An Introduction to Individual Participant Data Meta-analysis.

Meta-analysis using individual participant data (IPD-MA) from randomised controlled trials (RCTs) can strengthen evidence used for decision-making, and is considered the 'gold standard' approach. In this paper, we present the importance, properties, and main approaches of conducting an IPD-MA. We exemplify the main approaches of conducting an IPD-MA and how these can be used to obtain subgroup effects through estimation of interaction terms. IPD-MA has several benefits over traditional aggregate data meta-analysis. These include: standardization of definitions of outcomes and/or scales, re-analysis of eligible RCTs using the same analysis model across all studies, accounting for missing outcome data, detecting outliers, using participant-level covariates to explore intervention-by-covariate interactions, and tailoring intervention effects to participant characteristics. IPD-MA can be performed either in a two-stage or a one-stage approach. We exemplify the presented methods using two illustrative examples. The first real-life example includes six studies assessing sonothrombolysis with or without addition of microspheres against intravenous thrombolysis alone (i.e., control) in acute ischemic stroke participants with large vessel occlusions. The second real-life example includes seven studies evaluating the association between blood pressure levels after endovascular thrombectomy and functional improvement of acute ischemic stroke in patients with large vessel occlusion. IPD reviews can be associated with higher-quality statistical analysis and may differ from aggregate data reviews. Unlike individual trials that lack power, and aggregate data meta-analysis results which suffer from confounding and aggregation bias, the use of IPD allows us to explore intervention-by-covariate interactions. However, a key limitation of conducting an IPD-MA is retrieval of IPD from original RCTs. Time and resources should be carefully planned before embarking to retrieving IPD.