Handling related publications reporting real-world evidence in network meta-analysis: a case study in multiple sclerosis

Aim: The presence of two or more publications that report on overlapping patient cohorts poses a challenge for quantitatively synthesizing real-world evidence (RWE) studies. Thus, we evaluated eight approaches for handling such related publications in network meta-analyses (NMA) of RWE studies.Methods: Bayesian NMAs were conducted to estimate the annualized relapse rate (ARR) of disease -modifying therapies in multiple sclerosis. The NMA explored the impact of hierarchically selecting one pivotal study from related publications versus including all of them while adjusting for correlations.Results: When selecting one pivotal study from related publications, the ARR ratios were mostly similar regardless of the pivotal study selected. When including all related publications, there were shifts in the point estimates and the statistical significance. Conclusion: An a priori hierarchy should guide the selection among related publications in NMAs of RWE. Sensitivity analyses modifying the hierarchy should be considered for networks with few or small studies.Plain language summary What is this article about? Real-world evidence (RWE) studies are increasingly recognized as important data source in guiding clinical care and regulatory submissions. However, when synthesizing the evidence originating from RWE studies, different analyses of the same data source are identified often. To avoid the double-counting of patients in network meta-analysis, we investigated different approaches for synthesizing the evidence from such related publications. What were the results? When selecting a single pivotal publication by prioritizing different selection criteria (sample size, number of therapies evaluated, type of publication, or follow-up time), results were fairly similar. In the scenarios that included all related publications (instead of selecting single pivotal study), the overall conclusions sometimes changed inappropriately, even when correcting for correlations between publications. What do the results mean? When analyzing RWE data in a network meta-analysis, we recommend establishing selection criteria for choosing pivotal publications prior to quantitatively synthesizing the evidence from RWE studies. Tweetable abstract: This study in multiple sclerosis examines approaches for handling related publications in network meta-analyses of real-world evidence and recommends establishing an a priori hierarchy for selection.