Introduction to meta-analysis 2: dealing with heterogeneity
Anna Chaimani, Université Paris Cité, Center of Research in Epidemiology and Statistics, Inserm, Paris
The studies in a meta-analysis may vary in their included populations, how the intervention of interest was used, how outcomes were assessed, and in study design and conduct. This variability in study properties may lead to variability in the outcomes across studies. This variability in outcomes in a meta-analysis is called heterogeneity. Determining whether heterogeneity across studies is present in a meta-analysis and identifying its possible causes are critical components of any meta-analysis.
Objectives: To provide review authors with the knowledge to understand and investigate heterogeneity across studies in a meta-analysis and to recognise the limitations of the methods available. This workshop is part of a series of workshops delivered by the Cochrane Statistical Methods Group.
Description: We will address approaches to dealing with heterogeneity across studies in a meta-analysis. We first discuss potential sources of across-study variability and provide an overview of methods for identifying whether heterogeneity is present in a meta-analysis. We then focus on issues related to dealing with heterogeneity once it has been identified. In particular, we discuss whether or not to combine results; how heterogeneity is handled in the choice between fixed-effect and random-effects analyses; and the use of subgroup analyses (with a brief mention of meta-regression). Discussion will be supplemented with practical examples from the Cochrane Database of Systematic Reviews.