	Median OS for 65+ subgroup. HR and KM curve reported for this outcome in the 65+ subgroup. The proportion of patients in each arm of the trial who were alive at the end of the study also reported.	HR QoL was measured using EORTC QLQ‐C30 and BN20 questionnaires at baseline and every 3 months for up to 12 months. Mean QoL scores and mean change from baseline reported. Outcomes not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.	Not reported separately for elderly subgroup.
12	ARTE 2018 (All elderly patients in the trial – aged 65 years or over)	Median OS (ITT and PP) and 1 year survival rate. Also reported OS depending on molecular panel subtype. No HR reported for median OS differences. KM curves are shown.	Reported median deterioration free survival from baseline. Individual functional and symptom scores from EORTC QLQ‐C30/BN20 before tumour progression analysed in a generalised linear model that controlled for time treatment interactions.	Median PFS. ITT and PP. And %PFS at 6 months. Also had independent central review for some patients. (66 with MRI available). Also looked at PFS per molecular panel subtype.	Yes, reported in supplementary material. G3‐5 fatigue, seizures, headaches, other neuro, neuropysch, haematological, arterial hypertension, thromboembolic, all infections, cutaneous, GI.	MMSE at baseline and serial measurements (0,7,19 weeks).	KPS at baseline.	Fatigue reported as a category of QoL and as an adverse events.
CTCAE: Common Terminology Criteria for Adverse Events; EORTC: European Organisation for Research and Treatment of Cancer (EORTC) QLQ‐C30 or QLQ‐BN20 (specific for brain cancer); FACT‐Br: Functional Assessment of Cancer Therapy scale [specific for brain cancer]); HR: hazard ratio; HRQoL: Health‐related quality of life; ITT: Intention to treat; mITT: modified ITT; KM curve: Kaplan Meier curve; KPS: Karnofsky performance status; MMSE: mini mental state examination; mMGMT: Methylated MGMT; umMGMT: unmethylated MGMT; PFS: progression free survival; PP: per protocol; OS: overall survival.

Table 4. Table of outcomes reported in included studies

Table 5. Median survival associated with treatment options evaluated in included studies

Study ID	Age (years) and performance status	Treatment arm	Overall Survival (OS)			Progression Free Survival (PFS)
Study ID	Age (years) and performance status	Treatment arm	Median OS (months)	lower 95% CI (months)	upper 95% CI (months)	Median PFS (months)	lower 95% CI (months)	upper 95% CI (months)
Green 1983	Age ≥ 65 KPS ≥ 70%	4 arms; see Table 1	‐	‐	‐	‐	‐	‐
Keime‐Guibert 2007	Age ≥ 70 KPS ≥ 70%	Supportive care	3.9	3.1	5.0	1.3	1.0	1.8
Keime‐Guibert 2007	Age ≥ 70 KPS ≥ 70%	RT (50 Gy/28 fractions/5‐6 weeks) + supportive care	6.8	5.9	8.1	3.5	2.5	5.2
Roa 2004	Age ≥ 60 KPS ≥ 50%	RT (40 Gy/15 fractions/3 weeks)	5.6	‐	‐	‐	‐	‐
Roa 2004	Age ≥ 60 KPS ≥ 50%	RT (60 Gy/30 fractions/6 weeks)	5.1	‐	‐	‐	‐	‐
Roa 2015^a (elderly and/or frail)	Age ≥ 65 KPS 50‐70%	RT (25 Gy/5 fractions/1 week)	6.8	4.5	9.1	4.3	2.6	5.9
Roa 2015^a (elderly and/or frail)	Age ≥ 65 KPS 50‐70%	RT (40 Gy/15 fractions)	6.2	4.7	7.7	3.2	0.1	6.3
Roa 2015 (elderly and frail)	Age ≥ 65 KPS ≥ 50%	RT (25 Gy/5 fractions/1 week)	7.5	5.3	9.7	‐	‐	‐
Roa 2015 (elderly and frail)	Age ≥ 65 KPS ≥ 50%	RT (40 Gy/15 fractions)	6.7	4.5	8.9	‐	‐	‐
Roa 2015^a(elderly and non‐frail)	Age ≥ 65 KPS ≥ 80%	RT (25 Gy/5 fractions/1 week)	8.0	5.9	10.0	‐	‐	‐
Roa 2015^a(elderly and non‐frail)	Age ≥ 65 KPS ≥ 80%	RT (40 Gy/15 fractions/3 weeks)	8.0	5.3	10.3	‐	‐	‐
Malmstrom 2012	Age ≥ 70 WHO 0‐2	RT (60 Gy/30 fractions/6 weeks)	5.2	4.0	6.3	‐	‐	‐
		RT (34 Gy/10 fractions/2 weeks)	7.0	5.2	8.8	‐	‐	‐
		TMZ	9.0	6.2	11.8
Malmstrom 2012 (unmethylated)	Age ≥ 70 WHO 0‐2	RT (any schedule)	7.0	5.8	8.3	‐	‐	‐
Malmstrom 2012 (unmethylated)	Age ≥ 70 WHO 0‐2	TMZ	6.8	5.9	7.7	‐	‐	‐
Malmstrom 2012 (methylated)	Age ≥ 70 WHO 0‐2	RT (any schedule)	8.2	6.6	9.9	‐	‐	‐
Malmstrom 2012 (methylated)	Age ≥ 70 WHO 0‐2	TMZ	9.7	8.0	11.4	‐	‐	‐
Wick 2012^b	Age ≥ 65 KPS ≥ 60%	RT (60Gy/30 fractions/6 weeks)	9.6	8.2	10.8	4.7	4.2	5.2
Wick 2012^b	Age ≥ 65 KPS ≥ 60%	TMZ	8.6	7.3	10.2	3.3	3.2	4.1
Wick 2012 (unmethylated)	Age ≥ 65 KPS ≥ 60%	RT (60Gy/30 fractions/6 weeks)	10.4	8.0	11.6	4.6	3.7	6.3
Wick 2012 (unmethylated)	Age ≥ 65 KPS ≥ 60%	TMZ	7.0	5.7	8.7	3.3	3.0	3.5
Wick 2012 (methylated)	Age ≥ 65 KPS ≥ 60%	RT (60Gy/30 fractions/6 weeks)	9.6	6.4	not reached	4.6	4.2	5.0
Wick 2012 (methylated)	Age ≥ 65 KPS ≥ 60%	TMZ	not reached	10.1	not reached	8.4	5.5	11.7
Perry 2017	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks)	7.6	7.0	8.4	3.9	3.5	4.3
Perry 2017	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks) + TMZ + maintenance TMZ	9.3	8.3	10.3	5.3	4.6	6.2
Perry 2017^c (unmethylated)	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks)	7.9	‐	‐	‐	‐	‐
Perry 2017^c (unmethylated)	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/ 3 weeks) + TMZ + maintenance TMZ	10	‐	‐	‐	‐	‐
Perry 2017^c (methylated)	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks)	7.7	‐	‐	‐	‐	‐
Perry 2017^c (methylated)	Age ≥ 65 ECOG 0‐2	RT (40 Gy/15 fractions/3 weeks) + TMZ+ maintenance TMZ	13.5	‐	‐	‐	‐	‐
AVAglio 2014	Age ≥ 70 WHO 0‐2	RT (60 Gy/30 fractions) + TMZ + maintenance TMZ	‐	‐	‐	‐	‐	‐
AVAglio 2014	Age ≥ 70 WHO 0‐2	RT (60 Gy/30 fractions) + TMZ + BEV + maintenance	‐	‐	‐	‐	‐	‐
Stupp 2017a^d	Age ≥ 65 KPS ≥ 70%	RT (±60 Gy/30 fractions/6 weeks) + TMZ + maintenance TMZ	13.7	7.6	24.8	‐	‐	‐
Stupp 2017a^d	Age ≥ 65 KPS ≥ 70%	RT (±60 Gy/30 fractions/6 weeks) + TMZ + maintenance TMZ + TTFields	17.4	9.0	31.5	‐	‐	‐
ARTE 2018	Age ≥ 65 KPS ≥ 60%	RT (40 Gy/15 fractions/3 weeks)	12.2	9.2	15.2	4.8	3.0	6.6
ARTE 2018	Age ≥ 65 KPS ≥ 60%	RT (40 Gy/15 fractions/3 weeks) + BEV	12.1	10.2	14.0	7.6	6.2	9.0
GLARIUS 2016^e (unmethylated)	Age ≥ 65 KPS ≥ 70%	RT (60 Gy/30 fractions/6 weeks) + TMZ + maintenance	17.5	‐	‐	‐	‐	‐
GLARIUS 2016^e (unmethylated)	Age ≥ 65 KPS ≥ 70%	RT (60 Gy/30 fractions/6 weeks) + BEV + IRI + maintenance	13.4	‐	‐	‐	‐	‐
Weller 2017	Age ≥ 65 KPS ≥ 60%	RT (60 Gy/30 fractions/6 weeks) + TMZ + maintenance	‐	‐	‐	‐	‐	‐
Weller 2017	Age ≥ 65 KPS ≥ 60%	RT (60 Gy/30 fractions/6 weeks) + TMZ + maintenance + RIN	‐	‐	‐	‐	‐	‐
BEV: bevacizumab; ECOG: Eastern Cooperative Oncology Group; Gy: Grays; IRI: irinotecan; KPS: Karnofsky performance score; OS: overall survival; PFS: progression free survival; RIN: rindopepimut; RT: radiotherapy; TTF: tumour treating fields; TMZ: temozolomide; WHO: World Health Organization ^aData from the Guedes de Castro et al 2017 substudy report on patients ≥ 65 years only ^bThis study reported event free survival (EFS), not PFS. Findings by MGMT methylation status were reported by authors in the later (2017) publication. ^cFrom Perry 2012 abstract ^dMGMT promotor methylation status was not reported separately for the elderly subgroup. Median survival data were reported as time from randomisation not diagnosis. Randomisation in this trial occurred after concomitant chemoradiotherapy. ^eFrom substudy data reported in a conference proceeding abstract by Kebir et al, 2016.

Table 5. Median survival associated with treatment options evaluated in included studies

Table 6. CHEC list* for included economic studies

CHEC ITEM	Ghosh et al (2018)
Is the study population clearly described?	Y
Are competing alternatives clearly described?	Y
Is a well‐defined research question posed in answerable form?	Y
Is the economic study design appropriate to the stated objective?	Y
Is the chosen time horizon appropriate to include relevant costs and consequences?	N
Is the actual perspective chosen appropriate?	N
Are all important and relevant costs for each alternative identified	N
Are all costs measured appropriately in physical units?	Y
Are costs valued appropriately?	N
Are all important and relevant outcomes for each alternative identified?	N
Are all outcomes measured appropriately?	N
Are outcomes valued appropriately?	N
Is an incremental analysis of costs and outcomes of alternatives performed?	Y
Are all future costs and outcomes discounted appropriately?	N
Are all important variables, whose values are uncertain, appropriately subjected to sensitivity analysis?	N
Do the conclusions follow from the data reported?	N
Does the study discuss the generalizability of the results to other settings and patient/client groups?	N
Does the article indicate that there is no potential conflict of interest of study researcher(s) and funder(s)?	Y
Are ethical and distributional issues discussed appropriately?	N
*Evers 2005

Table 6. CHEC list* for included economic studies

Table 7. CHEERS checklist* of included studies

CHEERS Quality Checklist
Section of paper	Component	Reported on Page Number
Ghosh et al (2018)
Title and abstract	Identify the study as an economic evaluation or use more specific terms such as ‘‘cost‐effectiveness analysis’’, and describe the interventions compared.	114
Title and abstract	Provide a structured summary of objectives, perspective, setting, methods (including study design and inputs), results (including base case and uncertainty analyses), and conclusions.	114
Methods	Describe characteristics of the base case population and subgroups analysed, including why they were chosen.	Supplementary material
	State relevant aspects of the system(s) in which the decision(s) need(s) to be made.	Not reported
	Describe the perspective of the study and relate this to the costs being evaluated.	Not reported
	Describe the interventions or strategies being compared and state why they were chosen.	114‐115
	State the time horizon(s) over which costs and consequences are being evaluated and say why appropriate.	Partially, stated but not justified
	Report the choice of discount rate(s) used for costs and outcomes and say why appropriate.	Not reported
	Describe what outcomes were used as the measure(s) of benefit in the evaluation and their relevance for the type of analysis performed.	115
	Describe fully the design features of the single effectiveness study and why the single study was a sufficient source of clinical effectiveness data.	115‐116
	Describe approaches used to estimate resource use associated with the alternative interventions. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs.	Not reported
	Report the dates of the estimated resource quantities and unit costs. Describe methods for adjusting estimated unit costs to the year of reported costs if necessary. Describe methods for converting costs into a common currency base and the exchange rate.	115
	Describe and give reasons for the specific type of decision‐analytical model used. Providing a figure to show model structure is strongly recommended.	N/A
	Describe all structural or other assumptions underpinning the decision‐analytical model.	N/A
	Describe all analytical methods supporting the evaluation. This could include methods for dealing with skewed, missing, or censored data; extrapolation methods; methods for pooling data; approaches to validate or make adjustments (such as half cycle corrections) to a model; and methods for handling population heterogeneity and uncertainty.	Partial, uncertainty discussed.
Results	Report the values, ranges, references, and, if used, probability distributions for all parameters. Report reasons or sources for distributions used to represent uncertainty where appropriate. Providing a table to show the input values is strongly recommended.	Not reported
	For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost‐effectiveness ratios.	116‐118 & supplementary material
	Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact of methodological assumptions (such as discount rate, study perspective).	116
	If applicable, report differences in costs, outcomes, or cost‐effectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.	N/A
Discussion	Summarise key study findings and describe how they support the conclusions reached. Discuss limitations and the generalisability of the findings and how the findings fit with current knowledge.	118‐119
Other	Describe how the study was funded and the role of the funder in the identification, design, conduct, and reporting of the analysis. Describe other non‐monetary sources of support.	119
Other	Describe any potential for conflict of interest of study contributors in accordance with journal policy. In the absence of a journal policy, we recommend authors comply with International Committee of Medical Journal Editors recommendations.	119
*Evers 2005

Table 7. CHEERS checklist* of included studies

Table 8. Table of estimate effects and certainty ratings for overall survival

Comparison	Direct Evidence		Indirect Evidence		Network Evidence
Comparison	HR (95% CI)	Certainty	HR (95% CI)	Certainty	HR (95% CI)	Certainty
RT60* vs Supp Care	0.47 (0.29 to 0.76)	Moderate¹	Not estimable²	–	0.47 (0.29 to 0.76)	Moderate
RT40 vs Supp Care	–	–	0.44 (0.25 to 0.77)	Low³	0.44 (0.25 to 0.77)	Low
CRT vs Supp Care	–	–	–	–	0.30 (0.17 to 0.53)	Not graded⁴
TMZ vs Supp Care	–	–	0.42 (0.25 to 0.71)	Low³	0.42 (0.25 to 0.71)	Low
BEV_CRT vs Supp Care	–	–	–	–	0.25 (0.11 to 0.54)	Not graded⁴
BEV_RT vs Supp Care	–	–	–	–	0.48 (0.23 to 1.00)	Not graded⁴

CRT vs RT40	0.67 (0.56 to 0.80)	High	–	–	0.67 (0.56 to 0.80)	High
BEV_CRT vs RT40	–	–	0.56 (0.31 to 0.99)	Moderate⁵	0.56 (0.31 to 0.99)	Moderate
TMZ vs RT40**	0.72 (0.50 to 1.05)	Low⁶	–	–	0.95 (0.71 to 1.26)	Low
BEV_RT vs RT40	1.08 (0.65 to 1.78)	Low⁶	Not estimable²	–	1.08 (0.66 to 1.78)	Low

RT40 vs RT60	0.74 (0.55 to 1.01)	Low⁶	Not estimable²	–	0.94 (0.72 to 1.23)	Low
BEV_RT vs RT60	–	–	1.01 (0.58 to 1.79)	Very low⁷	1.01 (0.58 to 1.79)	Very low
BEV_CRT vs RT60	–	–	–	–	0.52 (0.28 to 0.98)	Not graded⁴
CRT vs RT60	–	–	0.63 (0.46 to 0.87)	Low⁸	0.63 (0.46 to 0.87)	Low
TMZ vs RT60	0.86 (0.68 to 1.09)	Very low⁹	–	–	0.89 (0.71 to 1.11)	Very low

BEV_RT vs CRT	–	–	1.61 (0.95 to 2.74)	Low¹⁰	1.61 (0.95 to 2.74)	Low
BEV_CRT vs CRT	0.83 (0.48 to 1.43)	Low⁶	Not estimable²	–	0.83 (0.48 to 1.44)	Low
TMZ vs CRT	–	–	1.42 (1.01 to 1.98)	Low¹⁰	1.42 (1.01 to 1.98)	Low

BEV_RT vs TMZ	–	–	1.14 (0.64 to 2.02)	Very low^10,11	1.14 (0.64 to 2.02)	Very low
BEV_CRT vs TMZ	–	–	–	–	0.59 (0.31 to 1.12)	Not graded⁴

BEV_CRT vs BEV_RT	–	–	–	–	0.52 (0.24 to 1.10)	Not graded⁴
¹ Evidence derived from a single small study ² Could not be estimated because the intervention was not connected via a loop in the evidence network ³ Contributing direct evidence was of moderate or low certainty ⁴ There was no direct evidence for this comparison, which did not connect via a common comparator, therefore the certainty of evidence was not graded. ⁵ Contributing direct evidence was of high or moderate certainty ⁶ Downgraded for study design limitations and imprecision ⁷ Contributing direct evidence was of low certainty; network estimate imprecise ⁸ Contributing direct evidence was of high or low certainty ⁹ Downgraded for imprecision, study design limitations and inconsistency ¹⁰ Contributing direct evidence was of high or low certainty ¹¹Downgraded for imprecision
RT50 (Keime‐Guibert 2007) coded as RT60. RT34 (Malmstrom 2012) coded as RT40. Abbreviations:* BEV_CRT = bevacizumab plus chemoradiotherapy; CI = confidence interval; ; CRT = chemoradiotherapy; RT40 = radiotherapy (40Gy in 15 fractions); RT60 = radiotherapy (60Gy in 30 fractions); Supp Care = supportive care; TMZ = temozolomide; TTF_AC = tumour treating fields with adjuvant chemotherapy (after concomitant chemotherapy)

Table 8. Table of estimate effects and certainty ratings for overall survival

Table 9. Overview of SUCRA rankings

Treatment option	Main NMA model	Sensitivity analysis A	Sensitivity analysis B	Sensitivity analysis C	Sensitivity analysis D
BEV_CRT	1.4	1.4	1.4	–	–
CRT	1.8	1.8	1.9	1.1*	1.1*
TMZ	3.8	3.8	3.7	2.8	2.8
RT40	4.3	4.3	4.2	3.3	3.3
BEV_RT	4.7	4.7	4.8	3.8	3.8
RT60	5.0	5.1	5.0	4.0	4.0
Supp_Care	7.0	7.0	7.0	6.0	6.0
*CRT40 BEV_CRT = chemoradiotherapy plus bevacizumab; CRT = chemoradiotherapy; RT40 = radiotherapy (40Gy in 15 fractions); RT60 = radiotherapy (60Gy in 30 fractions); Supp_Care = supportive care; TMZ = temozolomide; TTF_AC = tumour treating fields with adjuvant chemotherapy (after concomitant chemotherapy) Sensitivity analysis A: Keime‐Guibert 2007 study’s RT50 arm is coded as RT40; sensitivity analysis B: network without a loop due to exclusion of the Malmstrom 2012 TMZ vs RT40 arm to avoid duplication of data; sensitivity analysis C.1 and C.2: disconnected networks due to non‐pooling of CRT40 and CRT60 arms from different studies; sensitivity analysis D: disconnected network due to non‐pooling of CRT40 and CRT60 arms from different studies, and Keime‐Guibert 2007 study arm RT50 coded as RT40.

Table 9. Overview of SUCRA rankings

Table 10. Data extraction table for economic studies

Author(s), Year & Title

Type of Evaluations

Sources of effectiveness data

Sources of Cost Data

Sources of outcome valuations

Analytical approach

Ghosh 2018

Improved cost‐effectiveness of short‐course radiotherapy in elderly and/or frail patients with glioblastoma

Cost‐effectiveness analysis

Cost‐utility analysis

Effectiveness data was taken from a phase III randomised trial.

Direct unit medical costs were collected from each country participating in the trial (from 2015).

Restricted mean overall survival (RMOS)

Incremental cost‐effectiveness ratio (ICER).

Table 10. Data extraction table for economic studies

Table 11. Unit costs table for Included studies

Component	Study	Country	Estimated costs of resources used	Source	Currency	Average number of Resources Used Arm 1 / Arm 2		Source
Dexamethasone 4 mg tablet	Ghosh 2018	Belarus	0.27	Not reported	US Dollars (conversion not reported)	44	16	Trial Dataset
		Brazil (Porto Alegre)	0.06	‐ Not reported	US Dollars (conversion not reported)	44	60	Trial Dataset
		Georgia	0.3	Not reported	US Dollars (conversion not reported)	18	4	Trial Dataset
		India	0.01	Not reported	US Dollars (conversion not reported)	32	24	Trial Dataset
		Poland	1.36	Not reported	US Dollars (conversion not reported)	12	8	Trial Dataset
CT	Ghosh 2018	Belarus	269	Not reported	US Dollars (conversion not reported)	4	3	Trial Dataset
		Brazil (Porto Alegre)	119	Not reported	US Dollars (conversion not reported)	4	3	Trial Dataset
		Georgia	130	Not reported	US Dollars (conversion not reported)	1	1	Trial Dataset
		India	25	Not reported	US Dollars (conversion not reported)	1	1	Trial Dataset
		Poland	57	Not reported	US Dollars (conversion not reported)	4	1.5	Trial Dataset
MRI	Ghosh 2018	Belarus	330	Not reported	US Dollars (conversion not reported)	5	‐	Trial Dataset
		Brazil (Porto Alegre)	293	Not reported	US Dollars (conversion not reported)	4	‐	Trial Dataset
		Georgia	230	Not reported	US Dollars (conversion not reported)	1	3	Trial Dataset
		India	58.3	Not reported	US Dollars (conversion not reported)	3	1	Trial Dataset
		Poland	130	Not reported	US Dollars (conversion not reported)	3	3	Trial Dataset
RT (5 fractions)	Ghosh 2018	Belarus	600	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Brazil (Porto Alegre)	1046	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Georgia	2900	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		India	25	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Poland	3472	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
RT (15 fractions)	Ghosh 2018	Belarus	1800	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Brazil (Porto Alegre)	1985	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Georgia	3800	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		India	42	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset
		Poland	3472	Not reported	US Dollars (conversion not reported)	‐	‐	Trial Dataset

Table 11. Unit costs table for Included studies

Comparison 1. Radiotherapy (50 Gy) versus supportive care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 HRQOL Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.1 At 30 days	1	59	Mean Difference (IV, Random, 95% CI)	‐4.20 [‐6.33, ‐2.07]
1.2 At 60 days	1	45	Mean Difference (IV, Random, 95% CI)	‐4.70 [‐7.33, ‐2.07]
1.3 At 90 days	1	39	Mean Difference (IV, Random, 95% CI)	‐7.60 [‐11.03, ‐4.17]
1.4 At 135 days	1	26	Mean Difference (IV, Random, 95% CI)	10.70 [6.01, 15.39]
2 Cognition Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

2.1 At 30 days	1	59	Mean Difference (IV, Random, 95% CI)	‐0.40 [‐3.24, 2.44]
2.2 At 60 days	1	45	Mean Difference (IV, Random, 95% CI)	‐5.60 [‐9.20, 0.00]
2.3 at 90 days	1	39	Mean Difference (IV, Random, 95% CI)	‐21.0 [‐25.18, ‐16.82]
2.4 at 135 days	1	26	Mean Difference (IV, Random, 95% CI)	‐13.0 [‐18.84, ‐7.16]
3 Fatigue Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.1 At 30 days	1	59	Mean Difference (IV, Random, 95% CI)	2.10 [‐0.49, 4.69]
3.2 At 60 days	1	45	Mean Difference (IV, Random, 95% CI)	8.60 [4.77, 12.43]
3.3 At 90 days	1	39	Mean Difference (IV, Random, 95% CI)	11.70 [8.24, 15.16]
3.4 At 135 days	1	26	Mean Difference (IV, Random, 95% CI)	0.5 [‐6.68, 7.68]
4 Progression‐free survival Show forest plot	1	81	Hazard Ratio (Random, 95% CI)	0.28 [0.17, 0.46]

Comparison 1. Radiotherapy (50 Gy) versus supportive care

Comparison 2. Short‐course radiotherapy versus standard radiotherapy

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 HRQOL at 4 weeks Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.1 RT (40 Gy) vs RT (60 Gy)	1	85	Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
1.2 RT (25 Gy) vs RT (40 Gy)	1	40	Mean Difference (IV, Random, 95% CI)	3.40 [‐8.33, 15.13]
2 HRQOL at 8 weeks Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

2.1 RT (40 Gy) vs RT (60 Gy)	1	72	Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
2.2 RT (25 Gy) vs RT (40 Gy)	1	24	Mean Difference (IV, Random, 95% CI)	0.0 [‐13.58, 13.58]
3 Treatment toxicity G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

3.1 RT (40 Gy) vs RT (60 Gy)	1	61	Risk Ratio (M‐H, Random, 95% CI)	0.0 [0.0, 0.0]

Comparison 2. Short‐course radiotherapy versus standard radiotherapy

Comparison 3. Chemotherapy versus radiotherapy

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Progression‐free survival Show forest plot	1		Hazard Ratio (Random, 95% CI)	Subtotals only

1.1 TMZ vs RT (60 Gy)	1	373	Hazard Ratio (Random, 95% CI)	1.15 [0.92, 1.44]
2 Thromboembolic event G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	2.74 [1.26, 5.94]
3 Neutropenia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

3.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	7.30 [1.70, 31.31]
4 Lymphopenia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

4.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	41.99 [5.85, 301.31]
5 Thrombocytopenia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

5.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	3.19 [1.07, 9.53]
6 Infection G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

6.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	1.39 [0.86, 2.26]
7 Fatigue/asthenia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

7.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	1.10 [0.63, 1.91]
8 Nausea/vomiting G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

8.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	5.48 [0.67, 45.05]
9 Weight loss G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

9.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	4.57 [0.22, 94.47]
10 Neurological symptoms G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

10.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	1.31 [0.82, 2.10]
11 Seizures G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

11.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	1.19 [0.60, 2.39]
12 Elevated liver enzymes G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

12.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	1.71 [0.97, 3.03]
13 Cutaneous adverse event G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

13.1 TMZ vs RT (60 Gy)	1	373	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.06, 14.49]

Comparison 3. Chemotherapy versus radiotherapy

Comparison 4. Chemoradiotherapy versus radiotherapy

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Progression‐free survival Show forest plot	1		Hazard Ratio (Random, 95% CI)	Subtotals only

1.1 TMZ+RT (40 Gy) vs RT (40 Gy)	1	562	Hazard Ratio (Random, 95% CI)	0.50 [0.41, 0.61]
2 Neutropenia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.1 TMZ + RT (40 Gy) vs RT (40 Gy)	1	515	Risk Ratio (M‐H, Random, 95% CI)	10.30 [2.45, 43.34]
3 Thrombocytopenia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

3.1 TMZ + RT (40 Gy) vs RT (40 Gy)	1	527	Risk Ratio (M‐H, Random, 95% CI)	28.56 [3.92, 207.86]
4 Lymphopenia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

4.1 TMZ + RT (40 Gy) vs RT (40 Gy)	1	521	Risk Ratio (M‐H, Random, 95% CI)	2.65 [1.75, 4.01]
5 Leucopenia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

5.1 TMZ + RT (40 Gy) vs RT (40 Gy)	1	528	Risk Ratio (M‐H, Random, 95% CI)	18.16 [2.45, 134.64]
6 Anaemia G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

6.1 TMZ + RT (40 Gy) vs RT (40 Gy)	1	528	Risk Ratio (M‐H, Random, 95% CI)	6.69 [0.35, 128.88]
7 Treatment toxicity G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

7.1 TMZ + RT (40 Gy) vs RT (40 Gy)	1	528	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.68, 1.59]

Comparison 4. Chemoradiotherapy versus radiotherapy

Comparison 5. Other + chemoradiotherapy versus chemoradiotherapy

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Progression‐free survival Show forest plot	1		Hazard Ratio (Random, 95% CI)	Subtotals only

1.1 BEV+chemoRT (60 Gy) vs chemoRT	1	73	Hazard Ratio (Random, 95% CI)	0.78 [0.46, 1.32]
2 Thromboembolic events G3+ Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.1 BEV+chemoRT (60 Gy) vs chemoRT	1	73	Risk Ratio (M‐H, Random, 95% CI)	16.63 [1.00, 275.42]

Comparison 5. Other + chemoradiotherapy versus chemoradiotherapy

Comparison 6. Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Progression‐free survival Show forest plot	1		Hazard Ratio (Random, 95% CI)	Subtotals only

1.1 BEV+RT vs RT	1	75	Hazard Ratio (Random, 95% CI)	0.46 [0.27, 0.78]
2 Thromboembolic events G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	2.0 [0.46, 8.73]

3 Haematological events G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	2.55 [0.13, 51.17]

4 Infections G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	1.17 [0.33, 4.13]

5 Fatigue G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	0.75 [0.13, 4.20]

6 Seizures G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	4.59 [0.26, 82.01]

7 Headaches G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	0.17 [0.01, 4.03]

8 Neuropsychiatric events G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	2.0 [0.24, 16.97]

9 Neurological events G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	0.9 [0.34, 2.40]

10 Hypertension G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	1.0 [0.20, 5.09]

11 Cutaneous adverse events G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	0.0 [0.0, 0.0]

12 Gastrointestinal events G3+ Show forest plot	1	75	Risk Ratio (M‐H, Random, 95% CI)	0.75 [0.13, 4.20]

Comparison 6. Other + radiotherapy (40 Gy) versus radiotherapy (40 Gy)