Scolaris Content Display Scolaris Content Display

Cochrane Database of Systematic Reviews Review - Intervention

Prophylactic antibiotics for adults with chronic obstructive pulmonary disease: a network meta‐analysis

Authors' declarations of interest

Version published: 15 January 2021 Version history

https://doi.org/10.1002/14651858.CD013198.pub2
Collapse all Expand all

Background

Chronic obstructive pulmonary disease (COPD) is a chronic respiratory condition characterised by persistent respiratory symptoms and airflow limitation. Acute exacerbations punctuate the natural history of COPD and are associated with increased morbidity and mortality and disease progression. Chronic airflow limitation is caused by a combination of small airways (bronchitis) and parenchymal destruction (emphysema), which can impact day‐to‐day activities and overall quality of life. In carefully selected patients with COPD, long‐term, prophylactic use of antibiotics may reduce bacterial load, inflammation of the airways, and the frequency of exacerbations.

Objectives

To assess effects of different prophylactic antibiotics on exacerbations, quality of life, and serious adverse events in people with COPD in three separate network meta‐analyses (NMAs), and to provide rankings of identified antibiotics.

Search methods

To identify eligible randomised controlled trials (RCTs), we searched the Cochrane Airways Group Specialised Register of trials and clinical trials registries. We conducted the most recent search on 22 January 2020.

Selection criteria

We included RCTs with a parallel design of at least 12 weeks' duration evaluating long‐term administration of antibiotics prophylactically compared with other antibiotics, or placebo, for patients with COPD.

Data collection and analysis

This Cochrane Review collected and updated pair‐wise data from two previous Cochrane Reviews. Searches were updated and additional studies included. We conducted three separate network meta‐analyses (NMAs) within a Bayesian framework to assess three outcomes: exacerbations, quality of life, and serious adverse events. For quality of life, we collected data from St George's Respiratory Questionnaire (SGRQ). Using previously validated methods, we selected the simplest model that could adequately fit the data for every analysis. We used threshold analysis to indicate which results were robust to potential biases, taking into account each study’s contributions to the overall results and network structure. Probability ranking was performed for each antibiotic class for exacerbations, quality of life, and serious adverse events.

Main results

Characteristics of studies and participants

Eight trials were conducted at multiple sites that included hospital clinics or academic health centres. Seven were single‐centre trials conducted in hospital clinics. Two trials did not report settings. Trials durations ranged from 12 to 52 weeks. Most participants had moderate to severe disease. Mean age ranged from 64 years to 73 years, and more males were recruited (51% to 100%). Forced expiratory volume in one second (FEV₁) ranged from 0.935 to 1.36 L. Most participants had previous exacerbations. Data from 12 studies were included in the NMAs (3405 participants; 16 treatment arms including placebo). Prophylactic antibiotics evaluated were macrolides (azithromycin and erythromycin), tetracyclines (doxycyclines), quinolones (moxifloxacin) and macrolides plus tetracyclines (roxithromycin plus doxycycline).

Risk of bias and threshold analysis

Most studies were at low risk across domains, except detection bias, for which only seven studies were judged at low risk. In the threshold analysis for exacerbations, all comparisons in which one antibiotic was compared with another were robust to sampling variation, especially macrolide comparisons. Comparisons of classes with placebo were sensitive to potential bias, especially macrolide versus placebo, therefore, any bias in the comparison was likely to favour the active class, so any adjustment would bring the estimated relative effect closer to the null value, thus quinolone may become the best class to prevent exacerbations.

Exacerbations

Nine studies were included (2732 participants) in this NMA (exacerbations analysed as time to first exacerbation or people with one or more exacerbations). Macrolides and quinolones reduced exacerbations. Macrolides had a greater effect in reducing exacerbations compared with placebo (macrolides: hazard ratio (HR) 0.67, 95% credible interval (CrI) 0.60 to 0.75; quinolones: HR 0.89, 95% CrI 0.75 to 1.04), resulting in 127 fewer people per 1000 experiencing exacerbations on macrolides. The difference in exacerbations between tetracyclines and placebo was uncertain (HR 1.29, 95% CrI 0.66 to 2.41). Macrolides ranked first (95% CrI first to second), with quinolones ranked second (95% CrI second to third). Tetracyclines ranked fourth, which was lower than placebo (ranked third). Contributing studies were considered as low risk of bias in a threshold analysis.

Quality of life (SGRQ)

Seven studies were included (2237 participants) in this NMA. SGRQ scores improved with macrolide treatment compared with placebo (fixed effect‐fixed class effect: mean difference (MD) ‐2.30, 95% CrI ‐3.61 to ‐0.99), but the mean difference did not reach the minimally clinical important difference (MCID) of 4 points. Tetracyclines and quinolones did not improve quality of life any more than placebo, and we did not detect a difference between antibiotic classes.

Serious adverse events

Nine studies were included (3180 participants) in the NMA. Macrolides reduced the odds of a serious adverse event compared with placebo (fixed effect‐fixed class effect: odds ratio (OR) 0.76, 95% CrI 0.62 to 0.93). There was probably little to no difference in the effect of quinolone compared with placebo or tetracycline plus macrolide compared with placebo. There was probably little to no difference in serious adverse events between quinolones or tetracycline plus macrolide. With macrolide treatment 49 fewer people per 1000 experienced a serious adverse event compared with those given placebo. Macrolides ranked first, followed by quinolones. Tetracycline did not rank better than placebo.

Drug resistance

Ten studies reported drug resistance. Results were not combined due to variation in outcome measures. All studies concluded that prophylactic antibiotic administration was associated with the development of antimicrobial resistance.

Authors' conclusions

This NMA evaluated the safety and efficacy of different antibiotics used prophylactically for COPD patients. Compared to placebo, prolonged administration of macrolides (ranked first) appeared beneficial in prolonging the time to next exacerbation, improving quality of life, and reducing serious adverse events. No clear benefits were associated with use of quinolones or tetracyclines. In addition, antibiotic resistance was a concern and could not be thoroughly assessed in this review. Given the trade‐off between effectiveness, safety, and risk of antibiotic resistance, prophylactic administration of antibiotics may be best reserved for selected patients, such as those experiencing frequent exacerbations. However, none of the eligible studies excluded patients with previously isolated non‐tuberculous mycobacteria, which would contraindicate prophylactic administration of antibiotics, due to the risk of developing resistant non‐tuberculous mycobacteria.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Prophylactic antibiotics for people with COPD

Review question

Which preventative antibiotic is effective and safe for reducing exacerbations, improving quality of life, and reducing serious side effects in people with COPD?

What is COPD?

COPD is a lung condition that can cause long‐term breathing problems. Symptoms include shortness of breath, cough, and sputum production. Flare‐ups (so‐called exacerbations) can be triggered by infection or inflammation, causing worsening symptoms and lung damage. Frequent exacerbations can lead to reduced quality of life and can increase the risk of death.

Why did we do this review?

We wanted to find out if one type of preventative antibiotic was better than another in reducing exacerbations, improving quality of life, and reducing side effects. We did this by using information from two previous reviews and comparing different antibiotics with each other, and with a control treatment (called placebo), by creating networks. As information was limited, the networks allowed us to combine information and determine the best preventative antibiotics by ranking them in order of ability to reduce exacerbations, improve quality of life, and reduce serious side effects.

What evidence did we find?

We tested three types of antibiotics: macrolides, quinolones, and tetracyclines. Macrolides were better in reducing exacerbations compared to control treatment. There was no clear difference in exacerbations when quinolone or tetracycline was compared with a control treatment. Tetracyclines were ranked lower than placebo in reducing exacerbations. We used the data for each antibiotic group to rank antibiotic groups in order of their ability to reduce exacerbations. We found that macrolides ranked first, followed by quinolones (second). Tetracyclines were ranked fourth and were not better than control treatment (ranked third).

Macrolides improved quality of life compared with control treatment. Quinolones did not appear to impact quality of life, and tetracyclines may have been associated with worsening quality of life compared to control treatment.

Macrolides were more effective in reducing serious unwanted events. There was no clear benefit for serious unwanted events with quinolone, tetracycline, or combined macrolide plus tetracycline compared with control treatment.

We could not clearly show benefit or harm of preventative antibiotic use for microbial resistance.

Quality of the evidence

We did not find any concerns about the ways in which studies were carried out, except that for some studies, people collecting the information knew (1) which patient was included in which treatment group, and (2) patient results when treatments were completed. Overall, the numerical information was robust and was unlikely to be influenced by differences noted between individual studies.

Conclusion

We found that exacerbations were reduced, quality of life was improved, and unwanted events were fewer with macrolides compared with control treatment. We could not determine whether quinolones or tetracyclines were of benefit compared with control treatment. Macrolides were ranked highest, followed by quinolones, which ranked second. Tetracyclines were no better than control treatment (ranked fourth and third, respectively). Although these NMAs show some benefit of using macrolides, they are based on a limited number of studies, and concerns remain about antibiotic resistance with long‐term use of antibiotics.

Authors' conclusions

Implications for practice

The NMA in this review compared macrolides, quinolones, and tetracyclines with each other and with placebo in people with moderate to severe COPD, who were already taking concomitant medications (e.g. long‐acting beta agonist (LABA), long‐acting muscarinic antagonist (LAMA), inhaled corticosteroid (ICS), anticholinergics, short‐acting beta agonist (SABA)), and who may have experienced exacerbations, as not all studies reported previous exacerbations. Overall, this NMA shows that treatment with macrolide reduced exacerbations, improved quality of life scores, and reduced serious adverse events in comparison to placebo. This was reflected in ranking of antibiotics, as macrolide was ranked first in reducing exacerbations. Smaller benefit of taking quinolone compared to placebo was noted (ranked second); however, taking tetracycline was not better than taking placebo. It should be noted that certainty of these effects, as measured by precision, was reflected in the 95% CrI, and certainty with respect to robustness of results was seen in results of the threshold analysis.

Our findings are also in line with other guidance, specifically, that provided in GOLD 2020 and NICE 2018. Although we did not explore the impact of prophylactic antibiotics in different patient subgroups, given the trade‐off between effectiveness, safety, and risk of antibiotic resistance, it is only appropriate to consider prophylactic administration of antibiotics for selected patients, such as those experiencing frequent exacerbations. Decisions on antibiotic use would depend on clinical assessment and discussion with patients about benefits and risks associated with long‐term use of prophylactic antibiotics.

It it interesting to note that none of the eligible studies excluded patients with previously isolated non‐tuberculous mycobacteria, with a prolonged QTc on their electrocardiogram, or with hearing loss. In clinical practice, the former group would not be eligible for prophylactic administration of antibiotics due to the risk of developing resistant non‐tuberculous mycobacteria. Moreover, long‐term use of macrolides would be contraindicated for patients with prolonged QTc, and long‐term use of quinolones should be discouraged. Hearing loss is also a contraindication for the use of long‐term macrolides due to ototoxicity (Rubinstein 2002; Smith 2020).

Implications for research

Most of the evidence in this NMA comes from studies investigating macrolides and quinolones, with sparse data derived from studies of tetracyclines and combinations of drug classes. Larger studies of head‐to‐head comparisons between macrolides and quinolones may help to further clarify the relative benefit of each drug and to determine subgroups that may respond more favourably to alternative classes. Careful and transparent reporting of baseline characteristics and potential effect modifiers by trialists will assist future evidence synthesis. Evaluation of the impact of prophylactic antibiotics on more selected populations would also be useful, as would evaluation of their impact on different types of COPD exacerbations.

Although studies involving tetracyclines may be informative, given the lack of evidence of benefit and the suggestion of possible harm, it seems unlikely that such trials will be carried out in the future.

Studies investigating alternatives to oral antibiotics may be useful, for example, studies of inhaled antibiotics. This route of administration was not considered in this review but could be considered for future network meta‐analysis if sufficient evidence became available.

Given the anti‐inflammatory effects of macrolides, understanding their impact on infective versus non‐infective exacerbations also warrants further investigation.

Given that this NMA was based on two previous reviews that undertook pair‐wise analyses separately, it will be decided in the future whether we will update the two previous reviews, or update this NMA to incorporate the pair‐wise analyses to keep all information in one review.

Summary of findings

Open in table viewer
Summary of findings 1. Summary of findings: exacerbations

Prophylactic antibiotics compared with placebo for COPD

Patients or population: adults with COPD

Settings: hospital clinics, multi‐centre

Intervention: macrolide, tetracycline, or quinolone

Comparison: placebo or standard care

Treatment

Anticipated absolute effects (95% CrI)*

Relative effect
HR (95% CrI)

No. of participants
(studies)

Absolute rate of exacerbations: median (95% CrI)

Risk difference with treatment

(number of people experiencing exacerbations)

Macrolide

(weighted mean 50 weeks' duration)

1.34 (1.19 to 1.50)

127 fewer per 1000 (168 fewer to 87 fewer)

0.67 (0.60 to 0.75)

688 (6)

Tetracycline

(13 weeks' duration)

2.58 (1.33 to 4.81)

60 more per 1000 (129 fewer to 127 more)

1.29 (0.66 to 2.41)

25 (1)

Quinolone

(weighted mean 46.5 weeks' duration)

1.77 (1.50 to 2.08)

35 fewer per 1000 (87 fewer to 11 more)

0.89 (0.75 to 1.04)

594 (2)

*The basis for the anticipatedrisk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% CrI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CrI).
COPD: chronic obstructive pulmonary disease; CrI: credible interval; HR: hazard ratio.

*Absolute rate of exacerbations per year in the placebo arm = 2; 864 people per 1000 experienced exacerbations over a year.

Open in table viewer
Summary of findings 2. Summary of findings: change from baseline in SGRQ

Prophylactic antibiotics compared with placebo for COPD

Patients or population: adults with COPD

Settings: hospital clinics, multi‐centre

Intervention: macrolide, tetracycline, or quinolone

Comparison: placebo

Treatment

Anticipated absolute effects (95% CrI)*

No. of participants
(studies)

Absolute change from baseline in SGRQ (95% CrI)

Mean difference in change from baseline in SGRQ score with treatment**

Macrolide

(weighted mean 48 weeks' duration)

‐4.00 (‐5.51 to ‐2.68)

2.298 point improvement (3.605 to 0.985 point improvement)

578 (6)

Tetracycline

(13 weeks' duration)

‐0.52 (‐3.21 to 2.16)

1.179 point worsening (1.509 point improvement to 3.859 point worsening)

25 (1)

Quinolone

(weighted mean 46.5 weeks' duration)

‐3.03 (‐4.69 to ‐1.37)

1.33 point improvement (2.986 point improvement to 0.328 point improvement)

528 (2)

*The basis for the anticipatedrisk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% CrI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CrI).
COPD: chronic obstructive pulmonary disease; CrI: credible interval; SGRQ: St George's Respiratory Questionnaire.

*Absolute change from baseline in the placebo arm was ‐1.7 (1.7 point improvement).

**The minimally clinically important difference for SGRQ is 4 points.
Open in table viewer
Summary of findings 3. Summary of findings: serious adverse events

Prophylactic antibiotics compared with placebo for COPD

Patients or population: adults with COPD

Settings: hospital clinics, multi‐centre

Intervention: macrolide, tetracycline, or quinolone

Comparison: placebo

Treatment

Anticipated absolute effects (95% CrI)*

Relative effect
OR (95% CrI)

No. of participants
(studies)

Absolute probability of an SAE: median (95% CrI)

Risk difference with treatment*

Macrolide

(weighted mean 49 weeks' duration)

0.21 (0.18 to 0.25)

49.07 fewer per 1000 (81.18 fewer to 14.23 fewer)

0.76 (0.62 to 0.93)

971 (8)

Quinolone

(48 weeks' duration)

0.26 (0.20 to 0.32)

1.873 fewer per 1000 (57.88 fewer to 60.89 more)

1.00 (0.72 to 1.34)

569 (1)

Macrolide + tetracycline

(12 weeks' duration)

0.25 (0.15 to 0.37)

9.461 fewer per 1000 (1.07 fewer to 108.5 more)

0.97 (0.52 to 1.66)

101 (1)

*The basis for the anticipatedrisk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% CrI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CrI).
COPD: chronic obstructive pulmonary disease; CrI: credible interval; OR: odds ratio.

*Absolute probability of events in the placebo arm was 0.26; risk of an SAE with placebo was 260 per 1000.

Background

Description of the condition

Chronic obstructive pulmonary disease (COPD) is a common and preventable disease that is characterised by persistent respiratory symptoms and airflow obstruction, with or without alveolar abnormalities, usually caused by significant exposure to noxious particles or gases (GOLD 2020). Tobacco smoking is considered the main risk factor for COPD, but other factors, such as biomass fuel and air pollution, can also contribute to development of the disease. In addition, individuals with genetic abnormalities, abnormal lung development, and accelerated ageing are likely to be susceptible to COPD (GOLD 2020). Common respiratory symptoms include dyspnoea, cough with or without sputum production, and recurrent lower respiratory tract infection. People with COPD may experience intermittent worsening of symptoms, known as exacerbations. Exacerbations are associated with increased mortality (Soler‐Cataluña 2005), higher healthcare costs (Pasquale 2012), and a more rapid decline in lung function (Donaldson 2002), as well as negative impact on quality of life (Mathioudakis 2020; Seemungal 1998).

Description of the intervention

The ECLIPSE study has shown that frequent flare‐ups are associated with a moderate to severe COPD phenotype; as disease severity increases, the frequency of exacerbations also increases (Hurst 2010). One approach to reduce the frequency of exacerbations of COPD and reverse this potential ‘vicious cycle' of inflammation is the long‐term use of antibiotics to prevent exacerbations. Such antibiotics are usually given by mouth but can also be inhaled. Depending on the type of antibiotic, it can be taken daily or three times a week, or by ‘pulsed' administration (e.g. 'pulsed' antibiotic may be given daily for several days followed by a break) (Herath 2018).

Authors of a Cochrane Review investigated effects of macrolides and a quinolone compared with control treatment (Herath 2018). Long‐term use of antibiotics was associated with significantly fewer patients experiencing an exacerbation of COPD compared with those receiving control treatment. Patients on prophylactic antibiotics were more likely to experience adverse effects, such as hearing loss with azithromycin and gastrointestinal symptoms with moxifloxacin.

How the intervention might work

Effects of long‐term antibiotics are not completely understood. Antibiotics may offer both antibacterial and anti‐inflammatory effects (Martinez 2008), and therefore may reduce both bacterial load and inflammation as a result of exacerbations from bacteria, viruses, and environmental pollution. Studies have suggested that the lungs of people with COPD may be colonised with more pathogenic bacteria than are found in healthy lungs (Mathioudakis 2020; Sethi 2004). Bacteria are identified in the sputum of approximately 40% to 60% of people experiencing an acute exacerbation (Sethi 2004), and their overgrowth may be a precipitant of exacerbations (Sze 2014). Antibiotics may also reduce neutrophilic airway inflammation by reducing bacterial load, potentially providing clinical benefit (Siva 2014). Choice of prophylactic antibiotic may be guided by factors including clinician and patient preferences and prior experience, previously isolated bacteria, and side effect profiles. Organisms isolated from exacerbating patients include Haemophilus influenzae (11% of all patients), Streptococcus pneumoniae (10%), Moraxella catarrhalis (10%), Haemophilus parainfluenzae (10%), and Pseudomonas aeruginosa (4%) (Sapey 2006).

Prophylactic antibiotics may be of greatest benefit in a subset of patients (Mathioudakis 2017; Miravittles 2015). Compared to placebo, azithromycin (a macrolide antibiotic) reduces exacerbations most markedly in older patients, non‐smokers, and those not using oral or inhaled steroids at baseline (Albert 2011).

Why it is important to do this review

This Cochrane Review included a network meta‐analysis that will accompany the head‐to‐head pair‐wise meta‐analysis review of prophylactic antibiotics (Threapleton 2018); this was supplemented with the addition of antibiotic versus placebo data (Herath 2018). As comparisons of antibiotics for reducing exacerbations and improving quality of life for patients with COPD were limited, a network meta‐analysis (NMA) was important to identify which antibiotic was better for improving these outcomes.

Objectives

To assess effects of different prophylactic antibiotics on exacerbations, quality of life, and serious adverse events in people with COPD in three separate network meta‐analyses, and to provide rankings of identified antibiotics.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs), regardless of language or publication status. We included trials of minimum 12 weeks' intervention duration. This duration was considered an appropriate minimum cut‐off to allow for evaluation of the impact of interventions on COPD exacerbations. We excluded cross‐over trials due to carry‐over effects. We did not identify any cluster‐randomised trials.

Types of participants

We included adults 18 years of age and older who had been diagnosed with COPD according to validated criteria (e.g. European Respiratory Society, American Thoracic Society, Global Initiative for Obstructive Lung Disease (GOLD) criteria). We included studies enrolling patients with COPD during a stable disease state, or during exacerbations, provided that antibiotics were administered long‐term, prophylactically. We included study populations with mild, moderate, severe, or very severe COPD according to the GOLD criteria for airflow limitation (GOLD 1: ≥ 80% predicted forced expiratory volume in one second (FEV₁); GOLD 2: 50% to 79%; GOLD 3: 30% to 49%; GOLD 4 < 30%). We anticipated that trials would likely recruit patients with moderate to severe or very severe COPD (GOLD stages 2 to 4). Most patients included in the studies had moderate to very severe COPD. Only one study included patients with mild COPD (5%), but this number was very small and was unlikely to affect our analyses. We included trials that recruited participants with or without a recent history of exacerbations and explored this as a potential source of heterogeneity. We excluded patients with the following co‐morbidities or characteristics: a primary diagnosis of bronchiectasis, asthma, or genetic disease, such as cystic fibrosis or primary ciliary dyskinesia.

Types of interventions

We included any prophylactic oral antibiotic classes given for at least 12 weeks continuously, intermittently (e.g. three times per week), or pulsed, in keeping with the linked pair‐wise meta‐analyses (Herath 2018; Threapleton 2018). Pulsed antibiotics must have been given for a minimum of five consecutive days every eight weeks.

We included trials in which participants had access to the following background treatments provided they were not part of the randomised study treatments.

  • Short‐acting and long‐acting bronchodilators.

  • Inhaled corticosteroids.

  • Oral corticosteroids.

  • Oxygen.

  • Pulmonary rehabilitation.

  • Smoking cessation interventions.

  • Any other standard treatment for COPD.

Types of outcome measures

Primary outcomes

  • COPD exacerbation* (we extracted data on time until first exacerbation, estimated using hazard ratios (HRs) as a preference, followed by rate ratio data and numbers of participants with one or more exacerbations)

  • Quality of life (St George's Respiratory Questionnaire (SGRQ))

  • All‐cause serious adverse events (number of participants with one or more adverse event)

  • Drug resistance/Microbial sensitivity (we did not perform NMA on this outcome, but we reported results for this outcome narratively)

  • Mortality (we anticipated that events would be rare, so we did not perform NMA on this outcome, but we reported results for this outcome narratively)

We reported endpoint data for dichotomous outcomes. Continuous outcomes were extracted, and we reported then at the closest time points to 6 months and 12 months.

*Moderate and severe exacerbations were defined as worsening of respiratory status requiring treatment with systemic corticosteroids and/or antibiotics; severe exacerbations were defined as requiring hospitalisation (see Table 1).

Open in table viewer
Table 1. Characteristics of studies including prior exacerbation details

Author

Class comparison

Concomitant treatments (%, antibiotic/ placebo)

Dose/regimen

COPD severity

Included in NMA?

Exacberations in the previous 12 months before participation in study

Exacerbation definition

Risk of bias

Albert 2011

(N = 1142);

USA

(12 academic health centres)

52 weeks

Macrolide vs placebo

ICS only (4%/6%)

LAMAs only
(6%/8%)

LABAs only
(3%/1%)

ICS + LAMA (19%/22%)
ICS + LABA (4%/5%)

LABA + LAMA (5%/4%)
ICS + LABA + LAMA (49%/46%)

None (10%/8%)

AZM 250 mg daily

continuous

Moderate to severe

FEV₁ 1.11 L

Yes

Approximately 50% of participants in each treatment arm had required hospitalisation or an ED visit in the previous 12 months

Acute exacerbation of COPD: “a complex of respiratory symptoms (increased or new onset) of more than one of the following: cough, sputum, wheezing, dyspnoea, or chest tightness with a duration of at least 3 days requiring treatment with antibiotics or systemic steroids"

Low risk of bias across all domains except attrition (unclear reasoning of missing data for HRQoL)

Banerjee 2005
(N = 67); UK (clinics and lung function units from 2 hospitals)

13 weeks

Macrolide vs placebo

All participants:

ICS (100%),
LABAs (18%), inhaled anticholinergics (63%)

CLR 500 mg daily

continuous

Moderate to severe

No

NR

Not included in exacerbation analysis

Low risk of bias across all domains except detection bias, which was unclear

Berkhof 2013

(N = 84);

Netherlands

(1 teaching hospital)

12 weeks

Macrolide vs placebo

LABAs (81%/80%)

Long‐acting anticholinergics (64%/57%)

ICS (98%/83%)

AZM 250 mg 3 times a week

Intermittent

Moderate

FEV₁ 1.36 L

Yes

Participants had a median for 1 exacerbation (range 0 to 13) in the previous 12 months

Time to first exacerbation of COPD: sustained worsening of COPD, from stable state and beyond normal day‐to‐day variations, requiring treatment with prednisolone, antibiotics, or a combination of both

Unclear selection bias (allocation) but assumed done, low risk across all other domains

Blasi 2010

(N = 22);

Italy

(multi‐centre)

26 weeks

Macrolide vs placebo

Inhaled medication NR LTOT (46% in both groups)

AZM 500 mg 3 times a week

Intermittent

Severe

FEV₁ (not reported)

Yes

Participants in each treatment arm had a mean of 3 exacerbations in the previous 12 months

Worsening of symptoms requiring both a change in regular respiratory medication or medical assistance, or resulting in hospitalisation or ED treatment

Judged as high risk of bias for allocation concealment, performance, detection, attrition, and selective reporting; open‐label

Brill 2015 (N = 99);

UK

(1 outpatient

hospital department)

13 weeks

Quinolone

Tetracycline

Macrolide

vs placebo

ICS (84%/76%/72%)

ICS in placebo group 57%

MOX 400 mg daily for 5 days every 4 weeks (pulsed)

DOX 100 mg daily (continuous)

AZM 250 mg 3 times a week (intermittent)

Moderate to severe

FEV₁ 1.4 L

Yes

Participants had a mean of 2.5 (SD 2.1) exacerbations with moxifloxacin, 2.1 (SD 1.7) exacerbations with doxycycline, 2.8 (SD 4.0) exacerbations with azithromycin, and

1.5 (SD 1.4) exacerbations with placebo in previous 12 months

Exacerbations during the study: using diary card criteria, patient reporting to clinical health professionals or study team.

Exacerbation was not the primary outcome of the study

Unclear performance bias; detection bias judged as high

He 2010

(N = 36);

China

(1 university hospital)

26 weeks

Macrolide vs placebo

ICS (44%/38%)

Theophylline (61%/55%)

Inhaled anticholinergic (50%/55%)

Inhaled beta‐adrenergic (72%/77%)

ERY 125 mg 3 times daily (continuous)

Severe

FEV₁ 1.07 L at baseline

Yes

NR

Moderate exacerbation: sustained worsening of baseline respiratory symptoms for at least 2 days requiring increased treatment or additional therapy (e.g. OCS, antibiotics)

Severe exacerbation: all of the above plus requiring hospital admission

Randomisation and allocation unclear. Double‐blind study, but outcome assessment unclear. Funding not stated

Mygind 2010

(N = 575);

Denmark

156 weeks

Macrolide vs placebo

NR

AZM 500 mg for 3 days every month (pulsed)

NR

No

NR

Not included in exacerbation analysis

Unclear randomisation, allocation concealment, attrition domains. Blinding of participants, personnel, and outcome assessors were judged as low risk of bias

Seemungal 2008

(N = 109);

UK

(2 outpatient clinics in 2 hospitals)

52 weeks

Macrolide vs placebo

ICS (77% in both groups)

LABAs (66%/61%)

LAMAs (28%/38%)

Theophylline (7.5%/14%)

ERY 250 mg twice daily (continuous)

Moderate to severe

FEV₁ 1.31 L at baseline

Yes

35% of participants had 3 or more exacerbations in the previous 12 months

Moderate exacerbation: sustained worsening of baseline respiratory symptoms for at least 2 days requiring treatment with OCS (prednisolone) and/or antibiotics

Severe exacerbation: requiring admission to hospital

Low risk of bias across all domains. Funded by British Lung Foundation

Sethi 2010

(N = 1157);

(international

multi‐centre)

48 weeks

Quinolone

vs placebo

SABAs (71%/72%)

LABAs (44%/45%)

ICS (41%/43%)

Theophylline (29%/26%)

Systemic steroids (0.4%/0.2%)

Others (4.7%/5.7%)

ICS + long‐acting bronchodilators (25%/26%)

MOX 400 mg daily for 5 days every 8 weeks (pulsed)

Mild to severe

FEV₁ 1.2 L at baseline

Yes

NR

Any confirmed AECB: requiring intervention

(start of systemic antibiotic and/or start of systemic steroid and/or hospitalisation within 7 days of the start date of exacerbation) and with a minimum of 2 weeks between the start of 2 consecutive exacerbations

Unclear risk for selection bias (random sequence generation and allocation concealment). Low risk for performance bias and selective reporting

Shafuddin 2015

(N = 292); Australia and New Zealand (multi‐ centre)

12 weeks

Macrolide

Macrolide plus tetracycline

vs placebo

NR

ROX 300 mg daily (continuous)

DOX + ROX 100 mg daily plus 300 mg daily (continuous)

Moderate to severe

FEV₁ 0.935 L at baseline

Yes

Mean 5.11 (SD 2.4) exacerbations within 2 years

Not included in exacerbation analysis

Low risk of bias across all domain except attrition, which was unclear

Simpson 2014

(N = 30);

Australia

(1 tertiary care respiratory and sleep ambulatory care service, hospital)

12 weeks

Macrolide vs placebo

ICS (% NR)

AZM 250 mg daily (continuous)

Moderate

Yes

NR

Severe exacerbations of COPD: requiring unscheduled medical attention with treatment of OCS and/or antibiotics

Low risk of bias across all domains

Singh 2019 (N = 60); India

(1 outpatient department)

13 weeks

Tetracycline

vs placebo

NR

DOX 100 mg daily (continuous)

Moderate to severe

No (sensitivity analysis)

NR

Not included in exacerbation analysis

Low risk of bias for allocation concealment, high risk of bias for blinding of participants, personnel, and outcome assessors. Randomisation and selective reporting domains were unclear

Suzuki 2001 (N = 109);

Japan (setting NR)

13 weeks

Macrolide

vs placebo

NR

ERY 200 to 400 mg daily (continuous)

FEV₁ 1.47 L at baseline

No (sensitivity analysis)

NR

Not included in exacerbation analysis

Low risk of bias across most domains except for blinding of participants, personnel, and outcome assessors, which were judged as high risk of bias

Tan 2016

(N = 49);

China

(1 regional hospital)

52 weeks

Macrolide

vs placebo

ICS (44%/38%/44%)

Theophylline (55%/55%/61%)

Inhaled anticholinergic (55%/50%/50%)

Inhaled beta2‐adrenergic agonist (66%/66%/72%)

ERY 125 mg 3 times daily (continuous)

ERY 125 mg 3 times daily with 6 months' follow‐up (continuous)

Moderate to severe

FEV₁ 1.04 to 1.08 L

Yes

NR

Not included in exacerbation analysis

Unclear risk of bias across most domains, high risk of bias for blinding of participants, personnel, and outcome assessors

Uzun 2014

(N = 92);

Netherlands

(1 regional hospital)

52 weeks

Macrolide

vs placebo

LABA (96%/91%)

LAMA (89%/71%)

ICS (89%/96%)

SABA (68%/73%)

Prednisolone (28%/20%)

AZM 500 mg 3 times a week (intermittent)

Mild to severe

FEV₁ 1.1 L at baseline

Yes

Participants had a mean of 4 (SD 1.1) acute exacerbations in the previous 12 months

All exacerbations: defined according to Anthonisen criteria, and whether the patient needed treatment with steroids or antibiotics, or both.

Severe exacerbation: requiring hospital admission.

Mild exacerbation: requiring treatment at the outpatient department

Low risk of bias across all domains

Vermeersch 2019

(N = 301);

Italy

(5 centres across Italy)

13 weeks

Macrolide

vs placebo

LABA (93%/94%)

LAMA (80%/80%)

ICS (80%/80%)

SABA (73%/71%)

AZM 500 mg once daily (loading dose) for 3 days, followed by 250 mg every 2 days for 13 weeks (intermittent)

FEV₁ 0.925 L

Yes

NR

Not included in exacerbation analysis

Low risk of bias across all domains

Wang 2017

(N = 86);

China

(1 regional hospital)

26 weeks

Macrolide

vs placebo

NR

AZM 250 mg once daily plus 20 mg once daily simvastatin (continuous)

FEV₁ 0.67 L

No

NR

Not included in exacerbation analysis

Low risk of bias for randomisation, high risk of bias for blinding of participants, personnel, and outcome assessors

Abbreviations

AECB: acute exacerbation of chronic bronchitis; AZM: azithromycin; CLR: clarithromycin; COPD: chronic obstructive pulmonary disease; DOX: doxycycline; ED: emergency department; ERY: erythromycin; FEV₁: forced expiratory volume in one second; HRQoL: health‐related quality of life; ICS: inhaled corticosteroid; LABA: long‐acting beta agonist; LAMA: long‐acting muscarinic antagonist; LTOT: long‐term oxygen therapy; MOX: moxifloxacin; NMA: network meta‐analysis; NR: not reported; OCS: oral corticosteroids;ROX: roxithromycin; SABA: short‐acting beta agonist; SD: standard deviation.

Search methods for identification of studies

Electronic searches

We updated the searches for both Herath 2018 and Threapleton 2018.

For this NMA, we identified studies from the Cochrane Airways Trials Register, which is maintained by the Information Specialist of the Cochrane Airways Group. We carried out the first search in October 2018 and updated it on 22 January 2020. At the time of this review, the Cochrane Airways Trials Register contained studies identified from:

  • monthly searches of the Cochrane Central Register of Controlled Trials (CENTRAL), in the Cochrane Library, through the Cochrane Register of Studies (inception to Issue 12; 2019);

  • weekly searches of MEDLINE Ovid SP (1946 to January 2020);

  • weekly searches of Embase Ovid SP (1974 to January 2020);

  • monthly searches of PsycINFO Ovid SP (1967 to January 2020);

  • monthly searches of the Cumulative Index to Nursing and Allied Health Literature (CINAHL EBSCO; 1937 to January 2020);

  • monthly searches of the Allied and Complementary Medicine Database (AMED EBSCO; inception to January 2020); and

  • handsearches of major respiratory conference proceedings.

Studies contained in the Cochrane Airways Trials Register were identified through search strategies based on the scope of Cochrane Airways. Details of these strategies, as well as a list of handsearched conference proceedings, are presented in Appendix 1. See Appendix 2 for search terms used to identify studies for this review.

We searched the following trials registries.

  • US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov; searched 22 January 2020).

  • World Health Organization International Clinical Trials Registry Platform (apps.who.int/trialsearch; searched 22 January 2020).

We searched the Cochrane Airways Trials Register and additional sources with no restriction on language or type of publication.

Searching other resources

For this NMA, we checked the reference lists of all primary studies and review articles for additional references. We searched relevant manufacturers' websites for study information.

On 21 August 2020, we searched for errata or retractions from included studies published in full text on PubMed (www.ncbi.nlm.nih.gov/pubmed).

Data collection and analysis

This review was built on two existing Cochrane Reviews (Herath 2018; Threapleton 2018), in which data from included studies in each of the reviews had already been extracted by two pairs of independent review authors. For studies already identified from the two existing Cochrane Reviews that reported exacerbations outcome data, we checked and extracted hazard ratio data if these data were available. New studies that were not included in Herath 2018 and Threapleton 2018 were selected, and data were extracted as outlined below.

Selection of studies

Two review authors (SJ and CT) independently screened the titles and abstracts of search results and coded them as either ‘retrieve' (eligible or potentially eligible/unclear) or ‘do not retrieve'. We retrieved the full‐text study reports of all potentially eligible studies. Two review authors (SJ and CT) independently assessed these for inclusion and recorded the reasons for exclusion of ineligible studies. We selected studies that evaluated clinical efficacy and safety of any prophylactic antibiotic treatments in patients with COPD (e.g. macrolides/quinones, macrolides/tetracyclines, quinones/tetracyclines, combined macrolide plus tetracycline/macrolide). We resolved any disagreement through discussion or, if required, we consulted a third review author (RF). We identified and excluded duplicates and collated multiple reports of the same study, so that each study, rather than each report, was the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram and Characteristics of excluded studies tables (Moher 2009).

Data extraction and management

We used Microsoft Excel to manage outcome data for the NMA, which we had piloted on at least one trial included in the review. One review author (SJ) extracted the following study characteristics from included trials that had not already been included in the two existing Cochrane Reviews (Herath 2018; Threapleton 2018).

  • Methods: study design, total duration of study, details of any ‘run‐in' period, number of study centres and locations, study settings, withdrawals, dates of study.

  • Participants: N, mean age, age range, gender, severity of COPD, diagnostic criteria, baseline lung function, smoking history, inclusion criteria, exclusion criteria, previous history of exacerbations.

  • Interventions: intervention, comparison, concomitant medications, excluded medications.

  • Outcomes: primary and secondary outcomes specified and collected, time points reported.

  • Notes: funding for studies, notable conflicts of interest of trial authors.

Two review authors (SJ and CT) independently extracted outcome data from included trials that had not already been identified by the two existing Cochrane Reviews (Herath 2018; Threapleton 2018), which they managed in Microsoft Excel. We noted in the Characteristics of included studies table if outcome data were not reported in a useable way. We resolved disagreements by reaching consensus or by involving a third review author (RF). We double‐checked that data were entered correctly by comparing data presented in the systematic review against study reports. A second review author (CT) spot‐checked study characteristics for accuracy against the study report. We recorded on the data extraction sheet data extracted from previous Cochrane Reviews that were relevant for this NMA.

Assessment of risk of bias in included studies

Studies that had been identified from Herath 2018 and Threapleton 2018 had previously been assessed for risk of bias by two pairs of independent review authors. Trials that were not already included in Herath 2018 and Threapleton 2018, were assessed for risk of bias as outlined below.

Two review authors (SJ and CT) independently assessed risk of bias for each included trial using the criteria outlined in the recently updated Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2018). We resolved disagreements by discussion or by consultation with a third review author (RF). We assessed risk of bias according to the following domains.

  • Random sequence generation.

  • Allocation concealment.

  • Blinding of participants and personnel.

  • Blinding of outcome assessment.

  • Incomplete outcome data.

  • Selective outcome reporting.

  • Other bias.

We judged each potential source of bias as ‘high', ‘low', or ‘unclear' and provided a quote from the study report together with a justification for our judgement in the ‘Risk of bias' table. We summarised ‘Risk of bias' judgements across different trials for each of the domains listed. We considered blinding separately for different key outcomes when necessary (e.g. for unblinded outcome assessment, risk of bias for all‐cause mortality may be very different than for a patient‐reported pain scale). When information on risk of bias was related to unpublished data or correspondence with a trial author, we noted this in the ‘Risk of bias' table.

When considering treatment effects, we took into account the risk of bias for trials that contributed to those outcomes.

Assessment of bias in conducting the systematic review

We conducted the review according to the published protocol and justified any deviations from it in the Differences between protocol and review section of this systematic review.

Measures of treatment effect

Direct pair‐wise meta‐analysis methods

Briefly, two published Cochrane Reviews outlined the pair‐wise meta‐analyses from prophylactic versus placebo‐controlled trials and head‐to‐head antibiotic trials (Herath 2018; Threapleton 2018). These reviews were conducted by standard Cochrane methods. Dichotomous data were analysed as odds ratios and continuous data as mean differences (MDs) or standardised mean differences (SMDs). Data in Threapleton 2018 were insufficient for review authors to conduct meta‐analyses; however, pooled results in Herath 2018 with dichotomous variables were expressed as a random effects model odds ratio (OR) with 95% CI. Rate data were combined (e.g. number of exacerbations per participant per year) via generic inverse variance (GIV) and were expressed as a rate ratio (Herath 2018).

NMA methods

We conducted an NMA of clinical trials to compare all prophylactic antibiotics with each other and with placebo. Bayesian Markov chain Monte Carlo method was implemented in OpenBUGS 3.2.3 (Lunn 2009). We used a hierarchical model with classes of antibiotics composed of individual treatments, which allowed each treatment effect and the overall class mean to be estimated (Dias 2018; Kew 2014).

We combined dichotomous data that took into account exposure time with rate and hazard ratio data for the exacerbations outcome; dichotomous data were combined with hazard ratio or rate ratio data with the assumption that all exacerbations occurred at the same rate (i.e. a patient is not more likely to have a second exacerbation if he or she has had a previous exacerbation). This was done by using a shared parameter model in OpenBUGS, whereby data on the log hazard ratio of exacerbations were modelled with normal likelihood and an identity link. Dichotomous data on the number of patients with at least one exacerbation were modelled using a binomial likelihood with a cloglog link (Dias 2018). Depending on availability, we extracted hazard ratios as a preference because they accounted for time at risk and censoring. We pooled other dichotomous outcomes as odds ratios. We used mean differences for continuous outcomes.

Prior distributions

For all models, vague prior distributions were used for all trial baselines and for relative treatment or class effects (normal(0,100²)). For random treatment effects models, a minimally informative uniform prior distribution was used for the between‐study heterogeneity parameter, with lower limit of zero and upper limit of 5 for exacerbations and serious adverse events (SAEs), and upper limit of 15 for SGRQ. For exchangeable‐class models, a Uniform(0, 5) prior distribution was used for the within‐class standard deviation.

Where the number of studies per comparison is small (usually less than 5), empirically informative prior distributions for the heterogeneity parameter are recommended (Rhodes 2015; Turner 2015). In order to assess sensitivity of results of random‐treatment effects model to the the prior distribution for the heterogeneity, results using empirically based prior distributions were also presented for the change for baseline in SGRQ and SAE outcomes. No empirically based prior distributions were available for outcomes on the log‐hazard ratio scale, so these were not considered for the exacerbations outcome. Therefore, when few studies were available in all comparisons for the exacerbations outcome, a half‐normal prior distribution that expressed the prior belief that 95% of trials would give hazard ratios within a factor of 2 from the estimated median hazard ratio was considered: half‐N(0, 0.32²) (Dias 2018). For exchangeable or fixed class models, the minimally informative uniform prior distribution was used (Turner 2015).

In the random treatment effects model for the change from baseline in SGRQ, we used the empirically based t‐distribution for the log of the between‐study variance for comparisons of pharmacological therapies to placebo for quality of life outcomes in respiratory diseases (t(‐5.07, 2.512, 5) as reported by Rhodes 2015. Because this prior distribution was presented in Rhodes 2015 on a standardised mean difference scale, it was converted to the mean difference scale by multiplying by 14, which is approximately the standard deviation of the SGRQ scale in patients with COPD (Puhan 2006).

For the random treatment effects model for SAE, we used the empirically based log‐normal distribution for the between‐trial variance (LN(‐1.87, 1.522)), as reported by Turner 2015.

Fixed‐class models were chosen for these outcomes (for which there were comparisons with more than 5 studies), and minimally informative uniform prior distributions were used.

Model fit and choice

We chose a model and considered it as the primary analysis for NMAs using the following strategy.

  • Start with fixed class models (with random and fixed treatment effects). If both fit well (i.e. posterior mean of residual deviance is close to the number of data points), choose the model with the lowest deviance information criterion (DIC) (if the difference is less than 3, choose the fixed effect model) and stop .

  • If the fixed treatment effect‐fixed class model does not fit well, try the fixed treatment effect‐random class model – assess fit, compare to models in the first step here, and choose the model with the lowest DIC.

  • If neither of the models in the first or second step fit well, try also random treatment effects with random class model. Choose a final model based on DIC, but interpret with caution if model fit is poor.

  • Compare results of random class models to the equivalent treatment level model (i.e. no class), if networks are connected.

Threshold analysis

A contrast level threshold analysis was performed to examine the impact of bias on each treatment contrast (Phillippo 2018; Phillippo 2019). Thresholds are provided that quantify how much the evidence could change (due, for example, to potential biases, or simply sampling variation) before the best treatment changes, and what the revised ‘best’ treatment would be. If it is judged that the evidence could not plausibly change by more than this amount, then the ‘best’ treatment choice is considered robust; otherwise, this choice is sensitive to plausible changes in the evidence.

Unit of analysis issues

Pair‐wise analysis

Herath 2018 and Threapleton 2018 reported pair‐wise data. Neither Herath 2018 nor Threapleton 2018 identified any cross‐over or cluster‐randomised trials. Threapleton 2018 used participants rather than events as the unit of analysis (i.e. the number of people admitted to hospital, rather than the number of admissions per person).

NMA

For dichotomous outcomes, participants were used as the unit of analysis to eliminate risk of multiple counting of participants (i.e. number of COPD patients with one exacerbation). If exacerbation data were provided as rate ratios or HRs, these data were extracted and analysed accordingly. Data from cluster‐randomised trials were planned to be included provided the data had been, or could be, adjusted to take clustering into account.

Dealing with missing data

For both pair‐wise and NMA missing data, investigators or trial sponsors were contacted to verify key study characteristics and to obtain missing numerical outcome data when possible (e.g. when a trial is identified as an abstract only). When this was not possible and the missing data were thought to introduce serious bias, we planned to perform a sensitivity analysis to determine whether the missing data could introduce serious bias to the overall results of the NMA (Guyatt 2017). When possible, we used intention‐to‐treat (ITT) data from randomly assigned participants.

Assessment of heterogeneity

Pair‐wise meta‐analysis

Herath 2018 tested for heterogeneity when CIs did not overlap with each other. The I² statistic was used to measure heterogeneity among the studies in each analysis. When we identified heterogeneity (I² ≥ 40%), we explored this using a pre‐specified subgroup analysis. We used the following overlapping cut‐off to define heterogeneity (Higgins 2011).

  • 0% to 40%: might not be important.

  • 30% to 60%: may represent moderate heterogeneity.

  • 50% to 90%: may represent substantial heterogeneity.

  • 75% to 100%: considerable heterogeneity.

As Threapleton 2018 identified insufficient studies for meta‐analysis, the I² statistic was not used to measure heterogeneity nor to perform pre‐specified subgroup analyses.

NMA
Assessment of similarity of participants, interventions, and trial methods

We assessed clinical similarity of studies and statistical consistency (when possible). Note that incoherence, transitivity, and the presence of effect modifiers all relate to the same issue of consistency, which was addressed clinically and statistically (when possible).

Assessment of heterogeneity and statistical consistency in the network meta‐analysis

For the NMA, we assessed consistency by comparing the model fit and between‐trial heterogeneity from NMA models versus those from an unrelated effects (inconsistency) model (Dias 2013a; Dias 2013b). We would use this to determine the presence and area of inconsistency. In networks for exacerbations and SGRQ, all loops were formed by a single multi‐arm study (Brill 2015). For the SAE network, all loops were formed by one study (Shafuddin 2015). Therefore, there was no potential to detect inconsistency in these networks, and inconsistency checks were not carried out.

We planned to qualitatively compare results from direct pair‐wise meta‐analysis versus NMA estimates to check for broad agreement.

Assessment of reporting biases

Pair‐wise meta‐analysis

Both Herath 2018 and Threapleton 2018 aimed to pool data if they identified more than 10 studies, and to examine a funnel plot to explore possible small‐study and publication biases. Threapleton 2018 did not explore reporting bias, as review authors identified insufficient studies. Herath 2018 attempted to contact study authors to ask for missing data when reporting bias was suspected.

NMA

We aimed to minimise reporting bias from unpublished trials or selective outcome reporting by using a broad search strategy, and by checking references of included trials and relevant systematic reviews. For each outcome, we estimated and presented the proportion of trials that contributed to the NMA. When possible, we aimed to combine data reported as HR, rate ratio, or number of participants with, for example, at least one exacerbation to minimise reporting bias.

Data synthesis

Pair‐wise meta‐analysis

Herath 2018 subgrouped all meta‐analyses by regimen (continuous (daily), intermittent (two or three times per week), or pulsed (daily for five days every four weeks)). Meta‐analysis was performed only when study populations were sufficiently similar for pooling to make sense (Herath 2018). As Threapleton 2018 identified insufficient studies, it was not possible to perform meta‐analyses.

NMA

We considered all treatment dosages as individual treatments. We used a class model approach for the NMA (Dias 2018; Kew 2014). We pre‐specified five classes of interventions in the network: macrolides (e.g. azithromycin, erythromycin, roxithromycin), quinolones (ciprofloxacin, moxifloxacin), tetracyclines (doxycycline), combined tetracyclines/macrolides (e.g. roxithromycin/doxycyline), and placebo. We compared models that assumed all interventions within a class had the same effect to models for which effects within a class were exchangeable (i.e. similar) using the deviance information criterion (DIC) and taking into account any changes in estimated heterogeneity. We presented estimates for within‐class variability in treatment effects, as well as between‐class variability in treatment effects, when applicable. We also presented the ranking of each class in one of the five positions (from best to worst).

Subgroup analysis and investigation of heterogeneity

Pair‐wise meta‐analyses

Herath 2018 planned to carry out subgroup analysis for the primary outcome (number of exacerbations) by exploring severity of COPD according to FEV₁ and GOLD criteria, type of antibiotic, duration of antibiotic use, year of conduct of study, whether the antibiotic was used primarily as an antimicrobial or anti‐inflammatory agent, treatment regimen (dose, frequency, route of administration), and history of exacerbations. Threapleton 2018 explored exacerbation history and COPD severity in studies with 70% or more on long‐acting beta‐adrenoceptor agonist/long‐acting muscarinic antagonist/inhaled corticosteroid (LABA/LAMA/ICS) at baseline versus those with less than 70% on LABA/LAMA/ICS at baseline.

NMA

We planned to undertake a flexible and exploratory approach to investigate heterogeneity, depending on the data found. In the event of significant heterogeneity in the NMA, we considered exploring heterogeneity using pre‐specified factors, if extractable.

  • Exacerbation history: trials that recruit participants with a group mean < 1 versus 2 to 3 or 4 or more exacerbations in the preceding year.

  • COPD severity: participants predominantly classed as GOLD 1 or 2 versus those predominantly classed as GOLD 3 or 4.

  • Trials with ≥ 70% of participants on long‐acting beta‐agonists (LABAs) or long‐acting muscarinic receptor agonists (LAMAs) or inhaled corticosteroids (ICSs) at baseline.

  • Pseudomonas colonisation: trials that recruited participants colonised with Pseudomonas at baseline versus those not colonised with Pseudomonas at baseline.

  • Methodological issues with randomisation, allocation concealment, participant/personnel blinding, outcome assessor blinding, and attrition.

If data were insufficient for assessment of the pre‐specified factors, we planned to investigate differences (if any) by extracting key severity criteria for each trial, and to summarise data across pair‐wise comparisons.

Sensitivity analysis

Pair‐wise analyses

Herath 2018 and Threapleton 2018 planned to conduct a sensitivity analysis on the primary outcome (people with one or more exacerbations) by removing studies at high risk or unclear risk of sequence generation, allocation concealment, or blinding. Threapleton 2018 also planned to remove cross‐over studies. Herath 2018 used a random effects model for outcome measures.

NMA

We performed sensitivity analyses by primarily excluding from the main analysis studies that were at high risk of bias, then including these studies in a sensitivity analysis.

Reporting biases

We did not investigate reporting bias, as studies were too few for a contour‐adjusted funnel plot to be prepared.

Summary of findings and assessment of the certainty of the evidence

‘Summary of findings' tables were created for the following outcomes: exacerbations, quality of life (SGRQ) and serious adverse events. Judgement of the quality of the evidence was based on the ‘Risk of bias' assessment of included trials, estimates of heterogeneity, and assessment of model fit inconsistency.

Results

Description of studies

Results of the search

Pair‐wise meta‐analysis results

Herath 2018 included nine new studies from a search of 265 additional references for the 2018 update. The previous version of the review included seven studies, resulting in a total of 16 studies included in the review (Albert 2011; Banerjee 2005; Berkhof 2013; Brill 2015; He 2010; Mygind 2010; NCT00524095; NCT02628769; Seemungal 2008; Sethi 2010; Shafuddin 2015; Simpson 2014; Suzuki 2001; Tan 2016; Uzun 2014; Wang 2017). Fourteen studies (N = 3932 participants) were included in the pair‐wise meta‐analyses (Herath 2018).

Threapleton 2018 included for analysis two eligible studies (N = 391) from a search of 1415 references (Brill 2015; Shafuddin 2015).

Further details about the study characteristics of both reviews can be found in Characteristics of included studies, and a summary of the results can be found in Appendix 3. From this point onwards, we will describe the results of the NMA only.

NMA results

We identified 1120 records through database searching, which included the original search in 2018 and updated searches in 2019 and January 2020. We screened all 1120 records in the absence of any duplicate records. We excluded 1052 records on the basis of titles and abstracts, which resulted in 68 full texts to be assessed for eligibility. From the full‐text assessment, we identified 38 manuscripts, reporting on 17 studies that were eligible to be included in the review. The PRISMA flow diagram shows how the final selection of studies was made (Figure 1).

Figure 1
Study flow diagram.

Study flow diagram.

Included studies

NMA included studies

We identified 17 trials that were eligible for inclusion in this review (Albert 2011; Banerjee 2005; Berkhof 2013; Blasi 2010; Brill 2015; He 2010; Mygind 2010; Seemungal 2008; Sethi 2010; Shafuddin 2015; Simpson 2014; Singh 2019; Suzuki 2001; Tan 2016; Uzun 2014; Vermeersch 2019; Wang 2017). Details of each study can be found in Characteristics of included studies. Of these, two were multi‐arm studies from which we could make direct comparisons among all included antibiotic classes (Brill 2015; Shafuddin 2015). We also included 12 studies that compared a single antibiotic with placebo (Albert 2011; Berkhof 2013; Blasi 2010; He 2010; Seemungal 2008; Sethi 2010; Simpson 2014; Singh 2019; Suzuki 2001; Tan 2016; Uzun 2014; Vermeersch 2019).

Of the 17 trials, three studies were not included in the NMA (Banerjee 2005; Mygind 2010; Wang 2017). Banerjee 2005 reported SGRQ symptom score in a format that could not be included in the network; therefore, this was reported as a pair‐wise analysis. Mygind 2010 was a conference abstract for which we could not obtain any further data when we contacted study authors. Wang 2017 did not include data relevant to our outcome criteria.

Two of the studies were not eligible for inclusion in the main NMA analyses but were included in sensitivity analyses (Simpson 2014; Singh 2019). Therefore, of the 12 studies included in the main NMA analysis, a total of 3405 participants with a diagnosis of COPD were randomly assigned to 16 treatment arms of interest (including placebo) (Table 2).

Open in table viewer
Table 2. Treatments and corresponding abbreviations and classes

Treatment

Abbreviation

Class

Placebo

Pbo

NA

Azithromycin 250 mg once daily

AZM250 once daily

Macrolide

Azithromycin 250 mg once daily 3 times per week

AZM250 once daily (3x weekly)

Macrolide

Azithromycin 500 mg once daily 3 times per week

AZM500 once daily (3x weekly)

Macrolide

Azithromycin 500 mg once daily 3 times per month

AZM500 once daily (3x monthly)

Macrolide

Azithromycin 500 mg once daily (for first 3 days),

azithromycin 250 mg every 2 days (intermittent)

for rest of treatment duration

AZM500 once daily (3 days) then

AZM250 (alternating day days)

Macrolide

Clarithromycin 500 mg once daily

CLR500 once daily

Macrolide

Erythromycin 250 mg three times daily

ERY250 three times daily

Macrolide

Erythromycin 250 mg twice daily

ERY250 twice daily

Macrolide

Erythromycin 125 mg 3 times daily

ERY125 three times daily

Macrolide

Erythromycin 200 to 400 mg once daily

ERY200/400 once daily

Macrolide

Roxithromycin 300 mg once daily

ROX300 once daily

Macrolide

Doxycycline 100 mg once daily

DOX100 once daily

Tetracycline

Roxithromycin 300 mg once daily +

Doxycycline 100 mg once daily

ROX300 once daily + DOX100

once daily

Macrolide + tetracycline

Moxifloxacin 400 mg once daily

(5 days every 4 weeks)

MOX400 once daily

(5 days every 4 weeks)

Quinolone

Moxifloxacin 400 mg once daily

(5 days every 8 weeks)

MOX400 once daily

(5 days every 8 weeks)

Quinolone

Abbreviations

AZM: azithromycin; CLR: clarithromycin; ERY: erythromycin; DOX: doxycycline; MOX: moxifloxacin; NA: not applicable; Pbo: placebo; ROX: roxithromycin.

Baseline characteristics of participants in trials included in the NMA

We found that baseline characteristics of trial populations were fairly similar across all studies, and most trial participants had moderate to severe disease. Mean age was similar, ranging from 64 years to 73 years, and considerably more males than females were recruited. Lung function, specifically mean FEV₁, ranged from 0.935 to 1.36 L. Mean pack‐years ranged from 36 to 59 across all studies, and overall there were no serious concerns that there was any imbalance in characteristics expected to modify relative treatment effects. The mean number of exacerbations among trial participants in the 12 months before trial start ranged from two to five (Blasi 2010; Brill 2015; Seemungal 2008; Shafuddin 2015; Uzun 2014; Table 1). Berkhof 2013 reported a median of one exacerbation in the previous 12 months among study participants. In Albert 2011, 50% of study participants were hospitalised or visited the emergency department 12 months before the trial start.

Characteristics of interventions in the NMA
Studies included in the NMA

Across the 12 studies included in the NMA, 15 antibiotics were evaluated and categorised into four classes: macrolides, tetracycline, quinolone, and macrolide plus tetracycline. A description of all studies is found in Characteristics of included studies and in Table 2.

Excluded studies

Fourteen excluded studies are listed in the Characteristics of excluded studies table, along with reasons for exclusion.

Risk of bias in included studies

Judgements for risk of bias and reasons can be found in the Characteristics of included studies table, and an overview of judgements for risk of bias can be found in Figure 2.

Figure 2
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

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

Allocation

Random sequence generation

Of the included studies, 12 were judged as having low risk of bias for randomisation sequence generation. We judged five studies as having unclear risk, as they did not report methods for the randomisation process (He 2010; Mygind 2010; Sethi 2010; Singh 2019; Tan 2016).

Allocation concealment

We assessed 10 studies as having low risk of bias for allocation concealment. Seven studies were judged as having unclear risk (Berkhof 2013; Blasi 2010; He 2010; Mygind 2010; Sethi 2010; Tan 2016; Wang 2017), as no further information about the treatment allocation process was provided.

Blinding

Blinding of participants and personnel

Eleven studies were judged as having low risk of performance bias (Albert 2011; Banerjee 2005; Berkhof 2013; He 2010; Mygind 2010; Seemungal 2008; Sethi 2010; Shafuddin 2015; Simpson 2014; Uzun 2014; Vermeersch 2019). Blasi 2010, Singh 2019, Suzuki 2001, and Wang 2017 were open‐label trials and therefore were judged to be at high risk of bias. Tan 2016 did not provide any information about blinding of participants or personnel; therefore, we assumed that this was an open‐label study. Brill 2015 was judged as having unclear risk of bias for this domain because it is not clear whether study personnel were blinded.

Blinding of outcome assessors

Seven studies were judged as having low risk of bias for outcome assessment (Albert 2011; Berkhof 2013; Seemungal 2008; Shafuddin 2015; Simpson 2014; Uzun 2014; Vermeersch 2019). Three studies were rated as having high risk of bias (Blasi 2010; Singh 2019; Suzuki 2001), as they were open‐label trials with no documentation regarding blinding of outcome assessors. Brill 2015, Tan 2016, and Wang 2017 were also judged as having high risk of bias, as blinding of outcome assessors was not described in either study. No further information about blinding in this domain was described; therefore, Banerjee 2005, He 2010, Mygind 2010, and Sethi 2010 were judged as having unclear risk of bias.

Incomplete outcome data

Thirteen studies provided adequate descriptions of outcomes of study participants (Albert 2011; Banerjee 2005; Berkhof 2013; Brill 2015; He 2010; Seemungal 2008; Shafuddin 2015; Simpson 2014; Singh 2019; Suzuki 2001; Tan 2016; Uzun 2014; Vermeersch 2019), as they had described the flow of participants throughout the trial using a CONSORT diagram or by including this information in a specific paragraph or table. Blasi 2010 was judged as having high risk of bias because the analysis was not intention‐to‐treat, and participants who died were not included in the analysis. This may have led to an overestimation of beneficial outcomes.

Withdrawal rates in treatment and control arms of most studies were similar, with the exception of four studies that were judged as having unclear risk of bias (Albert 2011Sethi 2010; Shafuddin 2015; Tan 2016). Albert 2011 and Sethi 2010 did not report reasons for missing health‐related quality of life data. Shafuddin 2015 was judged as having unclear risk of bias for this domain, as more participants dropped out of the combined antibiotic treatment arm compared to the single antibiotic and placebo arms, although all participants were included in the intention‐to‐treat analysis. Tan 2016 was judged as having unclear risk of bias, as details of the number of people analysed at each time point were not reported.

Mygind 2010 provided limited information, as it was a conference abstract of unpublished data; therefore we judged this source as having unclear risk of bias. Wang 2017 did not provide any further information about missing data; therefore we judged this study as having unclear risk of bias for this domain.

Selective reporting

We judged 13 studies as having low risk of bias for this domain (Albert 2011; Banerjee 2005; Berkhof 2013; Brill 2015; He 2010; Mygind 2010; Seemungal 2008; Sethi 2010; Shafuddin 2015; Simpson 2014; Suzuki 2001; Uzun 2014; Vermeersch 2019). We judged Blasi 2010 as having high risk as outcomes in the publication were reported differently from those in the protocol on the trial registry website. SAEs were reported only in the antibiotic arm and not in the control arm; therefore it is not clear whether participants in the control group had any SAEs. We also judged Wang 2017 as having high risk of bias, as no prospective trial registration or protocol was identified, and dyspnoea grade was reported as measured in the abstract but there was no description in the methods or results of the publication. We judged two studies as having unclear risk of bias (Singh 2019; Tan 2016), as it was not clear if the outcomes were reported as planned.

No prospective trial registration or protocol was identified.

Other potential sources of bias

We did not consider industry sponsorship as necessarily increasing the risk of bias when studies were well designed. We did not identify any other potential sources of bias.

Effects of interventions

See: Summary of findings 1 Summary of findings: exacerbations; Summary of findings 2 Summary of findings: change from baseline in SGRQ; Summary of findings 3 Summary of findings: serious adverse events

Network meta‐analysis outcomes

In general, the NMA supported a common class effect for exacerbations and quality of life. Individual prophylactic antibiotics included in each NMA are provided in tables that have been referenced in the text below.

NMA 1. Primary outcome: exacerbations

Definitions of exacerbations reported in the studies included for this outcome are captured in Table 1. Overall, moderate exacerbations were described as sustained worsening of baseline respiratory symptoms for at least two days requiring treatment with systemic corticosteroids and/or antibiotics, with severe exacerbations requiring additional hospital admission.

Exacerbation data were reported either as time to first exacerbation (Albert 2011; Blasi 2010; He 2010; Seemungal 2008; Simpson 2014; Uzun 2014), or as the number of people with one or more exacerbations during the study period (Berkhof 2013; Brill 2015; Sethi 2010; Suzuki 2001) (Table 3; Table 4).

Open in table viewer
Table 3. Exacerbations: studies included with time to first exacerbation data

Study

Treatments compared

Log hazard ratio

Standard error

Albert 2011

Pbo

AZM 250 mg once daily

‐0.31

0.07

He 2010

Pbo

ERY125 mg 3 times daily

‐0.59

0.29

Seemungal 2008

Pbo

ERY 250 mg twice daily

‐0.45

0.14

Simpson 2014

Pbo

AZM 250 mg once daily

‐0.99

0.62

Uzun 2014

Pbo

AZM 500 mg once daily 3 times per week

‐0.54

0.16

Blasi 2010

Pbo

AZM 500 mg once daily 3 times per week

‐1.69

0.60

Abbreviations

AZM: azithromycin; ERY: erythromycin; Pbo: placebo.

Open in table viewer
Table 4. Exacerbations: studies included with the number of people with one or more exacerbations

Study

Treatment 1 (N)

No. of events

Treatment 2 (N)

No. of events

Treatment 3 (N)

No. of events

Treatment 4 (N)

No. of events

Berkhof 2013

Pbo

(42)

17

AZM 250 mg once daily 3 times a week

(42)

10

‐

‐

‐

‐

Brill 2015

Pbo

(24)

13

DOX100 mg once daily

(25)

15

AZM 250 mg once daily

3 times per week

(25)

10

MOX 400 mg once daily

(5 days every 4 weeks)

(25)

10

Sethi 2010

Pbo

(580)

295

MOX 400 mg once daily

(5 days every 8 weeks)

(569)

269

‐

‐

‐

‐

Suzuki 2001*

Pbo

(54)

30

ERY 200 to 400 mg once daily

(55)

6

‐

‐

‐

‐

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin;Pbo: placebo.

*This study was included only as a sensitivity analysis ‐ reported in Appendix 4.

Model selection

Both fixed class models with fixed and random treatment effects fit well. They had similar DIC values; therefore the simpler fixed effect model was chosen, although results for the random effects model were also displayed for comparison (DIC 52.17, SD 0.16, 95% CrI 0.006 to 0.519) (Table 5).

Open in table viewer
Table 5. Exacerbations: model fit statistics

DIC

SD (95% CrI)

Total residual deviance*

Fixed class effect models

Fixed effect model

51.31

‐

15.17

Random effects model

52.17

0.16 (0.006 to 0.519)

13.61

Abbreviations

CrI: credible interval; DIC: deviance information criterion; SD: standard deviation.

*Compared to 14 data points.

Results

The NMA included nine studies and nine interventions from three antibiotic classes (macrolides, quinolones, tetracyclines) and from control treatment (placebo or standard therapy) (2732 participants; Table 3; Table 4; Table 6). Figure 3 represents studies contributing to the NMA (a) at the individual intervention level and (b) at the antibiotic class level. summary of findings Table 1 shows the hazard ratio (HR) for each class compared to every other. Each class except tetracycline (HR 1.29, 95% CrI 0.66 to 2.41) reduced exacerbations compared to control (placebo or standard therapy). Evidence suggests that macrolides considerably reduced exacerbations compared to placebo or standard therapy (HR 0.67, 95% CrI 0.60 to 0.75), whereas quinolones showed smaller benefit and the 95% CrI included no effect (HR 0.89, 95% CrI 0.75 to 1.04). Furthermore, our analysis suggests that macrolides were superior to quinolones in reducing exacerbations (quinolone versus macrolide; HR 1.32, 95% CrI 1.08 to 1.61) (Table 7). Figure 4 presents the hazard ratios for both fixed effect and random effects models.

Open in table viewer
Table 6. Exacerbations: interventions and treatment classes

Intervention

Treatment class

N

1

Pbo

Placebo

1345

2

AZM 250 mg once daily

Macrolide

573

3

AZM 250 mg once daily 3 times per week

Macrolide

67

4

AZM 500 mg once daily 3 times per week

Macrolide

57

5

ERY 250 mg 3 times daily

Macrolide

53

6

ERY 125 mg 3 times daily

Macrolide

18

7

DOX 100 mg once daily

Tetracycline

25

8

MOX 400 mg once daily (5 days every 8 weeks)

Quinolone

569

9

MOX 400 mg once daily (5 days every 4 weeks)

Quinolone

25

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; Pbo: placebo.

Figure 3
Exacerbations: network diagram of interventions and classes. Treatment abbreviations are defined in . The size of the nodes is proportionate to the number of participants assigned to the intervention. The thickness of the lines is proportionate to the number of randomised trials that studied the respective comparison.

Exacerbations: network diagram of interventions and classes. Treatment abbreviations are defined in Table 1. The size of the nodes is proportionate to the number of participants assigned to the intervention. The thickness of the lines is proportionate to the number of randomised trials that studied the respective comparison.

Open in table viewer
Table 7. Exacerbations: number of trials, number of participants, and relative effect estimates for all class comparisons

Comparison

Hazard ratios

Number of trials

N

Intervention

Comparator

Median

95% CrI

Macrolide

Placebo

0.67

0.60 to 0.75

9

1509

Tetracycline

Placebo

1.29

0.66 to 2.41

1

49

Quinolone

Placebo

0.89

0.75 to 1.04

2

1198

Tetracycline

Macrolide

1.93

0.99 to 3.62

1

50

Quinolone

Macrolide

1.32

1.08 to 1.61

1

50

Quinolone

Tetracycline

0.69

0.37 to 1.34

1

50

Abbreviations

CrI: credible interval.

Figure 4
Exacerbations: forest plot of relative effects for each class comparison. Values less than 1 favour the first names class.

Exacerbations: forest plot of relative effects for each class comparison. Values less than 1 favour the first names class.

Table 8 shows rank statistics for the three antibiotic classes and for control (placebo or standard therapy). The highest ranked class was macrolide, with a median rank of 1 (95% CrI first to second), followed by quinolone (95% CrI second to third). Tetracycline was the worst ranked treatment for this outcome (95% CrI first to fourth). Although control (placebo or standard therapy) was ranked third, it had a 95% CrI of second to fourth, and a similar mean rank to tetracycline (3.1 versus 3.6), which reflected that the HR showed no clear evidence of a difference between tetracycline and control (placebo or standard therapy) (HR 1.29, 95% CrI 0.66 to 2.41) (Figure 4; Table 8).

Open in table viewer
Table 8. Exacerbations: number of participants and rank statistics for each class (sorted by mean rank)

Class

N

Mean

Median

95% CrI

Macrolide

768

1.0

1

1 to 2

Quinolone

594

2.2

2

2 to 3

Placebo

1345

3.1

3

2 to 4

Tetracycline

25

3.6

4

1 to 4

Abbreviations

CrI: credible interval.

Figure 5 represents the rank probabilities of each antibiotic class and control (placebo or standard therapy). The vertical axis shows the probability of being ranked best (first) to worst (fourth). The probability of macrolides being ranked first was 0.97 (Figure 5).

Figure 5
Exacerbations: plot of rank probabilities for each class.

Exacerbations: plot of rank probabilities for each class.

The absolute rate of exacerbations per person per year for each treatment class is reported in summary of findings Table 1, with the assumption that the absolute rate of exacerbations in the control (placebo or standard therapy) arm was two per person per year. Macrolides had a median rate of exacerbations of 1.34 (CrI 1.19 to 1.50) per person per year compared to control (placebo or standard therapy). Tetracycline was the only class that had a higher median rate of exacerbations per person per year than control (placebo or standard therapy) (2.58, 95% CrI 1.33 to 4.81), suggesting a probable lack of clinical effectiveness.

Threshold analysis and robustness of the evidence

We judged studies contributing to this analysis to be at low risk of bias in most domains. Figure 6 shows the forest plot for the threshold analysis. None of the comparisons had the upper or lower portion of the invariant interval within the 95% CrI of the effect estimate (Figure 6); thus the decision was not sensitive to the level of imprecision in this estimate, that is, the decision that the optimal treatment class to prevent exacerbations is macrolide was robust to sampling variation. However, this decision appeared sensitive to potential bias in comparisons of all classes to placebo (macrolide versus placebo, tetracycline versus placebo, quinolone versus placebo; Figure 6), as some of the thresholds were small. Upon inspection of these thresholds, we noted that only the comparison of macrolides to control (placebo or standard therapy) (2 versus 1; Figure 6) could change due to plausible bias adjustment. This is so because if there was any bias in the comparison, it was likely to favour the active class; therefore any adjustment would bring the estimated relative effect closer to the null value, meaning that quinolone may become the best class to prevent exacerbations. All other comparisons would require a bias that favoured placebo to be present, or an implausibly large bias before the optimal treatment changed. All other invariant intervals were very wide, so comparisons were robust to any changes in the evidence informing those comparisons.

Figure 6
Exacerbations: forest plot with threshold analysis for the log‐HR of exacerbations for each class. Base case optimal treatment set is 2. Class codes: 1 = placebo; 2 = macrolide; 3 = tetracycline; 4 = quinolone. Comparisons are macrolide versus placebo (2 versus 1); tetracycline versus placebo (3 versus 1); quinolone versus placebo (4 versus 1); tetracycline versus macrolide (3 versus 2); quinolone versus macrolide (4 versus 2); quinolone versus tetracycline (4 versus 3).

Exacerbations: forest plot with threshold analysis for the log‐HR of exacerbations for each class. Base case optimal treatment set is 2. Class codes: 1 = placebo; 2 = macrolide; 3 = tetracycline; 4 = quinolone. Comparisons are macrolide versus placebo (2 versus 1); tetracycline versus placebo (3 versus 1); quinolone versus placebo (4 versus 1); tetracycline versus macrolide (3 versus 2); quinolone versus macrolide (4 versus 2); quinolone versus tetracycline (4 versus 3).

Exacerbations: sensitivity analysis including Suzuki 2001

In an initial analysis, treatment‐specific comparisons in a fixed effect‐no class model show significantly lower risk of exacerbations, which resulted in macrolides as the highest ranking class‐specific treatment (Appendix 4; Table 6; online supplement Janjua 2021) and a probability of 0.96 for being ranked first. Due to suspicion of increased heterogeneity in the effects of macrolides compared to control treatment and other antibiotic treatments, we investigated further the relative estimates for all treatment comparisons in both fixed effect‐no class and fixed effect‐exchangeable class models. The risk of exacerbations was considerably lower with erythromycin 200 to 400 mg compared with placebo or other antibiotic treatments (Table 4; Appendix 4). We identified Suzuki 2001 as the study contributing to observed effects and investigated characteristics of the study that may have possibly contributed to the analysis result (Table 1). Suzuki 2001 was an open‐label study that was rated at high risk of bias due to lack of blinding. Study authors reported erythromycin dosage as ranging from 200 mg to 400 mg. Unfortunately, they provided no further information in their publication regarding the actual dose of erythromycin that participants received, and there was no possibility of contacting these study authors because the study was published in 2001. For these reasons, we decided to include Suzuki 2001 in a sensitivity analysis rather than in the main analysis.

NMA 2. Primary outcome: quality of life: change from baseline in SGRQ score
Model selection

Fixed class models with both fixed and random treatment effects fit well. Sensitivity to the prior distribution for heterogeneity was assessed by fitting a random effects model with an empirically based prior distribution converted to the mean difference scale (Rhodes 2015). We chose the fixed treatment effect model with fixed class effect (Table 9). We also reported results for the random treatment effects model with fixed class effect using the uniform prior distribution.

Open in table viewer
Table 9. Change from baseline in SGRQ: number of trials, number of participants, and relative effects for all class comparisons

Comparison

Number of trials

N

Fixed effects‐fixed class effect

Random effects‐fixed class effect (uniform prior)

Random effects‐fixed class effect (empirical prior)

MD

95% CrI

MD

95% CrI

MD

95% CrI

Macrolide vs placebo

6

1158

‐2.30

‐3.61 to ‐0.99

‐2.34

‐4.28 to ‐0.39

‐2.28

‐5.19 to 1.01

Tetracycline vs placebo

1

49

1.18

‐1.49 to 3.84

1.14

‐2.47 to 4.62

1.20

‐4.62 to 7.19

Quinolone vs placebo

2

1078

‐1.33

‐2.97 to 0.32

‐1.42

‐4.04 to 1.05

‐1.44

‐5.99 to 3.07

Tetracycline vs macrolide

1

50

3.47

0.92 to 6.03

3.47

0.01 to 6.83

3.47

‐2.38 to 9.22

Quinolone vs macrolide

1

50

0.97

‐0.10 to 2.95

0.91

‐2.01 to 3.71

0.84

‐4.24 to 5.51

Quinolone vs tetracycline

1

50

‐2.50

‐5.32 to 0.30

‐2.56

‐6.33 to 1.16

‐2.63

‐8.96 to 3.37

Abbreviations

CrI: credible interval;MD: mean difference; SGRQ: St George's Respiratory Questionnaire.

Results

All of the included studies were two‐arm studies except for one study, which had four treatment arms (Brill 2015). Mean differences across the four study arms are correlated; thus a co‐variance between mean differences ‐ V ‐ was required (Dias 2018; Franchini 2012). As this was not reported, we intended to use sampling error (SE²) for mean SGRQ at baseline for participants randomised to placebo (V = 15.04), and to use SE² for the change from baseline (assuming equal baseline and follow‐up variances and correlation of 0.7; V = 9.025) in a sensitivity analysis. However, only the latter (V = 9.025) produced a valid co‐variance matrix for observed mean differences; therefore, only this value was used.

The NMA included seven studies of eight interventions from three antibiotic classes (macrolide, quinolone, tetracycline) and placebo for this outcome (2237 participants; Table 9; Table 10). Figure 7 represents studies contributing to the NMA (a) at the individual intervention level and (b) at the antibiotic class level. Figure 8 shows the mean difference in change from baseline in SGRQ score for each class compared to every other for the preferred model (fixed treatment‐fixed class model), as well as the random treatment‐fixed class model for comparison. Evidence suggests that the macrolide class improved SGRQ score compared to placebo in both models, although only the main analysis yielded significant results (Figure 8; Table 9).

Open in table viewer
Table 10. Change from baseline in SGRQ: included studies

Study

Endpoint (weeks)

Treatments compared

Mean difference vs Placebo

SE of Mean difference

Albert 2011

52

Placebo

AZM 250 mg once daily

‐2.2

0.7853

Berkhof 2013

12

Placebo

AZM250 mg once daily 3 times per week

‐7.5

2.5456

He 2010

26

Placebo

ERY 125 mg 3 times daily

‐3

5.6801

Sethi 2010

48

Placebo

MOX 400 mg once daily (5 days every 8 weeks)

‐1.2

0.9231

Simpson 2014

12

Placebo

AZM 250 mg once daily

6.1

5.31927

Uzun 2014

52

Placebo

AZM500 mg once daily 3 times per week

‐0.61

2.622449

Brill 2015*

13

Placebo

a. DOX 100 mg once daily

b. AZM 250 mg once daily 3 times per week

c. MOX 400 mg once daily (5 days every 4 weeks)

a. 1.02

b. ‐2.29

c. ‐1.88

a. 3.135

b. 3.212

c. 3.426

a. 0.88

b. ‐2.35

c. ‐2.25

a. 3.132

b. 3.085

c. 3.233

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; SE: standard error; SGRQ: St George's Respiratory Questionnaire.

*Data in bold used in sensitivity analysis.

Figure 7
Quality of life: SGRQ network map.

Quality of life: SGRQ network map.

Figure 8
Change from baseline in SGRQ: forest plot of relative effects for each class comparison. Values less than 0 favour the first named class.

Change from baseline in SGRQ: forest plot of relative effects for each class comparison. Values less than 0 favour the first named class.

Bayesian probabilities of the mean difference exceeding the minimal clinically important difference (MCID) of 4 for SGRQ, when a macrolide is compared to placebo, tetracycline, or quinolone, were calculated as 0.005, 0.345, and 0.001, respectively, under the fixed treatment effect model with fixed class effect.

Table 11 shows rank statistics for the four classes. The highest ranked class was macrolide, with a median rank of 1 (95% CrI first to second). Figure 9 shows plots of rank probabilities for each class. The vertical axis shows the probability of being ranked best (first) to worst (fourth). The probability of macrolides being ranked first was 0.82 (Figure 9).

Open in table viewer
Table 11. Change from baseline in SGRQ: number of participants and rank statistics for each class (sorted by mean rank)

Treatment class

Number of participants

Mean

Median

95% CrI

Macrolide

578

1.17

1

1 to 2

Quinolone

528

1.93

2

1 to 3

Placebo

1106

3.14

3

2 to 4

Tetracycline

25

3.76

4

2 to 4

Abbreviations

CrI: credible interval; SGRQ: St. George's Respiratory Questionnaire.

Figure 9
Change from baseline in SGRQ: plot of rank probabilities for each class.

Change from baseline in SGRQ: plot of rank probabilities for each class.

We presented the change in SGRQ from baseline for macrolides, tetracyclines, and quinolones compared with placebo in summary of findings Table 2. We assumed that the absolute change from baseline for SGRQ in the placebo arm was ‐1.7, for a 1.7 point improvement. The absolute change in SGRQ from baseline with macrolide treatment was ‐4.00 (95% CrI ‐5.51 to ‐2.68), which translated to a 2.30 point improvement compared to placebo (3.61 to 0.99 point improvement). However this did not reach clinical significance (MCID of 4 point improvement). With quinolone treatment, the absolute change from baseline was ‐3.03 (95% CrI ‐4.69 to ‐1.37). This resulted in a 1.33 point improvement compared to placebo (2.99 to 0.33 point improvement). With tetracyclines, there was an absolute change in SGRQ of ‐0.52 (95% CrI ‐3.21 to 2.16), which resulted in a 1.18 point worsening in quality of life compared to placebo (1.51 improvement to 3.86 worsening). Thus, tetracycline was worse than placebo, but there was still improvement compared to baseline.

Quality of life: change from baseline in SGRQ score: sensitivity analysis including Brill 2015

Results of the sensitivity analysis are detailed in Appendix 5. The main NMA included fully adjusted effect estimates from Brill 2015; however, the study also reported relative effects, which were adjusted only for baseline values. With a fixed treatment‐fixed class model, the relative effect for SGRQ score in the sensitivity analysis was similar to the main NMA result and ranking; however, the 95% CrI in the sensitivity analysis no longer included zero when quinolone was compared to tetracycline.

Quality of life: change from baseline in SGRQ: sensitivity analysis including Singh 2019

Results of the sensitivity analysis can be found in Appendix 5. We included Singh 2019 in a sensitivity analysis rather than in the main analysis. Inclusion of Singh 2019 in the sensitivity analysis using a fixed effect‐fixed class model, with effect estimates of each class comparison, including tetracycline, led to a shift when compared to the main analysis (Figure 6; online supplement Janjua 2021, Figure 4.3. For example, the effect estimate for tetracycline versus placebo shifted to the right in the sensitivity analysis, as did the effect estimate for the comparison of tetracycline compared to macrolide (online supplement Janjua 2021; Figure 4.3). Ranking also changed, resulting in tetracycline ranking third and placebo now ranked fourth, in contrast to the main NMA (Appendix 5).

NMA 3. Primary outcome: serious adverse events
Model selection

Fixed class models with both fixed and random treatment effects fit well; thus we chose a fixed treatment effect model with fixed class effect. Results for the random effects model were also displayed for comparison (DIC 113.2, SD 0.44, 95% CrI 0.02 to 1.28) (Table 12). We also reported results for the random treatment effects model with fixed class effect for comparison using the uniform prior distribution (Figure 10; Table 13) and the empirical prior distribution (Table 13) for comparison.

Open in table viewer
Table 12. Serious adverse events: model fit statistics

DIC

Between‐study SD (95% CrI)

Total residual deviance*

Fixed class models

Fixed treatment effect

113.8

‐

21.58

Random treatment effects

113.2

0.44 (0.02 to 1.28)

18.59

Abbreviations

* Compare to 19 data points

CrI: credible interval; DIC: deviance information criterion; SD: standard deviation.

Figure 10
Serious adverse events: forest plot of relative effects for each class comparison. Values less than 1 favour the first named class.

Serious adverse events: forest plot of relative effects for each class comparison. Values less than 1 favour the first named class.

Open in table viewer
Table 13. Serious adverse events: number of trials, participants, and relative effects for all class comparisons

Treatment class comparison

Number of trials

N

Fixed effect‐fixed class effect model

Random effects‐fixed class effect (uniform prior)

Random effects‐fixed class effect (empirical prior)

OR

95% CrI

OR

95% CrI

OR

95% CrI

Macrolide vs placebo

8

1930

0.76

0.62 to 0.93

0.72

0.38 to 1.14

0.73

0.45 to 1.07

Quinolone vs placebo

1

1149

1.00

0.72 to 1.34

1.21

0.29 to 3.24

1.08

0.42 to 2.27

Macrolide + tetracycline vs placebo

1

195

0.97

0.52 to 1.66

1.12

0.27 to 2.84

1.00

0.36 to 2.19

Quinolone vs macrolide

0

0

1.32

0.90 to 1.89

1.88

0.41 to 5.67

1.56

0.55 to 3.62

Macrolide + tetracycline vs macrolide

1

198

1.28

0.68 to 2.19

1.67

0.42 to 4.31

1.41

0.52 to 3.15

Macrolide + tetracycline vs quinolone

0

0

1.00

0.49 to 1.82

1.72

0.17 to 4.67

1.13

0.26 to 3.11

Abbreviations

CrI: credible interval; OR: odds ratio.

Results

The NMA included nine studies, nine interventions from three antibiotic classes (macrolide, macrolide plus tetracycline, quinolone), and placebo for this outcome (3180 participants; Table 14; Table 15). Figure 11 represents studies contributing to the NMA (a) at the individual intervention level and (b) at the antibiotic class level. Brill 2015 was not included in the analysis, as no events were reported in any treatment arm. We presented the odds ratio (OR) of serious adverse events for each class compared to every other (Table 13). Evidence suggests that the macrolide class reduced the odds of having a serious adverse event compared to placebo when the fixed treatment effect model with fixed class effects was used (Table 13; Figure 10). However, use of the random treatment effects model with fixed class effect resulted in the 95% CrI crossing the line of no effect (Table 13; Figure 10).

Open in table viewer
Table 14. Serious adverse events: table of interventions and treatment classes

Intervention

Treatment class

N

1

Pbo

Pbo

1539

2

AZM 250 mg od

Macrolide

573

3

ERY 125 mg tds

Macrolide

54

4

ERY 250 mg bd

Macrolide

53

5

MOX 400 mg od (5 days every 8 weeks)

Quinolone

569

6

ROX 300 mg od + DOX 100 mg od

Macrolide + tetracycline

101

7

ROX 300 mg od

Macrolide

97

8

AZM 500 mg od (3x weekly)

Macrolide

47

9

AZM 500 mg od (for first 3 days),

AZM 250 mg every 2 days (intermittent)

for rest of treatment duration

Macrolide

147

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; Pbo: placebo; ROX: roxithromycin.

Open in table viewer
Table 15. Serious adverse events: studies included

Study name

Treatments compared

Number of participants

Number of events

Treatment 1

Treatment 2

Treatment 3

Treatment 1

Treatment 2

Treatment 3

Albert 2011

Pbo

AZM 250 mg once daily

559

558

NA

212

184

NA

He 2010

Pbo

ERY125 mg 3 times daily

18

18

NA

3

2

NA

Seemungal 2008

Pbo

ERY 250 mg twice daily

56

53

NA

12

14

NA

Sethi 2010

Pbo

MOX 400 mg once daily (5 days every 8 weeks)

580

569

NA

97

94

NA

Shafuddin 2015

Pbo

ROX 300 mg once daily

+ DOX 100 mg once daily

ROX 300 mg once daily

94

101

97

20

24

23

Simpson 2014

Pbo

AZM 250 mg once daily

15

15

NA

4

1

NA

Tan 2016

Pbo

ERY 125 mg 3 times daily

18

36

NA

3

2

NA

Uzun 2014

Pbo

AZM 500 mg once daily 3 times per week

45

47

NA

5

3

NA

Vermeersch 2019

Pbo

AZM 500 mg once daily (for first 3 days),

AZM 250 mg every 2 days (intermittent)

for rest of treatment duration

15

147

NA

48

25

NA

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; NA: not applicable; Pbo: placebo; ROX: roxithromycin.

Figure 11
Serious adverse events: network map.

Serious adverse events: network map.

Table 16 shows rank statistics for the four classes. The highest ranked class was macrolide, with a median rank of 1 (95% CrI first to second). Figure 12 represents the plots for ranking probabilities of each class. The vertical axis shows the probability of being ranked best (first) to worst (fourth). The probability of macrolides being ranked first was 0.70 (Figure 12).

Open in table viewer
Table 16. Serious adverse events: total number of participants and rank statistics for each class (sorted by mean rank)

Treatment class

N

Mean

Median

95% CrI

Macrolide

971

1.33

1

1 to 3

Macrolide + tetracycline

101

2.61

2

1 to 4

Quinolone

569

2.95

3

1 to 4

Pbo

1539

3.12

3

2 to 4

Abbreviations

CrI: credible interval; Pbo: placebo.

Figure 12
Serious adverse events: plot of rank probabilities for each antibiotic class

Serious adverse events: plot of rank probabilities for each antibiotic class

The relative effect and the absolute risk difference with macrolide, quinolone, or macrolide plus tetracycline compared with placebo are presented in summary of findings Table 3, with the assumption that the absolute probability of events in the placebo arm was 0.26, and the risk of serious adverse events with placebo was 260 per 1000. A greater magnitude of effect was evident upon treatment with macrolides compared with placebo. The relative effect was OR 0.76 (95% CrI 0.62 to 0.93), that is, 49 fewer people per 1000 experienced a serious adverse event with macrolides compared to placebo. For quinolones or macrolides plus tetracycline, there was uncertainty in effect between classes and placebo, as the 95% credible interval crossed the line of no effect.

Non‐NMA outcomes

Antimicrobial resistance

Antibiotic resistance was investigated by 10 studies (Albert 2011; Banerjee 2005; Berkhof 2013; Blasi 2010; Brill 2015; He 2010; Seemungal 2008; Sethi 2010; Uzun 2014; Vermeersch 2019) (Table 17).

Open in table viewer
Table 17. Drug resistance or microbial sensitivity reported in included studies

Study, drug, duration (weeks)

Drug resistance/microbial sensitivity methods

Results

Conclusion

Albert 2011

AZM (52)

Nasopharyngeal swabs and expectorated sputum samples taken at baseline and every 3 months, assessment for resistance to macrolides. Only 15% of participants were able to produce sputum by the third month; therefore, assessments were limited to nasopharyngeal swabs

Prevalence of resistance to macrolides was 52% and 57%, respectively (P = 0.64)

During the study, 81% AZM and 41% placebo and bacteria were resistant to macrolides (P < 0.001)

People receiving AZM were less likely to be colonised with respiratory pathogens compared to placebo but were more likely to become colonised with macrolide‐resistant organisms. No association with exacerbations

Banerjee 2005

CLR (13)

Sputum sample was tested for potential pathogenic microorganisms: H influenzae, S pneumoniae, M catarrhalis, H parainfluenzae, S aureus, P aeruginosa, K pneumoniae

At the start, 90% of isolates were due to S pneumoniae, H influenzae, M catarrhalis. Some patients had more than one PPM in sputum. After 3 months of CLR, number of people with sputum PPM increased from 12 to 15. Number of bacterial isolates did not increase. In the placebo group, this increased from 10 to 16, and the number of bacterial isolates increased from 15 to 25

CLR did not significantly reduce mean number of H influenzae, S pneumoniae, or M catarrhalis bacterial isolate compared to placebo

No multi‐resistant gram‐negative organisms emerged in the CLR group. CLR did not significantly change the mean CFU number per bacterial isolate compared to placebo

Contradicts other CLR studies that show the opposite due to lack of compliance in the trial. The study did not measure MIC90, which would have been ideal to detect changes in antibiotic susceptibility or resistance with time

Berkhof 2013

AZM (12)

Sputum samples collected

A reduction in respiratory pathogens was seen in the AZM group compared to the placebo group. One patient in the AZM group at 12 weeks had AZM‐resistant bacteria (S aureus)

Dose given seemed effective compared to other studies

Blasi 2010

AZM (26)

Minimum inhibitory concentration (MIC) used to determine bacterial counts

P aeruginosa became resistant to ceftazidime after 6 months of treatment in 1 patient in the AZM group. An ERY‐resistant S pneumoniae was found in 1 patient in the AZM group at 6 months

Not associated with significant effects of reduction in bacterial load or bacterial eradication. Patients with long‐term use of AZM had no resistance except for 1 person

Brill 2015

MOX, DOX, AZM (13)

Resistance to the 3 tested antibiotics, change in sputum bacterial load via quantitative culture (qPCR)

Pulsed MOX demonstrated the largest fall in bacterial numbers on culture but was associated with increased adverse events

Resistance was found in all 3 antibiotic arms of at least 3 times pre‐treatment values. With baseline adjustment of MIC, MOX was associated with an increase in MIC for isolates cultured in sputum compared to placebo (e.g. DOX group were more likely to be resistant to DOX vs placebo)

There was an increase in resistance of airway bacteria to all 3 antibiotics

He 2010

Sputum samples/bacteriology

9 ERY and 7 placebo patients had bacterial growth at baseline. 4 had > 1 organism. At 6 months, there was significant bacterial growth, and 3 specimens had > 1 organism. There was no detection difference in the rate of identifying the 3 main micro‐organisms between the 2 groups

Seemungal 2008

ERY (52)

Sputum samples

Sensitivity testing

H influenzae detection positive in 27% of stable samples and in 40% of exacerbation samples. All H influenzae were resistant to ERY. S pneumoniae was found in 7% and 10%, respectively. No difference in detection rate for any organism between both arms at any follow‐up time points.

Sensitivity testing found that 33/69 showed no growth at baseline. Those who tested positive at baseline were resistant to H influenzae (ERY = 10, P = 12), S pneumoniae (ERY = 1, P = 5 all sensitive), M catarrhalis (ERY = 1, P = 2 all sensitive)

At 12 months, 26/43 samples had no significant growth. Of those samples that were positive, H influenzae (ERY = 1, P = 3), S pneumoniae (ERY = 1 resistant, P = 2 all sensitive), M catarrhalis (2 = sensitive, P = 2)

Microbial resistance was not dependent on the use of ERY. Only 1 case of ERY resistance occurred in the macrolide group at 52 weeks. The number of participant in the study was small; therefore interpretation of these results is not definitive

Sethi 2010

MOX (48)

Sputum samples

Over the 48‐week treatment, there was a reduction in the number of participants with pathogens isolated, with greater reduction with MOX vs placebo. No difference in MIC increases that were sustained

Isolates showed that 1 patient in the MOX group was S pneumoniae resistant at week 40, which was not associated with exacerbations and was not persistent at further visits. For S aureus, 1 to 3 isolates were MOX resistant at baseline and at other time points but did not persist and were not related to exacerbations. Median MIC of MOX against P aeruginosa at 24 weeks was 4 mg/L but was reduced to 1 mg/L to levels at randomisation for the MOX group. Median MIC in placebo group increased from 0.5 to 2 mg/L among those who completed treatment

No further comments

Uzun 2014

AZM (52)

Macrolide resistance by sputum culture

32/47 AZM gave samples. 32/45 in placebo gave samples. Most common bacteria were H influenzae, S pneumoniae, and P aeruginosa. At follow‐up fewer in the AZM group had positive cultures compared to the placebo group

Macrolide resistance was seen in 3 AZM and in 11 placebo (P = 0.036)

The number of sputum samples overall was low. Like Albert 2011, AZM group was less likely to be colonised with respiratory pathogens and acquisition of macrolide‐resistant bacteria was significantly reduced

Vermeersch 2019

AZM

Sputum samples

Bacterial samples obtained contained H influenzae, S pneumoniae, P aeruginosa, M catarrhalis, and S aureus. At follow‐up, there were no significant group differences (AZM or placebo) for positive sputum cultures with acquired pathogens, neither for acquired macrolide‐resistant bacteria

Macrolide resistance was monitored, but induced sputum was not required per protocol; the limited number of spontaneous sputum samples did not allow for thorough evaluation of antibiotic resistance induced by AZM on top of standard treatment

Abbreviations

AZM: azithromycin; B catarrhalis :Branhamella catarrhalis;CLR: clarithromycin; CFU: colony‐forming unit; DOX: doxycycline; ERY: erythromycin; H influenzae: Haemophius influenzae;MIC: minimum inhibitory concentration; MIC90: MIC required to inhibit growth of 90% or organisms; M catarrhalis: Morexella catarrhalis;MOX: moxifloxacin; NA: not applicable; P aeruginosa: Pseudomonas aeruginosa;PPM: parts per million; qPCR: quantitative polymerase chain reaction; S aureus:Staphylococcus aureus;S pneumoniae: Streptococcus pneumoniae.

Macrolide: azithromycin

Albert 2011 was a 52‐week trial that assessed azithromycin 250 mg once daily compared with placebo (1142 randomised participants). Organisms most commonly identified in the antibiotic group and in the placebo group were Staphylococcus aureus (azithromycin N = 60 (10.7%), placebo N = 71 (12.7%)), Moraxella species (azithromycin N = 13 (2.3%), placebo N = 6 (1%)), and Streptococcus pneumoniae (azithromycin N = 6 (1.1%), placebo N = 6 (1.1%)). S aureus was cultured more frequently, as would be expected from nasopharyngeal sampling. Participants who were not colonised at the start of the study (N = 66 azithromycin, N = 172 placebo) became colonised during the course of the study, and resistance to macrolides was higher in the antibiotic arm than in the placebo arm (81% versus 41%; P < 0.001).

Berkhof 2013 was a 12‐week trial (84 randomised participants) that reported a reduction in respiratory pathogens in sputum samples of those taking azithromycin (250 mg once daily, 3 times a week). At 12 weeks, only one participant in the azithromycin group had antibiotic‐resistant S aureus. Study authors did not report mean inhibitory concentration (MIC)90, which would have allowed detection of changes in antibiotic resistance over time.

Uzun 2014 (92 randomised participants) found that at 52 weeks, fewer people had resistant bacteria when taking azithromycin (500 mg, 3 times a week) compared with placebo (3 versus 11; P = 0.036).

Vermeersch 2019 (301 randomised participants) did not find significant group differences between azithromycin (500 mg once daily for 3 days, followed by 250 mg every 2 days) and placebo for acquired macrolide‐resistant bacteria. At 13 weeks, only one participant in the placebo treatment group had newly acquired macrolide‐resistant bacteria.

Blasi 2010 (22 randomised participants) found that one participant in the azithromycin group (500 mg daily, 3 times a week) had erythromycin‐resistant S pneumoniae at 26 weeks.

Brill 2015 (99 randomised participants) measured sputum bacterial load and antibiotic resistance at 13 weeks post antibiotic treatment (azithromycin 250 mg, 3 times a week) compared with placebo. The most common organisms were S pneumoniae and Streptococcus species. Antibtiotic resistance was increased, with a factor increase of mean inhibitory concentration of 6.23 (95% confidence interval (CI) 1.66 to 23.35; P = 0.01) compared with placebo for sputum‐isolated cultures (Brill 2015).

Macrolide: clarithromycin

Banerjee 2005 (67 randomised participants) found no multi‐resistant gram‐negative pathogens among those taking long‐acting clarithromycin 500 mg daily compared with placebo at 13 weeks.

Macrolide: erythromycin

At 26 weeks of treatment, He 2010 (36 randomised participants) did not report any significant group differences (erythromycin 250 mg, 3 times a day versus placebo) in the emergence of antibiotic‐resistant organisms.

Seemungal 2008 (109 randomised participants) found that at the end of 52 weeks of treatment, only one participant had colonisation of S pneumoniae resistant to erythromycin in the erythromycin (250 mg twice daily) treatment arm. All Haemophilus influenzae isolates (22/109) were found to be resistant to erythromycin.

Quinolone: moxifloxacin

Brill 2015 (99 randomised participants) reported that moxifloxacin (400 mg daily for 5 days, every 4 weeks) was associated with a factor increase in MIC of 4.82 (95% CI 1.44 to 16.19; P = 0.01) compared to placebo from baseline to 13 weeks (Brill 2015). The odds of isolates cultured in sputum being resistant to moxifloxacin was above 2, but this result was not statistically significant.

Sethi 2010 (1149 randomised participants) reported a reduction in the number of participants with pathogens isolated over 48 weeks of treatment, with a more significant reduction with moxifloxacin treatment (400 mg daily for 5 days, repeated every 8 weeks) compared to placebo. One participant taking moxifloxacin had resistant Streptococcus pneumoniae at 40 weeks, which was not associated with exacerbations and did not persist at subsequent visits. This was similarly found in up to three isolates from participants taking moxifloxacin who were positive for S aureus at baseline and also at various time points in the study. The median MIC of moxifloxacin against Pseudomonas aeruginosa at 24 weeks was 4 mg/L, which was reduced further to 1 mg/L during the rest of the treatment period. The opposite was observed in the placebo group, in which median MIC increased from 0.5 mg/L to 2 mg/L among people who completed treatment.

Tetracycline: doxycycline

Brill 2015 (99 randomised participants) reported a change in MIC of 3.74 (95% CI 1.46 to 16.19) with doxycyline (100 mg once daily) compared with placebo at 13 weeks. Moreover, isolates from participants taking doxycycline were more likely to be resistant to doxycycline than from those taking placebo (OR 5.77, 95% CI 1.40 to 23.74; P = 0.02).

Mortality

Overall, eight studies reported mortality data (Albert 2011; Blasi 2010; Mygind 2010; Seemungal 2008; Sethi 2010; Shafuddin 2015; Uzun 2014; Vermeersch 2019) (Table 18). Refer to Appendix 3 for Herath 2018 and Threapleton 2018 results.

Open in table viewer
Table 18. Mortality: numbers of deaths in treatment and placebo groups in included studies

Study ID

Antibiotic class

Antibiotic

Placebo or control or standard treatment

Albert 2011

Macrolide

AZM: 18/570 (3%)

20/572 (4%)

Banerjee 2005

Macrolide

CLR: 0/31

0/36

Berkhof 2013

Macrolide

AZM: 0/42

0/42

Blasi 2010

Macrolide

AZM: 1/11 (9%)

5/11 (45%)

Brill 2015

Quinolone

Tetracycline

Macrolide

MOX: 0/25

DOX: 0/25

AZM: 0/25

0/24

He 2010

Macrolide

ERY: 0/18

0/18

Mygind 2010

Macrolide

AZM: 74/287 (25%)

81/288 (28%)

Seemungal 2008

Macrolide

ERY: 0/53 (0%)

1/56 (2%)

Sethi 2010

Quinolone

MOX: 13/753 (2%)

13/584 (2%)

Shafuddin 2015

Macrolide+

tetracycline

ROX: 3/97 (3%)

DOX+ROX: 5/101 (5%)

5/94 (5%)

Simpson 2014

Macrolide

AZM: 0/15

0/15

Singh 2019

Tetracycline

DOX: 0/30

0/30

Suzuki 2001

Macrolide

ERY: 0/55

0/54

Tan 2016

Macrolide

ERY: 0/18

0/18

Uzun 2014

Macrolide

AZM: 0/47 (0%)

2/45 (4%)

Vermeersch 2019

Macrolide

AZM: 3/147 (2%)

6/154 (3%)

Wang 2017

Macrolide

AZM: 0/43

0/43

Abbreviations

AZM: azithromycin; CLR: clarithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; ROX: roxithromycin.

Overall, the total number of deaths with prophylactic antibiotic was 117/2394 participants compared with 133/2066 participants (Table 18). Death rates reported in the included studies ranged from 2% to 5% with antibiotics (Albert 2011; Sethi 2010; Shafuddin 2015; Vermeersch 2019), and from 2% to 5% with placebo. However, two studies showed higher rates of death compared with other studies (Blasi 2010; Mygind 2010). Mygind 2010, a 156‐week trial, included participants with moderate to severe COPD and reported more than 20% deaths. The number of deaths was similar between treatment groups. Blasi 2010, a 26‐week trial that included participants with severe COPD, showed a significantly higher rate of death in the placebo group compared to the azithromycin group during the treatment period (placebo 45%, azithromycin 9%).

Discussion

Summary of main results

We identified 17 studies that met the inclusion criteria for this review. Twelve studies assessed effects of macrolides (azithromycin and erythromycin) , quinolone (moxifloxacin), tetracycline (doxycycline), or tetracyclines plus macrolides (roxithromycin plus doxycycline) compared with placebo or standard therapy and were included in the network meta‐analysis (NMA) (3405 total participants).

We investigated the safety and clinical effectiveness of prophylactic antibiotics for adults with chronic obstructive pulmonary disease (COPD), specifically focusing on differences among antibiotic classes and individual antibiotics identified. Specifically, we evaluated the impact of different antibiotic classes and individual treatments on the frequency of exacerbations, quality of life evaluated using St George's Respiratory Questionnaire (SGRQ), and serious adverse events (SAEs). Overall, evidence included in the NMA was at low risk of bias and generally supported that macrolides were the highest ranking antibiotic treatment for reducing exacerbations and SAEs, and for improving quality of life, among people with moderate to severe COPD. More specifically, macrolides appear to be superior to placebo in reducing exacerbations and SAEs, with a probability of 0.97 and 0.7, respectively. In addition, macrolides were superior to placebo in improving quality of life, with a probability of 0.82, although the difference from placebo did not exceed the minimum clinically important difference (MCID).

Overall, results of the NMA are consistent with findings of the two previous pair‐wise reviews (Herath 2018; Threapleton 2018).

Exacerbations

Macrolides were more beneficial than placebo in reducing exacerbations (hazard ratio (HR) 0.67, 95% credible interval (CrI) 0.60 to 0.75), with an absolute risk reduction of 127 per 1000 people per year, that is, 127 fewer people had exacerbations per 1000 people treated for a year. The effect between quinolones and placebo was probably smaller (HR 0.89, 95% CrI 0.75 to 1.04). Exacerbations may be increased when tetracyclines are compared to placebo (HR 1.29, 95% CrI 0.66 to 2.41). When the four treatment classes were ranked, macrolides were ranked first, followed byquinolones. Placebo ranked third, although it was only marginally better than tetracyclines. As tetracyclines ranked fourth, this was reflected in the absolute risk, which resulted in an increase in the number of exacerbations by 60 per 1000 people per year compared with placebo. However, there was uncertainty in this result because confidence intervals ranged from 129 fewer to 127 more, and this evidence was based on data from one study. In comparison with each other, macrolides were also superior to quinolones and were borderline superior to tetracyclines in reducing exacerbations.

Quality of life

Treatment with macrolides improved quality of life compared to placebo by 2.29 points on the SGRQ (3.61 to 0.99 point improvement). Smaller improvement was seen with quinolone treatment compared with placebo (1.33 points on the SGRQ (2.99 to 0.33 point improvement)). There may be little to no difference in quality of life with tetracycline treatment compared to placebo (1.18 point worsening in quality of life compared to placebo (1.51 improvement to 3.86 worsening)), although the results were very uncertain. These results are reflected in ranking of these treatments, in which macrolides were ranked first, followed by quinolone. Tetracycline was ranked fourth after placebo.

Serious adverse events

Evidence suggests that macrolides reduced the odds of having an SAE compared to placebo only when the fixed treatment‐fixed class effect model was used. The relative effect of SAEs compared with placebo was odds ratio (OR) 0.76 (95% CrI 0.62 to 0.93). There was probably little to no difference in the effect with quinolones (OR 1.00, 95% CrI 0.72 to 1.34) or with macrolide plus tetracycline (OR 0.97, 95% CrI 0.52 to 1.66). When ranked, macrolides were ranked the highest (i.e. the highest probability of having the largest reduction in SAEs), with a probability of 0.7 of being ranked first. Macrolide plus tetracycline ranked second, quinolone ranked third, and placebo ranked fourth. These results were reflected in the absolute risk of each antibiotic treatment compared with placebo. Absolute risk in the placebo group was 260 per 1000. Absolute risk was 49 fewer per 1000 with macrolide, approximately 10 fewer per 1000 with macrolide plus tetracycline, and approximately 2 fewer per 1000 with quinolone.

Overall completeness and applicability of evidence

To date, this is the first NMA investigating the effectiveness of prophylactic antibiotics for people with COPD. For this reason, we decided to include placebo‐controlled trials to indirectly compare antibiotic classes with each other or with placebo. We identified four antibiotic classes: macrolides, quinolones, tetracyclines, and macrolide plus tetracycline; however most included studies compared macrolides with placebo.

Participants included in the NMA overall were similar, and although there could have been variation in history of exacerbations and maintenance inhaled therapies, there were no concerns for inconsistency across studies. It should be noted, however, that results from the NMA for exacerbations, quality of life, and SAEs can be generalised only to the subgroup of moderate to severe COPD (forced expiratory volume (FEV₁) ranging from 0.935 to 1.36 L) between 64 and 73 years of age.

Analyses show that treatment with macrolides overall reduced exacerbations compared to placebo or standard treatment. We assumed that with placebo or standard treatment, individuals would be expected to experience two exacerbations per year. Based on this assumption, the absolute risk was considerably lower with macrolides compared with placebo (127 per 1000 compared with 864 per 1000, respectively) (summary of findings Table 1). With quinolones and tetracyclines, there was uncertainty about the difference in exacerbations, as credible intervals crossed the line of no effect. In current clinical practice, long‐term macrolide treatment could be considered for people with COPD who have more than three acute exacerbations per year, and an accurate assessment of baseline exacerbation rate should be determined before long‐term antibiotics are started (Smith 2020). National Institute for Health and Care Excellence (NICE) guidance suggests that the macrolide azithromycin (250 mg 3 times a week) could be used on the condition that individuals who continue to have frequent exacerbations or prolonged exacerbations, or exacerbations requiring hospitalisation, are not smoking, and that such treatment should be continued only as required, and if benefits outweigh risks (NICE 2018).

Similarly, the SGRQ analysis shows that treatment with macrolides resulted in greater improvement in quality of life compared to placebo (mean difference (MD) 2.298), as did treatment with quinolones (MD 1.33). Although these results did not reach the MCID of 4 points, there was modest benefit for quality of life. This result is in line with findings of a previous review showing modest improvement on the SGRQ but not reaching clinical significance with macrolides (i.e. a decrease of 2.12 points) (Ni 2015). Overall, from these findings, it is unclear how much long‐term antibiotics may impact individuals' quality of life beyond 12 months, as the duration of treatment in our analysis ranged from 12 weeks to 52 weeks. Only three studies lasted longer than 48 to 52 weeks (Albert 2011; Sethi 2010; Uzun 2014). Therefore, it may be important to take into consideration the risk and benefit for each individual with regular monitoring at 6 and 12 months using the SGRQ tool, as suggested in clinical guidance (Smith 2020). We did not investigate responder analysis data for SGRQ in trials or other systematic reviews, which could have been informative; however, this could be revisited in the future.

When SAEs were assessed, it was noted that fewer people experienced adverse events (AEs) with macrolides (49 fewer per 1000 people with macrolides versus placebo treatment (260 per 1000)). Ni 2015 reported a slightly higher rate of AEs in the macrolide group compared to the placebo group (OR 1.55, 95% confidence interval (CI) 1.003 to 2.39; P = 0.049). There was no difference in the effects of taking quinolones or macrolide plus tetracycline compared with placebo. It is important to note that the absolute risk of events in the placebo arm was calculated from the included studies and may not be applicable in general. Furthermore, we did not investigate the association between duration of antibiotic treatment and the impact of SAEs that people may experience. One recent systematic review found that macrolide use for 3 or 12 months resulted in more side effects than control treatment, but at 6 months, there was no difference between antibiotics and control treatment (Cui 2018). We did not investigate the impact of antibiotics on individual side effects; however, a previous Cochrane Review assessed this in detail (Herath 2018). Among the studies that we included in this review, hearing impairment and gastrointestinal problems were more commonly associated with long‐term use of macrolides. Exacerbations could have been reported as AEs or SAEs; however, exacerbations were reported separately from AEs among the studies included in the analysis (Table 19).

Open in table viewer
Table 19. Adverse events across all studies

Study ID, drug, dose, schedule (weeks' duration)

Adverse events

Antibiotic (n)

Comparator (n)

Reporting of exacerbations as AEs

Albert 2011, AZM, 250 mg once daily (52)

Discontinuation due to: audiogram‐confirmed hearing decrement (142), tinnitus (4), gastrointestinal tract (11), QTc prolongation (6), allergic reaction (5), abnormal laboratory test (4), other (10)

Discontinuation due to: audiogram‐confirmed hearing decrement, tinnitus (4), neoplasm (3), GI tract (6), QTc prolongation (4), allergic reaction (8), abnormal laboratory test (3), other (17)

Exacerbation was not reported as an AE

He 2010, ERY, 125 mg 3 times daily (26)

Discontinued due to: abdominal pain (1), complication of left heart failure (1)

Discontinued due to: respiratory insufficiency (2), other (1)

Exacerbation was not reported as an AE

Seemungal 2008 ERY, 250 mg twice daily (52)

Upper gastrointestinal (5), lower gastrointestinal (3), rash (3), other (3)

Upper gastrointestinal (5), lower gastrointestinal (3), rash (2), other (2)

Exacerbation was not reported as an AE

Sethi 2010, MOX, 400 mg once daily (5 days every 8 weeks) (48)

Cardiac disorders (3), gastrointestinal (diarrhoea, nausea, vomiting) (27), general disorders/administration site conditions (4), asthenia (3), immune system disorders (4), hypersensitivity (3), infections and infestations (5), musculoskeletal and connective tissue disorders (3), nervous system disorders (6), dizziness (3), respiratory, thoracic and mediastinal disorders (8), dyspnoea (4), skin and subcutaneous tissue disorders (5), AEs leading to discontinuation (26)

Cardiac disorders (1), gastrointestinal (diarrhoea, nausea, vomiting) (4), general disorders and administration site conditions (2), asthenia (0), hypersensitivity (0), infections and infestations (3), musculoskeletal and connective tissue disorders (1), nervous system disorders (4), dizziness (1), respiratory, thoracic, and mediastinal disorders (0), dyspnoea (0), skin and subcutaneous tissue disorders (5), AEs leading to discontinuation (16)

Exacerbation was not reported as an AE

Shafuddin 2015, ROX 300 mg once daily + DOX 100 mg once daily; or ROX 300 mg once daily (12)

Roxithromycin + doxycycline: nausea (12), diarrhoea (2), headache (4), abdominal pain (3), reflux (2), vomiting (1), abnormal liver function (1), abnormal ECG (1), rash (1), dyspnoea (0), dizziness (0), oral candidiasis (0), gastrointestinal upset (0)

Roxithromycin alone: nausea (13), diarrhoea (3), headache (1), abdominal pain (1), reflux (1), vomiting (3), abnormal liver function (2), abnormal ECG (0), rash (1), dyspnoea (1), dizziness (4), oral candidiasis (2), gastrointestinal upset (2)

Nausea (1), diarrhoea (1), headache (1), abdominal pain (1), reflux (0), vomiting (0), abnormal liver function (0), abnormal ECG (0), dyspnoea (2), dizziness (0), oral candidiasis (3), gastrointestinal upset (2)

Exacerbation was not reported as an AE

Simpson 2014, AZM, 250 mg once daily (12)

Diarrhoea (1), abdominal pain (0), nausea (0), vomiting (0), fever (0), headache (0), rash (0), hearing loss (0), other (10)

Diarrhoea (5), abdominal pain (0), nausea (0), vomiting (0), fever (0), headache (0), rash (0), hearing loss (0), other (9)

Exacerbation was not reported as an AE

Tan 2016, ERY, 125 mg 3 times daily (52)

Withdrawal due to: abdominal pain (1), left‐sided heart failure (1)

Withdrawal due to: respiratory insufficiency (2), unknown (1)

Exacerbation was not an outcome in the study

Uzun 2014, AZM, 500 mg once daily 3 times per week (52)

Diarrhoea (9), nausea/vomiting (3), other (4)

Diarrhoea (1), nausea/vomiting (2), other (7)

Exacerbaiton was not reported as an AE. 2 fatal SAEs occurred due to COPD respiratory failure, which were counted in the mortality outcome

Vermeersch 2019, AZM, 500 mg once daily (for first 3 days), AZM 250 mg every 2 days for rest of treatment duration (13)

Diarrhoea (20), nausea (12), anorexia (9), hearing loss (1), syncope (1)

Diarrhoea (15), nausea (11), anorexia (8), hearing loss (6), syncope (2)

Exacerbation was not reported as an AE

Abbreviations

AE: adverse event;AZM: azithromycin; DOX: doxycyline; ECG: electrocardiogram; ERY: erythromycin; MOX: moxifloxacin; NR: not reported; QTc: corrected QT interval; ROX: roxithromycin; SAE: serious adverse event.

Most data on microbial resistance that we identified assessed macrolides (azithromycin) with varying durations of treatment. Macrolide use led to microbial resistance in Albert 2011 at 52 weeks; however, other studies were of shorter duration, and they found no difference between antibiotic and placebo treatments (Vermeersch 2019), or they reported microbial resistance in only one participant taking macrolide (Blasi 2010). With quinolone treatment, fewer individuals had pathogens compared with placebo (Sethi 2010), and although some quinolone‐resistant bacteria were isolated, they did not persist at the end of treatment. Based on limited evidence, concern is ongoing that the association of antibiotic treatment and microbial resistance may increase over time, prompting careful clinical assessment of risk and benefit before such treatment is started, and regular monitoring once antibiotic treatment is under way.

Limited data are available regarding the persistence of microbial resistance after discontinuation of prophylactic antibiotics for patients with COPD. This question was not rigorously assessed in any of the included studies. Indirect data are available from the AZISAST trial, which evaluated microbial resistance after discontinuation of prophylactic macrolides in exacerbation‐prone patients with severe asthma (Brusselle 2013). In the AZISAST trial, the percentage of macrolide‐resistant streptococci was reduced from 74% to 46% within four weeks from macrolide discontinuation. In the same period, microbiome characteristics returned to the pretreatment condition. As a result of this observation, annual brief periods of discontinuation of prophylactic antibiotics, preferably during summer, when disease is better controlled, are often suggested for patients with COPD (Miravittles 2015; Smith 2020). Such strategies will need to be evaluated by well‐designed randomised controlled trials (RCTs).

Overall, we found that death was a rare event in clinical studies. In most studies, no deaths occurred in either treatment group. When deaths were reported, they were similar in antibiotic and treatment groups. Mygind 2010 recorded a higher death rate (although similar in each treatment group) (azithromycin 25%, placebo 28%), but the study provided no further information. The death rate in Blasi 2010 was higher in the placebo group compared with the azithromycin group (azithromycin 9%, placebo 45%).

In summary, findings of this NMA show that macrolides are superior to other prophylactic antibiotics identified in the review. The superiority of macrolides compared to other antibiotics in preventing exacerbations has long been suspected and may be attributed to the anti‐inflammatory and immunomodulatory effects that this class of antibiotics exerts, in contrast to the other evaluated classes (Loukides 2013). In line with current clinical guidance, azithromycin at the doses identified may be of benefit for reducing exacerbations and improving quality of life in a subgroup of patients who have moderate to severe COPD that is managed by inhaled therapies but who encounter frequent exacerbations that may result in hospitalisation.

Current clinical guidance recommends that prophylactic antibiotic treatment should be considered for a minimum of 6 to 12 months while changes in exacerbations are observed (Smith 2020). The duration of antibiotic use among studies in the NMA ranged from 12 weeks to 52 weeks, which could be considered long enough to detect differences in exacerbations, but also in quality of life and SAEs. Studies in the NMA represent a wide range of populations geographically (China, Europe, United Kingdom, and USA); however, generalisation of prophylactic antibiotics may be impaired by our inability to include two studies from India and Japan in the main analyses (Singh 2019; Suzuki 2001).

We did not investigate the cost‐effectiveness of one antibiotic compared to another; however, these treatments are generally considered to be of low cost, and exacerbations pose a significant health and economic burden. It should be noted that although such treatments may be cost‐effective, potential costs of monitoring and follow‐up of those taking antibiotics over a long time would need to be considered but are likely to be off‐set by potential benefits for health status.

However, significant gaps in the evidence need to be addressed. First, we did not identify any data on the impact of long‐term antibiotic administration in different types of exacerbations (i.e. exacerbations triggered by bacteria or viruses, or those associated with enhanced eosinophilic inflammation) (Mathioudakis 2020). Although it is clear that prophylactic antibiotics are effective only for a subgroup of selected patients with COPD, this finding has not clearly emerged from available evidence in this or previous Cochrane Reviews (Herath 2018, Threapleton 2018). In addition, we did not identify data on the effectiveness of other antibiotics that may be used to treat exacerbations. For example, penicillins are often used to treat acute exacerbations, but we did not identify any trials in which penicillins are used for prophylaxis. Limited information is available about tetracyclines and combined antibiotic treatment; however, available evidence indicates that these antibiotic combinations are not necessarily superior (Shafuddin 2015).

Quality of the evidence

The methodological quality of all included studies was assessed and overall risk of bias was deemed low across the five domains. Some studies lack clarity regarding randomisation, allocation concealment, and attrition. We did not contact study authors for further information for unclear domains, as these studies were also included in Herath 2018 and Threapleton 2018, and study authors would have been contacted already. We were more confident in our findings from the main analyses when two studies at high risk of bias were excluded (Singh 2019; Suzuki 2001). These studies were instead included in sensitivity analyses that did not significantly change our findings. In all networks, all loops were formed by a single multi‐arm study; therefore, there was no potential to detect inconsistency, and inconsistency checks were not carried out. For most networks, the fixed effect model was selected, as there was no evidence of heterogeneity. When random effects models were used, the standard deviation for between‐study heterogeneity was reported along with its credible interval (exacerbations: Table 5; SAEs: Table 12). Imprecision was reflected in the 95% CrI, which was reported and commented on when appropriate.

GRADE headings were not used to assess validity of the evidence, as we did not think this approach would be informative because more than two treatments were being compared in the NMA. Similarly, we did not consider use of Confidence in Network Meta‐Analysis (CINeMA) because this application cannot be used when a Bayesian analysis is conducted. We used threshold analysis to examine the impact of bias on each treatment contrast (relative effect of each treatment comparison) (Phillippo 2019), which quantified how much the evidence could change before the best treatment changed, and what the new 'best' treatment would be. This approach indicates which results were robust to potential biases in the evidence, taking into account the contribution of each study to the overall results and network structure. We interpreted threshold results with respect to the risk of bias identified for each study, and this is reflected in the conclusions. For exacerbations, the comparison of macrolides to control could change due to plausible bias adjustment, suggesting the possibility that quinolones might be the best class of treatment agents for preventing exacerbations. All other antibiotic versus placebo comparisons were robust to any changes in the evidence, as no implausibly large bias was present that would favour placebo. Consequently, direct comparison of macrolides with quinolones in future RCTs could be informative.

Potential biases in the review process

We followed our pre‐published protocol when carrying out this network meta‐analysis. We included reviews that intended to compare placebo‐controlled trials and head‐to‐head antibiotic comparisons. We checked our included studies with two previously published Cochrane Reviews (Herath 2018; Threapleton 2018). As most of the studies that we identified were also included in these reviews, we were confident that we had identified all relevant studies for the NMA. These studies had already been assessed for risk of bias, and data had been extracted by two review authors; however, we arranged the data in the format required for NMAs. Updated searches identified two new studies that were published in 2019, which met the inclusion criteria for the NMAs. We did not use GRADE nor CINeMA to assess certainty of evidence, as we used the threshold analysis as an alternative approach.

Agreements and disagreements with other studies or reviews

To date, no published network meta‐analyses have specifically investigated prophylactic antibiotics for people with COPD. However, several published systematic reviews and meta‐analyses have investigated the effectiveness of prophylactic antibiotics for a subgroup of people with moderate to severe COPD (Cui 2018; Donath 2013; Herath 2018; Huckle 2018; Lee 2013; Ni 2015; Wang 2018; Yao 2013).

Pair‐wise comparisons of prophylactic antibiotics with placebo or with each other were previously published in two Cochrane Reviews (Herath 2018; Threapleton 2018). Herath 2018 compared prophylactic antibiotics with placebo and included 14 trials (3932 participants). These review authors found that prophylactic antibiotics, specifically macrolides (continuous or intermittent), were beneficial in reducing exacerbations among COPD patients. There was probable benefit for patient‐reported quality of life with antibiotics compared with placebo, but this did not reduce the number of deaths due to any cause nor frequency of hospitalisation nor lung function loss during the study period (Herath 2018). Threapleton 2018 compared different classes of prophylactic antibiotics with each other and identified only two trials (391 participants) of short duration in which antibiotics were compared head‐to‐head in COPD patients. There was no clear difference between one antibiotic and another in reducing exacerbations or quality of life. No deaths were reported in one study, but eight people died in the other study. Very similar numbers of people in both studies experienced serious side effects. These numbers were small, and overall it is unclear whether one antibiotic treatment type caused more side effects than another (Threapleton 2018).

Herath 2018 concluded that prophylactic macrolide antibiotics used up to 12 months are likely to reduce the number of people who experience one or more exacerbations (exacerbation frequency) and to increase the median time to first exacerbation, while improving health‐related quality of life. As the head‐to‐head comparison of antibiotics was not clear due to insufficient information, no specific inferences were made in the review (Threapleton 2018). Evidence for exacerbations in Herath 2018 was of moderate certainty and shows overall benefit of antibiotics in reducing exacerbations using GRADE. This review did not conduct subgroup analysis according to treatment class, but when antibiotics were subgrouped according to antibiotic regimen, macrolides seemed to be most effective in reducing the number of people who had exacerbations (Herath 2018).

In our search of relevant evidence, we have identified the same published trials that were included in published reviews, with the exception of some new trials published in 2019. By using these placebo‐controlled trials, we have been able to indirectly compare different antibiotic classes with each other and with placebo treatment. Previous published evidence shows that macrolides can be of benefit for reducing exacerbations in people with moderate to severe COPD. In addition, British Thoracic Society (BTS) and NICE guidance recommends macrolide use for the moderate to severe COPD population subgroup who have frequent exacerbations requiring steroid therapy, who do not currently smoke, and who have had at least one exacerbation requiring hospitalisation per year (NICE 2016; Smith 2020). BTS guidance suggests that long‐term macrolide treatment could be considered for a minimum of 6 months up to 12 months until its impact on exacerbations is assessed (Smith 2020). Our NMA results show that macrolides rank higher than quinolones, tetracyclines, and placebo in reducing exacerbations, even though the duration of treatment across studies varies from 12 to 52 weeks.

Previous published evidence suggests that macrolides can improve quality of life in people with moderate to severe COPD who are taking macrolides (Cui 2018; Herath 2018; Ni 2015; Wang 2018); however, this improvement was not shown to reach the MCID of 4 points. Similarly, results from the NMA show some improvement in quality of life with macrolide treatment, as measured by SGRQ, and in line with published evidence, this improvement does not reach the MCID.

Published reviews have reported increased risk of side effects associated with longer antibiotic treatment duration (Lee 2013; Ni 2015; Wang 2018; Yao 2013). Results from our NMA indicate that with macrolide treatment, the number of people experiencing one or more SAE is reduced compared to placebo. However, we did not investigate all adverse events or side effects, which may be increased overall by antibiotic treatment. Adverse effects most commonly associated with macrolide treatment were hearing impairment, gastrointestinal problems, and nausea, as well as others (Table 19). As suggested by BTS guidance, the risk‐to‐benefit balance needs to be considered by clinicians when they administer macrolides (Smith 2020).

Study flow diagram.

Figures and Tables -
Figure 1

Study flow diagram.

Navigate to figure in ReviewOpen in new tab

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

Figures and Tables -
Figure 2

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

Navigate to figure in ReviewOpen in new tab

Exacerbations: network diagram of interventions and classes. Treatment abbreviations are defined in . The size of the nodes is proportionate to the number of participants assigned to the intervention. The thickness of the lines is proportionate to the number of randomised trials that studied the respective comparison.

Figures and Tables -
Figure 3

Exacerbations: network diagram of interventions and classes. Treatment abbreviations are defined in Table 1. The size of the nodes is proportionate to the number of participants assigned to the intervention. The thickness of the lines is proportionate to the number of randomised trials that studied the respective comparison.

Navigate to figure in ReviewOpen in new tab

Exacerbations: forest plot of relative effects for each class comparison. Values less than 1 favour the first names class.

Figures and Tables -
Figure 4

Exacerbations: forest plot of relative effects for each class comparison. Values less than 1 favour the first names class.

Navigate to figure in ReviewOpen in new tab

Exacerbations: plot of rank probabilities for each class.

Figures and Tables -
Figure 5

Exacerbations: plot of rank probabilities for each class.

Navigate to figure in ReviewOpen in new tab

Exacerbations: forest plot with threshold analysis for the log‐HR of exacerbations for each class. Base case optimal treatment set is 2. Class codes: 1 = placebo; 2 = macrolide; 3 = tetracycline; 4 = quinolone. Comparisons are macrolide versus placebo (2 versus 1); tetracycline versus placebo (3 versus 1); quinolone versus placebo (4 versus 1); tetracycline versus macrolide (3 versus 2); quinolone versus macrolide (4 versus 2); quinolone versus tetracycline (4 versus 3).

Figures and Tables -
Figure 6

Exacerbations: forest plot with threshold analysis for the log‐HR of exacerbations for each class. Base case optimal treatment set is 2. Class codes: 1 = placebo; 2 = macrolide; 3 = tetracycline; 4 = quinolone. Comparisons are macrolide versus placebo (2 versus 1); tetracycline versus placebo (3 versus 1); quinolone versus placebo (4 versus 1); tetracycline versus macrolide (3 versus 2); quinolone versus macrolide (4 versus 2); quinolone versus tetracycline (4 versus 3).

Navigate to figure in ReviewOpen in new tab

Quality of life: SGRQ network map.

Figures and Tables -
Figure 7

Quality of life: SGRQ network map.

Navigate to figure in ReviewOpen in new tab

Change from baseline in SGRQ: forest plot of relative effects for each class comparison. Values less than 0 favour the first named class.

Figures and Tables -
Figure 8

Change from baseline in SGRQ: forest plot of relative effects for each class comparison. Values less than 0 favour the first named class.

Navigate to figure in ReviewOpen in new tab

Change from baseline in SGRQ: plot of rank probabilities for each class.

Figures and Tables -
Figure 9

Change from baseline in SGRQ: plot of rank probabilities for each class.

Navigate to figure in ReviewOpen in new tab

Serious adverse events: forest plot of relative effects for each class comparison. Values less than 1 favour the first named class.

Figures and Tables -
Figure 10

Serious adverse events: forest plot of relative effects for each class comparison. Values less than 1 favour the first named class.

Navigate to figure in ReviewOpen in new tab

Serious adverse events: network map.

Figures and Tables -
Figure 11

Serious adverse events: network map.

Navigate to figure in ReviewOpen in new tab

Serious adverse events: plot of rank probabilities for each antibiotic class

Figures and Tables -
Figure 12

Serious adverse events: plot of rank probabilities for each antibiotic class

Navigate to figure in ReviewOpen in new tab
Summary of findings 1. Summary of findings: exacerbations

Prophylactic antibiotics compared with placebo for COPD

Patients or population: adults with COPD

Settings: hospital clinics, multi‐centre

Intervention: macrolide, tetracycline, or quinolone

Comparison: placebo or standard care

Treatment

Anticipated absolute effects (95% CrI)*

Relative effect
HR (95% CrI)

No. of participants
(studies)

Absolute rate of exacerbations: median (95% CrI)

Risk difference with treatment

(number of people experiencing exacerbations)

Macrolide

(weighted mean 50 weeks' duration)

1.34 (1.19 to 1.50)

127 fewer per 1000 (168 fewer to 87 fewer)

0.67 (0.60 to 0.75)

688 (6)

Tetracycline

(13 weeks' duration)

2.58 (1.33 to 4.81)

60 more per 1000 (129 fewer to 127 more)

1.29 (0.66 to 2.41)

25 (1)

Quinolone

(weighted mean 46.5 weeks' duration)

1.77 (1.50 to 2.08)

35 fewer per 1000 (87 fewer to 11 more)

0.89 (0.75 to 1.04)

594 (2)

*The basis for the anticipatedrisk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% CrI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CrI).
COPD: chronic obstructive pulmonary disease; CrI: credible interval; HR: hazard ratio.

*Absolute rate of exacerbations per year in the placebo arm = 2; 864 people per 1000 experienced exacerbations over a year.

Figures and Tables -
Summary of findings 1. Summary of findings: exacerbations
Navigate to table in Review
Summary of findings 2. Summary of findings: change from baseline in SGRQ

Prophylactic antibiotics compared with placebo for COPD

Patients or population: adults with COPD

Settings: hospital clinics, multi‐centre

Intervention: macrolide, tetracycline, or quinolone

Comparison: placebo

Treatment

Anticipated absolute effects (95% CrI)*

No. of participants
(studies)

Absolute change from baseline in SGRQ (95% CrI)

Mean difference in change from baseline in SGRQ score with treatment**

Macrolide

(weighted mean 48 weeks' duration)

‐4.00 (‐5.51 to ‐2.68)

2.298 point improvement (3.605 to 0.985 point improvement)

578 (6)

Tetracycline

(13 weeks' duration)

‐0.52 (‐3.21 to 2.16)

1.179 point worsening (1.509 point improvement to 3.859 point worsening)

25 (1)

Quinolone

(weighted mean 46.5 weeks' duration)

‐3.03 (‐4.69 to ‐1.37)

1.33 point improvement (2.986 point improvement to 0.328 point improvement)

528 (2)

*The basis for the anticipatedrisk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% CrI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CrI).
COPD: chronic obstructive pulmonary disease; CrI: credible interval; SGRQ: St George's Respiratory Questionnaire.

*Absolute change from baseline in the placebo arm was ‐1.7 (1.7 point improvement).

**The minimally clinically important difference for SGRQ is 4 points.
Figures and Tables -
Summary of findings 2. Summary of findings: change from baseline in SGRQ
Navigate to table in Review
Summary of findings 3. Summary of findings: serious adverse events

Prophylactic antibiotics compared with placebo for COPD

Patients or population: adults with COPD

Settings: hospital clinics, multi‐centre

Intervention: macrolide, tetracycline, or quinolone

Comparison: placebo

Treatment

Anticipated absolute effects (95% CrI)*

Relative effect
OR (95% CrI)

No. of participants
(studies)

Absolute probability of an SAE: median (95% CrI)

Risk difference with treatment*

Macrolide

(weighted mean 49 weeks' duration)

0.21 (0.18 to 0.25)

49.07 fewer per 1000 (81.18 fewer to 14.23 fewer)

0.76 (0.62 to 0.93)

971 (8)

Quinolone

(48 weeks' duration)

0.26 (0.20 to 0.32)

1.873 fewer per 1000 (57.88 fewer to 60.89 more)

1.00 (0.72 to 1.34)

569 (1)

Macrolide + tetracycline

(12 weeks' duration)

0.25 (0.15 to 0.37)

9.461 fewer per 1000 (1.07 fewer to 108.5 more)

0.97 (0.52 to 1.66)

101 (1)

*The basis for the anticipatedrisk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% CrI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CrI).
COPD: chronic obstructive pulmonary disease; CrI: credible interval; OR: odds ratio.

*Absolute probability of events in the placebo arm was 0.26; risk of an SAE with placebo was 260 per 1000.

Figures and Tables -
Summary of findings 3. Summary of findings: serious adverse events
Navigate to table in Review
Table 1. Characteristics of studies including prior exacerbation details

Author

Class comparison

Concomitant treatments (%, antibiotic/ placebo)

Dose/regimen

COPD severity

Included in NMA?

Exacberations in the previous 12 months before participation in study

Exacerbation definition

Risk of bias

Albert 2011

(N = 1142);

USA

(12 academic health centres)

52 weeks

Macrolide vs placebo

ICS only (4%/6%)

LAMAs only
(6%/8%)

LABAs only
(3%/1%)

ICS + LAMA (19%/22%)
ICS + LABA (4%/5%)

LABA + LAMA (5%/4%)
ICS + LABA + LAMA (49%/46%)

None (10%/8%)

AZM 250 mg daily

continuous

Moderate to severe

FEV₁ 1.11 L

Yes

Approximately 50% of participants in each treatment arm had required hospitalisation or an ED visit in the previous 12 months

Acute exacerbation of COPD: “a complex of respiratory symptoms (increased or new onset) of more than one of the following: cough, sputum, wheezing, dyspnoea, or chest tightness with a duration of at least 3 days requiring treatment with antibiotics or systemic steroids"

Low risk of bias across all domains except attrition (unclear reasoning of missing data for HRQoL)

Banerjee 2005
(N = 67); UK (clinics and lung function units from 2 hospitals)

13 weeks

Macrolide vs placebo

All participants:

ICS (100%),
LABAs (18%), inhaled anticholinergics (63%)

CLR 500 mg daily

continuous

Moderate to severe

No

NR

Not included in exacerbation analysis

Low risk of bias across all domains except detection bias, which was unclear

Berkhof 2013

(N = 84);

Netherlands

(1 teaching hospital)

12 weeks

Macrolide vs placebo

LABAs (81%/80%)

Long‐acting anticholinergics (64%/57%)

ICS (98%/83%)

AZM 250 mg 3 times a week

Intermittent

Moderate

FEV₁ 1.36 L

Yes

Participants had a median for 1 exacerbation (range 0 to 13) in the previous 12 months

Time to first exacerbation of COPD: sustained worsening of COPD, from stable state and beyond normal day‐to‐day variations, requiring treatment with prednisolone, antibiotics, or a combination of both

Unclear selection bias (allocation) but assumed done, low risk across all other domains

Blasi 2010

(N = 22);

Italy

(multi‐centre)

26 weeks

Macrolide vs placebo

Inhaled medication NR LTOT (46% in both groups)

AZM 500 mg 3 times a week

Intermittent

Severe

FEV₁ (not reported)

Yes

Participants in each treatment arm had a mean of 3 exacerbations in the previous 12 months

Worsening of symptoms requiring both a change in regular respiratory medication or medical assistance, or resulting in hospitalisation or ED treatment

Judged as high risk of bias for allocation concealment, performance, detection, attrition, and selective reporting; open‐label

Brill 2015 (N = 99);

UK

(1 outpatient

hospital department)

13 weeks

Quinolone

Tetracycline

Macrolide

vs placebo

ICS (84%/76%/72%)

ICS in placebo group 57%

MOX 400 mg daily for 5 days every 4 weeks (pulsed)

DOX 100 mg daily (continuous)

AZM 250 mg 3 times a week (intermittent)

Moderate to severe

FEV₁ 1.4 L

Yes

Participants had a mean of 2.5 (SD 2.1) exacerbations with moxifloxacin, 2.1 (SD 1.7) exacerbations with doxycycline, 2.8 (SD 4.0) exacerbations with azithromycin, and

1.5 (SD 1.4) exacerbations with placebo in previous 12 months

Exacerbations during the study: using diary card criteria, patient reporting to clinical health professionals or study team.

Exacerbation was not the primary outcome of the study

Unclear performance bias; detection bias judged as high

He 2010

(N = 36);

China

(1 university hospital)

26 weeks

Macrolide vs placebo

ICS (44%/38%)

Theophylline (61%/55%)

Inhaled anticholinergic (50%/55%)

Inhaled beta‐adrenergic (72%/77%)

ERY 125 mg 3 times daily (continuous)

Severe

FEV₁ 1.07 L at baseline

Yes

NR

Moderate exacerbation: sustained worsening of baseline respiratory symptoms for at least 2 days requiring increased treatment or additional therapy (e.g. OCS, antibiotics)

Severe exacerbation: all of the above plus requiring hospital admission

Randomisation and allocation unclear. Double‐blind study, but outcome assessment unclear. Funding not stated

Mygind 2010

(N = 575);

Denmark

156 weeks

Macrolide vs placebo

NR

AZM 500 mg for 3 days every month (pulsed)

NR

No

NR

Not included in exacerbation analysis

Unclear randomisation, allocation concealment, attrition domains. Blinding of participants, personnel, and outcome assessors were judged as low risk of bias

Seemungal 2008

(N = 109);

UK

(2 outpatient clinics in 2 hospitals)

52 weeks

Macrolide vs placebo

ICS (77% in both groups)

LABAs (66%/61%)

LAMAs (28%/38%)

Theophylline (7.5%/14%)

ERY 250 mg twice daily (continuous)

Moderate to severe

FEV₁ 1.31 L at baseline

Yes

35% of participants had 3 or more exacerbations in the previous 12 months

Moderate exacerbation: sustained worsening of baseline respiratory symptoms for at least 2 days requiring treatment with OCS (prednisolone) and/or antibiotics

Severe exacerbation: requiring admission to hospital

Low risk of bias across all domains. Funded by British Lung Foundation

Sethi 2010

(N = 1157);

(international

multi‐centre)

48 weeks

Quinolone

vs placebo

SABAs (71%/72%)

LABAs (44%/45%)

ICS (41%/43%)

Theophylline (29%/26%)

Systemic steroids (0.4%/0.2%)

Others (4.7%/5.7%)

ICS + long‐acting bronchodilators (25%/26%)

MOX 400 mg daily for 5 days every 8 weeks (pulsed)

Mild to severe

FEV₁ 1.2 L at baseline

Yes

NR

Any confirmed AECB: requiring intervention

(start of systemic antibiotic and/or start of systemic steroid and/or hospitalisation within 7 days of the start date of exacerbation) and with a minimum of 2 weeks between the start of 2 consecutive exacerbations

Unclear risk for selection bias (random sequence generation and allocation concealment). Low risk for performance bias and selective reporting

Shafuddin 2015

(N = 292); Australia and New Zealand (multi‐ centre)

12 weeks

Macrolide

Macrolide plus tetracycline

vs placebo

NR

ROX 300 mg daily (continuous)

DOX + ROX 100 mg daily plus 300 mg daily (continuous)

Moderate to severe

FEV₁ 0.935 L at baseline

Yes

Mean 5.11 (SD 2.4) exacerbations within 2 years

Not included in exacerbation analysis

Low risk of bias across all domain except attrition, which was unclear

Simpson 2014

(N = 30);

Australia

(1 tertiary care respiratory and sleep ambulatory care service, hospital)

12 weeks

Macrolide vs placebo

ICS (% NR)

AZM 250 mg daily (continuous)

Moderate

Yes

NR

Severe exacerbations of COPD: requiring unscheduled medical attention with treatment of OCS and/or antibiotics

Low risk of bias across all domains

Singh 2019 (N = 60); India

(1 outpatient department)

13 weeks

Tetracycline

vs placebo

NR

DOX 100 mg daily (continuous)

Moderate to severe

No (sensitivity analysis)

NR

Not included in exacerbation analysis

Low risk of bias for allocation concealment, high risk of bias for blinding of participants, personnel, and outcome assessors. Randomisation and selective reporting domains were unclear

Suzuki 2001 (N = 109);

Japan (setting NR)

13 weeks

Macrolide

vs placebo

NR

ERY 200 to 400 mg daily (continuous)

FEV₁ 1.47 L at baseline

No (sensitivity analysis)

NR

Not included in exacerbation analysis

Low risk of bias across most domains except for blinding of participants, personnel, and outcome assessors, which were judged as high risk of bias

Tan 2016

(N = 49);

China

(1 regional hospital)

52 weeks

Macrolide

vs placebo

ICS (44%/38%/44%)

Theophylline (55%/55%/61%)

Inhaled anticholinergic (55%/50%/50%)

Inhaled beta2‐adrenergic agonist (66%/66%/72%)

ERY 125 mg 3 times daily (continuous)

ERY 125 mg 3 times daily with 6 months' follow‐up (continuous)

Moderate to severe

FEV₁ 1.04 to 1.08 L

Yes

NR

Not included in exacerbation analysis

Unclear risk of bias across most domains, high risk of bias for blinding of participants, personnel, and outcome assessors

Uzun 2014

(N = 92);

Netherlands

(1 regional hospital)

52 weeks

Macrolide

vs placebo

LABA (96%/91%)

LAMA (89%/71%)

ICS (89%/96%)

SABA (68%/73%)

Prednisolone (28%/20%)

AZM 500 mg 3 times a week (intermittent)

Mild to severe

FEV₁ 1.1 L at baseline

Yes

Participants had a mean of 4 (SD 1.1) acute exacerbations in the previous 12 months

All exacerbations: defined according to Anthonisen criteria, and whether the patient needed treatment with steroids or antibiotics, or both.

Severe exacerbation: requiring hospital admission.

Mild exacerbation: requiring treatment at the outpatient department

Low risk of bias across all domains

Vermeersch 2019

(N = 301);

Italy

(5 centres across Italy)

13 weeks

Macrolide

vs placebo

LABA (93%/94%)

LAMA (80%/80%)

ICS (80%/80%)

SABA (73%/71%)

AZM 500 mg once daily (loading dose) for 3 days, followed by 250 mg every 2 days for 13 weeks (intermittent)

FEV₁ 0.925 L

Yes

NR

Not included in exacerbation analysis

Low risk of bias across all domains

Wang 2017

(N = 86);

China

(1 regional hospital)

26 weeks

Macrolide

vs placebo

NR

AZM 250 mg once daily plus 20 mg once daily simvastatin (continuous)

FEV₁ 0.67 L

No

NR

Not included in exacerbation analysis

Low risk of bias for randomisation, high risk of bias for blinding of participants, personnel, and outcome assessors

Abbreviations

AECB: acute exacerbation of chronic bronchitis; AZM: azithromycin; CLR: clarithromycin; COPD: chronic obstructive pulmonary disease; DOX: doxycycline; ED: emergency department; ERY: erythromycin; FEV₁: forced expiratory volume in one second; HRQoL: health‐related quality of life; ICS: inhaled corticosteroid; LABA: long‐acting beta agonist; LAMA: long‐acting muscarinic antagonist; LTOT: long‐term oxygen therapy; MOX: moxifloxacin; NMA: network meta‐analysis; NR: not reported; OCS: oral corticosteroids;ROX: roxithromycin; SABA: short‐acting beta agonist; SD: standard deviation.

Figures and Tables -
Table 1. Characteristics of studies including prior exacerbation details
Navigate to table in Review
Table 2. Treatments and corresponding abbreviations and classes

Treatment

Abbreviation

Class

Placebo

Pbo

NA

Azithromycin 250 mg once daily

AZM250 once daily

Macrolide

Azithromycin 250 mg once daily 3 times per week

AZM250 once daily (3x weekly)

Macrolide

Azithromycin 500 mg once daily 3 times per week

AZM500 once daily (3x weekly)

Macrolide

Azithromycin 500 mg once daily 3 times per month

AZM500 once daily (3x monthly)

Macrolide

Azithromycin 500 mg once daily (for first 3 days),

azithromycin 250 mg every 2 days (intermittent)

for rest of treatment duration

AZM500 once daily (3 days) then

AZM250 (alternating day days)

Macrolide

Clarithromycin 500 mg once daily

CLR500 once daily

Macrolide

Erythromycin 250 mg three times daily

ERY250 three times daily

Macrolide

Erythromycin 250 mg twice daily

ERY250 twice daily

Macrolide

Erythromycin 125 mg 3 times daily

ERY125 three times daily

Macrolide

Erythromycin 200 to 400 mg once daily

ERY200/400 once daily

Macrolide

Roxithromycin 300 mg once daily

ROX300 once daily

Macrolide

Doxycycline 100 mg once daily

DOX100 once daily

Tetracycline

Roxithromycin 300 mg once daily +

Doxycycline 100 mg once daily

ROX300 once daily + DOX100

once daily

Macrolide + tetracycline

Moxifloxacin 400 mg once daily

(5 days every 4 weeks)

MOX400 once daily

(5 days every 4 weeks)

Quinolone

Moxifloxacin 400 mg once daily

(5 days every 8 weeks)

MOX400 once daily

(5 days every 8 weeks)

Quinolone

Abbreviations

AZM: azithromycin; CLR: clarithromycin; ERY: erythromycin; DOX: doxycycline; MOX: moxifloxacin; NA: not applicable; Pbo: placebo; ROX: roxithromycin.

Figures and Tables -
Table 2. Treatments and corresponding abbreviations and classes
Navigate to table in Review
Table 3. Exacerbations: studies included with time to first exacerbation data

Study

Treatments compared

Log hazard ratio

Standard error

Albert 2011

Pbo

AZM 250 mg once daily

‐0.31

0.07

He 2010

Pbo

ERY125 mg 3 times daily

‐0.59

0.29

Seemungal 2008

Pbo

ERY 250 mg twice daily

‐0.45

0.14

Simpson 2014

Pbo

AZM 250 mg once daily

‐0.99

0.62

Uzun 2014

Pbo

AZM 500 mg once daily 3 times per week

‐0.54

0.16

Blasi 2010

Pbo

AZM 500 mg once daily 3 times per week

‐1.69

0.60

Abbreviations

AZM: azithromycin; ERY: erythromycin; Pbo: placebo.

Figures and Tables -
Table 3. Exacerbations: studies included with time to first exacerbation data
Navigate to table in Review
Table 4. Exacerbations: studies included with the number of people with one or more exacerbations

Study

Treatment 1 (N)

No. of events

Treatment 2 (N)

No. of events

Treatment 3 (N)

No. of events

Treatment 4 (N)

No. of events

Berkhof 2013

Pbo

(42)

17

AZM 250 mg once daily 3 times a week

(42)

10

‐

‐

‐

‐

Brill 2015

Pbo

(24)

13

DOX100 mg once daily

(25)

15

AZM 250 mg once daily

3 times per week

(25)

10

MOX 400 mg once daily

(5 days every 4 weeks)

(25)

10

Sethi 2010

Pbo

(580)

295

MOX 400 mg once daily

(5 days every 8 weeks)

(569)

269

‐

‐

‐

‐

Suzuki 2001*

Pbo

(54)

30

ERY 200 to 400 mg once daily

(55)

6

‐

‐

‐

‐

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin;Pbo: placebo.

*This study was included only as a sensitivity analysis ‐ reported in Appendix 4.

Figures and Tables -
Table 4. Exacerbations: studies included with the number of people with one or more exacerbations
Navigate to table in Review
Table 5. Exacerbations: model fit statistics

DIC

SD (95% CrI)

Total residual deviance*

Fixed class effect models

Fixed effect model

51.31

‐

15.17

Random effects model

52.17

0.16 (0.006 to 0.519)

13.61

Abbreviations

CrI: credible interval; DIC: deviance information criterion; SD: standard deviation.

*Compared to 14 data points.
Figures and Tables -
Table 5. Exacerbations: model fit statistics
Navigate to table in Review
Table 6. Exacerbations: interventions and treatment classes

Intervention

Treatment class

N

1

Pbo

Placebo

1345

2

AZM 250 mg once daily

Macrolide

573

3

AZM 250 mg once daily 3 times per week

Macrolide

67

4

AZM 500 mg once daily 3 times per week

Macrolide

57

5

ERY 250 mg 3 times daily

Macrolide

53

6

ERY 125 mg 3 times daily

Macrolide

18

7

DOX 100 mg once daily

Tetracycline

25

8

MOX 400 mg once daily (5 days every 8 weeks)

Quinolone

569

9

MOX 400 mg once daily (5 days every 4 weeks)

Quinolone

25

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; Pbo: placebo.

Figures and Tables -
Table 6. Exacerbations: interventions and treatment classes
Navigate to table in Review
Table 7. Exacerbations: number of trials, number of participants, and relative effect estimates for all class comparisons

Comparison

Hazard ratios

Number of trials

N

Intervention

Comparator

Median

95% CrI

Macrolide

Placebo

0.67

0.60 to 0.75

9

1509

Tetracycline

Placebo

1.29

0.66 to 2.41

1

49

Quinolone

Placebo

0.89

0.75 to 1.04

2

1198

Tetracycline

Macrolide

1.93

0.99 to 3.62

1

50

Quinolone

Macrolide

1.32

1.08 to 1.61

1

50

Quinolone

Tetracycline

0.69

0.37 to 1.34

1

50

Abbreviations

CrI: credible interval.

Figures and Tables -
Table 7. Exacerbations: number of trials, number of participants, and relative effect estimates for all class comparisons
Navigate to table in Review
Table 8. Exacerbations: number of participants and rank statistics for each class (sorted by mean rank)

Class

N

Mean

Median

95% CrI

Macrolide

768

1.0

1

1 to 2

Quinolone

594

2.2

2

2 to 3

Placebo

1345

3.1

3

2 to 4

Tetracycline

25

3.6

4

1 to 4

Abbreviations

CrI: credible interval.

Figures and Tables -
Table 8. Exacerbations: number of participants and rank statistics for each class (sorted by mean rank)
Navigate to table in Review
Table 9. Change from baseline in SGRQ: number of trials, number of participants, and relative effects for all class comparisons

Comparison

Number of trials

N

Fixed effects‐fixed class effect

Random effects‐fixed class effect (uniform prior)

Random effects‐fixed class effect (empirical prior)

MD

95% CrI

MD

95% CrI

MD

95% CrI

Macrolide vs placebo

6

1158

‐2.30

‐3.61 to ‐0.99

‐2.34

‐4.28 to ‐0.39

‐2.28

‐5.19 to 1.01

Tetracycline vs placebo

1

49

1.18

‐1.49 to 3.84

1.14

‐2.47 to 4.62

1.20

‐4.62 to 7.19

Quinolone vs placebo

2

1078

‐1.33

‐2.97 to 0.32

‐1.42

‐4.04 to 1.05

‐1.44

‐5.99 to 3.07

Tetracycline vs macrolide

1

50

3.47

0.92 to 6.03

3.47

0.01 to 6.83

3.47

‐2.38 to 9.22

Quinolone vs macrolide

1

50

0.97

‐0.10 to 2.95

0.91

‐2.01 to 3.71

0.84

‐4.24 to 5.51

Quinolone vs tetracycline

1

50

‐2.50

‐5.32 to 0.30

‐2.56

‐6.33 to 1.16

‐2.63

‐8.96 to 3.37

Abbreviations

CrI: credible interval;MD: mean difference; SGRQ: St George's Respiratory Questionnaire.

Figures and Tables -
Table 9. Change from baseline in SGRQ: number of trials, number of participants, and relative effects for all class comparisons
Navigate to table in Review
Table 10. Change from baseline in SGRQ: included studies

Study

Endpoint (weeks)

Treatments compared

Mean difference vs Placebo

SE of Mean difference

Albert 2011

52

Placebo

AZM 250 mg once daily

‐2.2

0.7853

Berkhof 2013

12

Placebo

AZM250 mg once daily 3 times per week

‐7.5

2.5456

He 2010

26

Placebo

ERY 125 mg 3 times daily

‐3

5.6801

Sethi 2010

48

Placebo

MOX 400 mg once daily (5 days every 8 weeks)

‐1.2

0.9231

Simpson 2014

12

Placebo

AZM 250 mg once daily

6.1

5.31927

Uzun 2014

52

Placebo

AZM500 mg once daily 3 times per week

‐0.61

2.622449

Brill 2015*

13

Placebo

a. DOX 100 mg once daily

b. AZM 250 mg once daily 3 times per week

c. MOX 400 mg once daily (5 days every 4 weeks)

a. 1.02

b. ‐2.29

c. ‐1.88

a. 3.135

b. 3.212

c. 3.426

a. 0.88

b. ‐2.35

c. ‐2.25

a. 3.132

b. 3.085

c. 3.233

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; SE: standard error; SGRQ: St George's Respiratory Questionnaire.

*Data in bold used in sensitivity analysis.
Figures and Tables -
Table 10. Change from baseline in SGRQ: included studies
Navigate to table in Review
Table 11. Change from baseline in SGRQ: number of participants and rank statistics for each class (sorted by mean rank)

Treatment class

Number of participants

Mean

Median

95% CrI

Macrolide

578

1.17

1

1 to 2

Quinolone

528

1.93

2

1 to 3

Placebo

1106

3.14

3

2 to 4

Tetracycline

25

3.76

4

2 to 4

Abbreviations

CrI: credible interval; SGRQ: St. George's Respiratory Questionnaire.

Figures and Tables -
Table 11. Change from baseline in SGRQ: number of participants and rank statistics for each class (sorted by mean rank)
Navigate to table in Review
Table 12. Serious adverse events: model fit statistics

DIC

Between‐study SD (95% CrI)

Total residual deviance*

Fixed class models

Fixed treatment effect

113.8

‐

21.58

Random treatment effects

113.2

0.44 (0.02 to 1.28)

18.59

Abbreviations

* Compare to 19 data points

CrI: credible interval; DIC: deviance information criterion; SD: standard deviation.

Figures and Tables -
Table 12. Serious adverse events: model fit statistics
Navigate to table in Review
Table 13. Serious adverse events: number of trials, participants, and relative effects for all class comparisons

Treatment class comparison

Number of trials

N

Fixed effect‐fixed class effect model

Random effects‐fixed class effect (uniform prior)

Random effects‐fixed class effect (empirical prior)

OR

95% CrI

OR

95% CrI

OR

95% CrI

Macrolide vs placebo

8

1930

0.76

0.62 to 0.93

0.72

0.38 to 1.14

0.73

0.45 to 1.07

Quinolone vs placebo

1

1149

1.00

0.72 to 1.34

1.21

0.29 to 3.24

1.08

0.42 to 2.27

Macrolide + tetracycline vs placebo

1

195

0.97

0.52 to 1.66

1.12

0.27 to 2.84

1.00

0.36 to 2.19

Quinolone vs macrolide

0

0

1.32

0.90 to 1.89

1.88

0.41 to 5.67

1.56

0.55 to 3.62

Macrolide + tetracycline vs macrolide

1

198

1.28

0.68 to 2.19

1.67

0.42 to 4.31

1.41

0.52 to 3.15

Macrolide + tetracycline vs quinolone

0

0

1.00

0.49 to 1.82

1.72

0.17 to 4.67

1.13

0.26 to 3.11

Abbreviations

CrI: credible interval; OR: odds ratio.

Figures and Tables -
Table 13. Serious adverse events: number of trials, participants, and relative effects for all class comparisons
Navigate to table in Review
Table 14. Serious adverse events: table of interventions and treatment classes

Intervention

Treatment class

N

1

Pbo

Pbo

1539

2

AZM 250 mg od

Macrolide

573

3

ERY 125 mg tds

Macrolide

54

4

ERY 250 mg bd

Macrolide

53

5

MOX 400 mg od (5 days every 8 weeks)

Quinolone

569

6

ROX 300 mg od + DOX 100 mg od

Macrolide + tetracycline

101

7

ROX 300 mg od

Macrolide

97

8

AZM 500 mg od (3x weekly)

Macrolide

47

9

AZM 500 mg od (for first 3 days),

AZM 250 mg every 2 days (intermittent)

for rest of treatment duration

Macrolide

147

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; Pbo: placebo; ROX: roxithromycin.

Figures and Tables -
Table 14. Serious adverse events: table of interventions and treatment classes
Navigate to table in Review
Table 15. Serious adverse events: studies included

Study name

Treatments compared

Number of participants

Number of events

Treatment 1

Treatment 2

Treatment 3

Treatment 1

Treatment 2

Treatment 3

Albert 2011

Pbo

AZM 250 mg once daily

559

558

NA

212

184

NA

He 2010

Pbo

ERY125 mg 3 times daily

18

18

NA

3

2

NA

Seemungal 2008

Pbo

ERY 250 mg twice daily

56

53

NA

12

14

NA

Sethi 2010

Pbo

MOX 400 mg once daily (5 days every 8 weeks)

580

569

NA

97

94

NA

Shafuddin 2015

Pbo

ROX 300 mg once daily

+ DOX 100 mg once daily

ROX 300 mg once daily

94

101

97

20

24

23

Simpson 2014

Pbo

AZM 250 mg once daily

15

15

NA

4

1

NA

Tan 2016

Pbo

ERY 125 mg 3 times daily

18

36

NA

3

2

NA

Uzun 2014

Pbo

AZM 500 mg once daily 3 times per week

45

47

NA

5

3

NA

Vermeersch 2019

Pbo

AZM 500 mg once daily (for first 3 days),

AZM 250 mg every 2 days (intermittent)

for rest of treatment duration

15

147

NA

48

25

NA

Abbreviations

AZM: azithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; NA: not applicable; Pbo: placebo; ROX: roxithromycin.

Figures and Tables -
Table 15. Serious adverse events: studies included
Navigate to table in Review
Table 16. Serious adverse events: total number of participants and rank statistics for each class (sorted by mean rank)

Treatment class

N

Mean

Median

95% CrI

Macrolide

971

1.33

1

1 to 3

Macrolide + tetracycline

101

2.61

2

1 to 4

Quinolone

569

2.95

3

1 to 4

Pbo

1539

3.12

3

2 to 4

Abbreviations

CrI: credible interval; Pbo: placebo.

Figures and Tables -
Table 16. Serious adverse events: total number of participants and rank statistics for each class (sorted by mean rank)
Navigate to table in Review
Table 17. Drug resistance or microbial sensitivity reported in included studies

Study, drug, duration (weeks)

Drug resistance/microbial sensitivity methods

Results

Conclusion

Albert 2011

AZM (52)

Nasopharyngeal swabs and expectorated sputum samples taken at baseline and every 3 months, assessment for resistance to macrolides. Only 15% of participants were able to produce sputum by the third month; therefore, assessments were limited to nasopharyngeal swabs

Prevalence of resistance to macrolides was 52% and 57%, respectively (P = 0.64)

During the study, 81% AZM and 41% placebo and bacteria were resistant to macrolides (P < 0.001)

People receiving AZM were less likely to be colonised with respiratory pathogens compared to placebo but were more likely to become colonised with macrolide‐resistant organisms. No association with exacerbations

Banerjee 2005

CLR (13)

Sputum sample was tested for potential pathogenic microorganisms: H influenzae, S pneumoniae, M catarrhalis, H parainfluenzae, S aureus, P aeruginosa, K pneumoniae

At the start, 90% of isolates were due to S pneumoniae, H influenzae, M catarrhalis. Some patients had more than one PPM in sputum. After 3 months of CLR, number of people with sputum PPM increased from 12 to 15. Number of bacterial isolates did not increase. In the placebo group, this increased from 10 to 16, and the number of bacterial isolates increased from 15 to 25

CLR did not significantly reduce mean number of H influenzae, S pneumoniae, or M catarrhalis bacterial isolate compared to placebo

No multi‐resistant gram‐negative organisms emerged in the CLR group. CLR did not significantly change the mean CFU number per bacterial isolate compared to placebo

Contradicts other CLR studies that show the opposite due to lack of compliance in the trial. The study did not measure MIC90, which would have been ideal to detect changes in antibiotic susceptibility or resistance with time

Berkhof 2013

AZM (12)

Sputum samples collected

A reduction in respiratory pathogens was seen in the AZM group compared to the placebo group. One patient in the AZM group at 12 weeks had AZM‐resistant bacteria (S aureus)

Dose given seemed effective compared to other studies

Blasi 2010

AZM (26)

Minimum inhibitory concentration (MIC) used to determine bacterial counts

P aeruginosa became resistant to ceftazidime after 6 months of treatment in 1 patient in the AZM group. An ERY‐resistant S pneumoniae was found in 1 patient in the AZM group at 6 months

Not associated with significant effects of reduction in bacterial load or bacterial eradication. Patients with long‐term use of AZM had no resistance except for 1 person

Brill 2015

MOX, DOX, AZM (13)

Resistance to the 3 tested antibiotics, change in sputum bacterial load via quantitative culture (qPCR)

Pulsed MOX demonstrated the largest fall in bacterial numbers on culture but was associated with increased adverse events

Resistance was found in all 3 antibiotic arms of at least 3 times pre‐treatment values. With baseline adjustment of MIC, MOX was associated with an increase in MIC for isolates cultured in sputum compared to placebo (e.g. DOX group were more likely to be resistant to DOX vs placebo)

There was an increase in resistance of airway bacteria to all 3 antibiotics

He 2010

Sputum samples/bacteriology

9 ERY and 7 placebo patients had bacterial growth at baseline. 4 had > 1 organism. At 6 months, there was significant bacterial growth, and 3 specimens had > 1 organism. There was no detection difference in the rate of identifying the 3 main micro‐organisms between the 2 groups

Seemungal 2008

ERY (52)

Sputum samples

Sensitivity testing

H influenzae detection positive in 27% of stable samples and in 40% of exacerbation samples. All H influenzae were resistant to ERY. S pneumoniae was found in 7% and 10%, respectively. No difference in detection rate for any organism between both arms at any follow‐up time points.

Sensitivity testing found that 33/69 showed no growth at baseline. Those who tested positive at baseline were resistant to H influenzae (ERY = 10, P = 12), S pneumoniae (ERY = 1, P = 5 all sensitive), M catarrhalis (ERY = 1, P = 2 all sensitive)

At 12 months, 26/43 samples had no significant growth. Of those samples that were positive, H influenzae (ERY = 1, P = 3), S pneumoniae (ERY = 1 resistant, P = 2 all sensitive), M catarrhalis (2 = sensitive, P = 2)

Microbial resistance was not dependent on the use of ERY. Only 1 case of ERY resistance occurred in the macrolide group at 52 weeks. The number of participant in the study was small; therefore interpretation of these results is not definitive

Sethi 2010

MOX (48)

Sputum samples

Over the 48‐week treatment, there was a reduction in the number of participants with pathogens isolated, with greater reduction with MOX vs placebo. No difference in MIC increases that were sustained

Isolates showed that 1 patient in the MOX group was S pneumoniae resistant at week 40, which was not associated with exacerbations and was not persistent at further visits. For S aureus, 1 to 3 isolates were MOX resistant at baseline and at other time points but did not persist and were not related to exacerbations. Median MIC of MOX against P aeruginosa at 24 weeks was 4 mg/L but was reduced to 1 mg/L to levels at randomisation for the MOX group. Median MIC in placebo group increased from 0.5 to 2 mg/L among those who completed treatment

No further comments

Uzun 2014

AZM (52)

Macrolide resistance by sputum culture

32/47 AZM gave samples. 32/45 in placebo gave samples. Most common bacteria were H influenzae, S pneumoniae, and P aeruginosa. At follow‐up fewer in the AZM group had positive cultures compared to the placebo group

Macrolide resistance was seen in 3 AZM and in 11 placebo (P = 0.036)

The number of sputum samples overall was low. Like Albert 2011, AZM group was less likely to be colonised with respiratory pathogens and acquisition of macrolide‐resistant bacteria was significantly reduced

Vermeersch 2019

AZM

Sputum samples

Bacterial samples obtained contained H influenzae, S pneumoniae, P aeruginosa, M catarrhalis, and S aureus. At follow‐up, there were no significant group differences (AZM or placebo) for positive sputum cultures with acquired pathogens, neither for acquired macrolide‐resistant bacteria

Macrolide resistance was monitored, but induced sputum was not required per protocol; the limited number of spontaneous sputum samples did not allow for thorough evaluation of antibiotic resistance induced by AZM on top of standard treatment

Abbreviations

AZM: azithromycin; B catarrhalis :Branhamella catarrhalis;CLR: clarithromycin; CFU: colony‐forming unit; DOX: doxycycline; ERY: erythromycin; H influenzae: Haemophius influenzae;MIC: minimum inhibitory concentration; MIC90: MIC required to inhibit growth of 90% or organisms; M catarrhalis: Morexella catarrhalis;MOX: moxifloxacin; NA: not applicable; P aeruginosa: Pseudomonas aeruginosa;PPM: parts per million; qPCR: quantitative polymerase chain reaction; S aureus:Staphylococcus aureus;S pneumoniae: Streptococcus pneumoniae.

Figures and Tables -
Table 17. Drug resistance or microbial sensitivity reported in included studies
Navigate to table in Review
Table 18. Mortality: numbers of deaths in treatment and placebo groups in included studies

Study ID

Antibiotic class

Antibiotic

Placebo or control or standard treatment

Albert 2011

Macrolide

AZM: 18/570 (3%)

20/572 (4%)

Banerjee 2005

Macrolide

CLR: 0/31

0/36

Berkhof 2013

Macrolide

AZM: 0/42

0/42

Blasi 2010

Macrolide

AZM: 1/11 (9%)

5/11 (45%)

Brill 2015

Quinolone

Tetracycline

Macrolide

MOX: 0/25

DOX: 0/25

AZM: 0/25

0/24

He 2010

Macrolide

ERY: 0/18

0/18

Mygind 2010

Macrolide

AZM: 74/287 (25%)

81/288 (28%)

Seemungal 2008

Macrolide

ERY: 0/53 (0%)

1/56 (2%)

Sethi 2010

Quinolone

MOX: 13/753 (2%)

13/584 (2%)

Shafuddin 2015

Macrolide+

tetracycline

ROX: 3/97 (3%)

DOX+ROX: 5/101 (5%)

5/94 (5%)

Simpson 2014

Macrolide

AZM: 0/15

0/15

Singh 2019

Tetracycline

DOX: 0/30

0/30

Suzuki 2001

Macrolide

ERY: 0/55

0/54

Tan 2016

Macrolide

ERY: 0/18

0/18

Uzun 2014

Macrolide

AZM: 0/47 (0%)

2/45 (4%)

Vermeersch 2019

Macrolide

AZM: 3/147 (2%)

6/154 (3%)

Wang 2017

Macrolide

AZM: 0/43

0/43

Abbreviations

AZM: azithromycin; CLR: clarithromycin; DOX: doxycycline; ERY: erythromycin; MOX: moxifloxacin; ROX: roxithromycin.

Figures and Tables -
Table 18. Mortality: numbers of deaths in treatment and placebo groups in included studies
Navigate to table in Review
Table 19. Adverse events across all studies

Study ID, drug, dose, schedule (weeks' duration)

Adverse events

Antibiotic (n)

Comparator (n)

Reporting of exacerbations as AEs

Albert 2011, AZM, 250 mg once daily (52)

Discontinuation due to: audiogram‐confirmed hearing decrement (142), tinnitus (4), gastrointestinal tract (11), QTc prolongation (6), allergic reaction (5), abnormal laboratory test (4), other (10)

Discontinuation due to: audiogram‐confirmed hearing decrement, tinnitus (4), neoplasm (3), GI tract (6), QTc prolongation (4), allergic reaction (8), abnormal laboratory test (3), other (17)

Exacerbation was not reported as an AE

He 2010, ERY, 125 mg 3 times daily (26)

Discontinued due to: abdominal pain (1), complication of left heart failure (1)

Discontinued due to: respiratory insufficiency (2), other (1)

Exacerbation was not reported as an AE

Seemungal 2008 ERY, 250 mg twice daily (52)

Upper gastrointestinal (5), lower gastrointestinal (3), rash (3), other (3)

Upper gastrointestinal (5), lower gastrointestinal (3), rash (2), other (2)

Exacerbation was not reported as an AE

Sethi 2010, MOX, 400 mg once daily (5 days every 8 weeks) (48)

Cardiac disorders (3), gastrointestinal (diarrhoea, nausea, vomiting) (27), general disorders/administration site conditions (4), asthenia (3), immune system disorders (4), hypersensitivity (3), infections and infestations (5), musculoskeletal and connective tissue disorders (3), nervous system disorders (6), dizziness (3), respiratory, thoracic and mediastinal disorders (8), dyspnoea (4), skin and subcutaneous tissue disorders (5), AEs leading to discontinuation (26)

Cardiac disorders (1), gastrointestinal (diarrhoea, nausea, vomiting) (4), general disorders and administration site conditions (2), asthenia (0), hypersensitivity (0), infections and infestations (3), musculoskeletal and connective tissue disorders (1), nervous system disorders (4), dizziness (1), respiratory, thoracic, and mediastinal disorders (0), dyspnoea (0), skin and subcutaneous tissue disorders (5), AEs leading to discontinuation (16)

Exacerbation was not reported as an AE

Shafuddin 2015, ROX 300 mg once daily + DOX 100 mg once daily; or ROX 300 mg once daily (12)

Roxithromycin + doxycycline: nausea (12), diarrhoea (2), headache (4), abdominal pain (3), reflux (2), vomiting (1), abnormal liver function (1), abnormal ECG (1), rash (1), dyspnoea (0), dizziness (0), oral candidiasis (0), gastrointestinal upset (0)

Roxithromycin alone: nausea (13), diarrhoea (3), headache (1), abdominal pain (1), reflux (1), vomiting (3), abnormal liver function (2), abnormal ECG (0), rash (1), dyspnoea (1), dizziness (4), oral candidiasis (2), gastrointestinal upset (2)

Nausea (1), diarrhoea (1), headache (1), abdominal pain (1), reflux (0), vomiting (0), abnormal liver function (0), abnormal ECG (0), dyspnoea (2), dizziness (0), oral candidiasis (3), gastrointestinal upset (2)

Exacerbation was not reported as an AE

Simpson 2014, AZM, 250 mg once daily (12)

Diarrhoea (1), abdominal pain (0), nausea (0), vomiting (0), fever (0), headache (0), rash (0), hearing loss (0), other (10)

Diarrhoea (5), abdominal pain (0), nausea (0), vomiting (0), fever (0), headache (0), rash (0), hearing loss (0), other (9)

Exacerbation was not reported as an AE

Tan 2016, ERY, 125 mg 3 times daily (52)

Withdrawal due to: abdominal pain (1), left‐sided heart failure (1)

Withdrawal due to: respiratory insufficiency (2), unknown (1)

Exacerbation was not an outcome in the study

Uzun 2014, AZM, 500 mg once daily 3 times per week (52)

Diarrhoea (9), nausea/vomiting (3), other (4)

Diarrhoea (1), nausea/vomiting (2), other (7)

Exacerbaiton was not reported as an AE. 2 fatal SAEs occurred due to COPD respiratory failure, which were counted in the mortality outcome

Vermeersch 2019, AZM, 500 mg once daily (for first 3 days), AZM 250 mg every 2 days for rest of treatment duration (13)

Diarrhoea (20), nausea (12), anorexia (9), hearing loss (1), syncope (1)

Diarrhoea (15), nausea (11), anorexia (8), hearing loss (6), syncope (2)

Exacerbation was not reported as an AE

Abbreviations

AE: adverse event;AZM: azithromycin; DOX: doxycyline; ECG: electrocardiogram; ERY: erythromycin; MOX: moxifloxacin; NR: not reported; QTc: corrected QT interval; ROX: roxithromycin; SAE: serious adverse event.

Figures and Tables -
Table 19. Adverse events across all studies
Navigate to table in Review