Rationale

Chronic obstructive pulmonary disease (COPD) affects around 174 million persons worldwide.1 The most common cause is smoking, and patients exhibit airflow obstruction that is not fully reversible. The disease is progressive, with declining lung function and a worsening of symptoms. Most troublesome are acute exacerbations manifested as a sudden worsening of symptoms, for example, severe coughing, shortness of breath and chest congestion, requiring urgent treatment and possibly hospitalisation. While smoking cessation remains the most effective intervention, the rate of exacerbation can be reduced by regular medication such as combination of long-acting beta-adrenoceptor agonists (LABAs),2 and inhaled corticosteroids (ICS)3 4 or long-acting muscarinic antagonists (LAMAs).5

COPD treatment guidelines are largely informed by randomised controlled trial (RCT) results,6 but we do not know if these findings apply to large patient populations not studied in trials. While RCTs will continue to be the gold standard for assessing the efficacy of medical interventions, they are expensive to conduct, and for practical and ethical reasons usually involve testing treatments in patient populations and within contexts which are sometimes very different from real life. In particular, RCT participants with COPD usually have no comorbidities and are otherwise healthy. Whereas often the very patients who are most commonly seen in a COPD outpatient clinic are those that would be excluded from clinical trials due to their comorbidities. As patient populations with chronic diseases such as COPD become more complex, the studies used to generate clinical evidence must reflect this.

Evidence of medication effects in routine care is vital for understanding the balance of treatment risks and benefits. RCTs have unique strengths, but results from trials are not always a good guide to the effects of drugs in routine clinical practice.7 National drug licensing authorities are now demanding better real-world evidence on which to make decisions and have introduced legislation mandating studies of both effectiveness and risk to be conducted in routine clinical care rather than the narrow and optimal confines of most randomised trials.8 9 While the conduct of observational studies to investigate possible drug harms is well established, the use of these studies to estimate treatment effectiveness is in its infancy. Issues of treatment channelling and indication bias mean that measuring the intended benefit of a treatment is beset with difficulties. While others have demonstrated it can be done in certain circumstances, we need more certainty about the methodology as it is applied in each disease area, since the issues of bias are likely to vary considerably.9

There is increasing evidence that dual long-acting bronchodilator therapy (LABA+LAMA) is more clinically beneficial compared with monotherapy or placebo10; however, there is also evidence that adding an ICS to LABA and LABA as triple therapy is better at reducing exacerbations and maintaining lung function.11 Furthermore, new treatments for reducing exacerbation frequency have become more evidenced based in recent years with commonly inhaled therapies (LABAs, LAMAs, ICS and combination devices) being prescribed alongside other treatments such as mucolytics (carbocysteine, N-acetylcysteine),12 antibiotics (macrolides) and inhaled colomycin13 14 and phosodiesterase-4 inhibitors such as roflumilast.15

Until recently, dual combination therapy (LABA+ICS) was the recommended treatment for highly symptomatic COPD patients at risk of exacerbations or severe airflow limitation.16 However, in the Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines (2017)17 triple therapy (combined LABA+LAMA+ ICS) is only recommended for patients categorised as severe. Given this changing landscape, and the future launch of combination triple therapies, it is important to identify and synthesise the available evidence from clinical trials and observational studies.

Objectives

This systematic literature review will assess clinical benefits and adverse events from maintenance treatment combinations in COPD described by RCTs and real-world evidence (RWE) studies. A network meta-analyses of outcomes from RCTs and RWE will be integrated if feasible.

The primary research question is: What are the clinical benefits of maintenance treatment combinations in COPD in terms of reducing exacerbations?

Secondary research questions include:

1. What is the long-term effectiveness of treatment combinations on improving lung function as measured by forced expiratory volume in 1 s (FEV₁)?

2. What is the long-term effectiveness of treatment combinations in improving patient-reported outcomes of respiratory-related quality of life, as measured by the St Georges Respiratory Questionnaire or the COPD Assessment Test?

3. What are the harms of each treatment combination including mortality, cardiovascular events, pneumonia, diabetes and bone fractures?

Eligibility criteria

Information sources

MEDLINE (Ovid interface, 1980 onwards) and Embase (Ovid interface, 1980 onwards) will be searched for potentially relevant articles using pre-defined search strategies. The International Prospective Register of Systematic Reviews (PROSPERO) will be searched periodically for relevant ongoing and completed systematic reviews. A manual search of the reference lists of all included studies will be conducted to check for other possibly relevant articles.

Manual searches will also be conducted to capture data not reported in the main publications and data from recent studies not yet published. Manual searches are anticipated to include conferences of interest and proceedings published from 2015 onwards, based on the assumption that trials are usually published within 2 to 3 years following presentation to a conference. In addition, we will search agencies’ websites (National Institute for Health and Care Excellence etc) and clinical trial registry clinicaltrials.gov.

Studies will be excluded if they investigate only single short-acting inhaled therapy versus placebo. Additionally, studies will be excluded if they exclude patients on the basis of comorbidities that are unlikely to be affected by the trialled drug.

Search strategy

The list of proposed search terms (table 2) will be reviewed by the primary and secondary reviewer and research group lead. The final search strategy will be checked for accuracy by the guarantor using the Peer Review of Electronic Search Strategies (PRESS) checklist.19 The search strategy will be developed in the first instance for MEDLINE and then adapted for use with Embase. Search filters limiting to studies published in 1980 onwards, studies published in English and studies in humans will be used. The disease terms, inhaler terms and outcome searches will be combined using the Boolean logic operator AND, while different choices of terms within these categories using will be combined using OR. The literature review searches will be updated at the end of the process.

Study records

Data items

We will use the PICOs (Population, Intervention, Comparator, Outcomes and Study characteristics) framework to systematise our data extraction. Within each of these five domains, information will be recorded on the following items:

• The study characteristics including setting, design, period of study, aims and objectives.

• The population including characteristics (including size, sex and age distribution, ethnicity, country), recruitment and sampling methods, inclusion/exclusion criteria.

• The intervention status definition and identification, number of exposed subjects, any exclusions.

• The comparators’ identification and definition of unexposed individuals, number of unexposed subjects, any exclusions.

• The outcomes definitions and identification of primary (exacerbation) and secondary outcomes (FEV₁ change from baseline, mortality, respiratory-related quality of life, adverse events), number of subjects, any exclusions, length of follow-up.

Outcomes and prioritisation

The following list of endpoints is anticipated; to protect against study selection bias for the length of studies, we anticipate splitting our analyses of exacerbations and lung function by length of study.

1. Number of moderate-to-severe exacerbations

   1. Short-term treatment (<20 weeks)

   2. Long-term treatment (≥20 weeks)

2. FEV₁ change from baseline

   1. Short-term treatment (<12 weeks)

   2. Long-term treatment (≥12 weeks)

3. Mortality

4. Respiratory-related quality of life (SGRQ, CAT)

5. Number of adverse and serious adverse events including:

   1. Cardiovascular events

   2. Pneumonia

   3. Diabetes

   4. Bone fractures

Risk of bias assessment

The primary reviewer will assess the risk of bias of each study included using the seven-item Cochrane Risk of Bias Tool.20 The Critical Appraisal Skills Programme21 quality assessment tools will be used to assess the quality of observational studies. Risk of bias assessment will be conducted independently by a second reviewer on a quarter of the selected studies and disagreements will be resolved in the first instance by discussion and then through consultation with a third reviewer. On completion of this process, the remaining studies will be assessed for the risk of bias by one reviewer. The Alternative Regression Test22 will be applied to evidence to test for publication bias.

Data synthesis

Three key aspects in the assessment of feasibility of network meta-analysis (NMA) (RCTs, RWE) will be examined:

1. The validity of an NMA of RCTs may be compromised by looking at: (1) clinical heterogeneity, defined as the extent of confounding bias due to an RCT’s similarity and consistency violations and (2) methodological heterogeneity arising from the quality of the individual RCTs that make up the evidence network. The assessment of clinical heterogeneity will be based on evaluation of the following data elements extracted from the RCTs:

   1. The study design, including the quality of methods employed.

   2. The admissibility criteria in terms of confounding factors (including subgroups of interest).

   3. The measurement and reporting of outcomes (eg, definitions of progression and time points, testing assumptions for underlying distributions).

   4. The comparators (common reference intervention) and dosages, schedules and routes of administration.

2. The validity of an NMA of RWE by looking at the similarity of patients populations across studies, similarity of treatments and differences in outcomes.

3. The use of complex statistical methods to control for confounding such as propensity score, disease risk score, and stabilised inverse probability weighting method.

Descriptive statistics will be conducted to describe the following:

• The main characteristics of the study populations at baseline (eg, age, sex, proportion of smokers, disease category as determined by the GOLD standard (A,B,C,D), FEV₁ status, etc.)

• Study designs (eg, details specific to RCT and RWE).

• Whenever possible, graphs showing the characteristics of studies by treatment or study will be developed. These findings will be used as a basis to identify potential sources of heterogeneity and to detect any potential outlier.

The problem of bias due to systematically differing entry criteria is not something we can pre-requisitely prevent; however, should there be large differences in population characteristics, we will conduct appropriate subgroup analyses.

Network meta-analysis

This NMA will incorporate the scenarios of both indirect and mixed treatment comparisons. The NMA will initially trial in feasibility design whether RCTs can be compared solely with other RCT designs of similar methodology or whether other types of study design such as observational studies can be included in direct comparison. Heterogeneity and inconsistency will be tested using the Wald test for inconsistency.

Potential differences across studies may include:

• The treatment dose/schedule or outcome definitions (exacerbation description).

• The distribution of each study and patient characteristics.

• Baseline risks associated with each treatment.

• Observed treatment effects.

In order to account for these differences, methods of meta-regression, sub-group and sensitivity analysis may be employed.

Should it not be possible to compare all study designs in a single NMA to show best-efficacy ranking of interventions, then separate NMAs will be performed for RCTs and the other study designs. The geometry of the network for each outcome measure will be presented in a network diagram of treatment nodes of variable sizes, comparison lines of variable thickness and loops to identify treatments with both direct and indirect comparisons.23

We plan to provide an overall combined ranking of interventions depending on their effect on exacerbation reduction compared with the combined weight of harm per intervention. Each harm will be weighted based on the severity of any harm reported as a result of using each intervention in each study. The study with best effect and least harm will be ranked first.

Statistical analysis

Due to the nature of this NMA, it is assumed that a random effects meta-analysis model will be chosen for the primary outcome of rate of exacerbation and continuous outcomes of health related quality of life and FEV₁. To estimate patient years for exacerbations, deaths and adverse events we plan to use methods previously reported.24 For change in continuous outcomes, that is, FEV₁ and SGRQ, we will either use the reported change or calculate the change from baseline based on the reported values. We will calculate the SD of change from the reported standard errors or confidence intervals. The primary outcome of risk of exacerbation will be reported as rate ratio and other continuous outcome measures will be reported as mean differences assuming normal distribution of data. The 95% CI for each summary statistics will be used to assess the statistical significance of each summary effect. The ranking of interventions will be based on their surface under cumulative ranking area, allowing clinicians to make treatment decisions based on relative effectiveness and risk of harm.25 Statistical analysis will be performed using Stata (StataCorp LP, College Station, TX, USA) and R (R Foundation for Statistical Computing, Vienna, Austria).

Confidence in cumulative evidence

The quality of evidence for our primary and secondary research questions will be assessed using those domains of the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) guidelines that are pertinent to observational studies, notably those relating to methodological flaws within the component studies and the consistency of results across different studies. Additional domains may be considered if deemed appropriate. The strength of evidence will be judged as ‘high’ (further research is unlikely to change our conclusion), ‘moderate’ (further research is likely to alter our conclusion) or ‘low’ (further studies are required to answer the research question with a high degree of confidence/increase confidence).

Patient involvement

No patients were involved in the design or analysis of this study.