Assessing the comparative effects of interventions in COPD: a tutorial on network meta-analysis for clinicians

Abstract To optimize patient outcomes, healthcare decisions should be based on the most up-to-date high-quality evidence. Randomized controlled trials (RCTs) are vital for demonstrating the efficacy of interventions; however, information on how an intervention compares to already available treatments and/or fits into treatment algorithms is sometimes limited. Although different therapeutic classes are available for the treatment of chronic obstructive pulmonary disease (COPD), assessing the relative efficacy of these treatments is challenging. Synthesizing evidence from multiple RCTs via meta-analysis can help provide a comprehensive assessment of all available evidence and a “global summary” of findings. Pairwise meta-analysis is a well-established method that can be used if two treatments have previously been examined in head-to-head clinical trials. However, for some comparisons, no head-to-head studies are available, for example the efficacy of single-inhaler triple therapies for the treatment of COPD. In such cases, network meta-analysis (NMA) can be used, to indirectly compare treatments by assessing their effects relative to a common comparator using data from multiple studies. However, incorrect choice or application of methods can hinder interpretation of findings or lead to invalid summary estimates. As such, the use of the GRADE reporting framework is an essential step to assess the certainty of the evidence. With an increasing reliance on NMAs to inform clinical decisions, it is now particularly important that healthcare professionals understand the appropriate usage of different methods of NMA and critically appraise published evidence when informing their clinical decisions. This review provides an overview of NMA as a method for evidence synthesis within the field of COPD pharmacotherapy. We discuss key considerations when conducting an NMA and interpreting NMA outputs, and provide guidance on the most appropriate methodology for the data available and potential implications of the incorrect application of methods. We conclude with a simple illustrative example of NMA methodologies using simulated data, demonstrating that when applied correctly, the outcome of the analysis should be similar regardless of the methodology chosen.