Advanced meta-analysis 1: Random-effects methods to be implemented in RevMan
Joanne McKenzie, Monash University
Elizabeth Korevaar, Methods in Evidence Synthesis Unit, School of Public Health and Preventive Medicine, Monash University
Simon Turner, Methods in Evidence Synthesis Unit, School of Public Health and Preventive Medicine, Monash University
Background: Meta-analysis is typically used to estimate the mean effect size of an outcome of interest. However, also of interest is estimation of the variability in the effect sizes (heterogeneity), because this allows assessment of the consistency of effects across studies and estimation of the likely effect of the intervention in an individual setting. The DerSimonian and Laird random-effects method is widely used (and is currently the only method available in RevMan) to estimate heterogeneity, however, its performance has long been challenged.
Objectives: To provide review authors with knowledge of new random-effects methods to be implemented in RevMan Web. This workshop is part of a series of workshops delivered by the Cochrane Statistical Methods Group and requires knowledge of the material presented at the workshop ‘Introduction to meta-analysis 2: dealing with heterogeneity’.
Description: Several alternative methods to the DerSimonian and Laird random-effects method have been proposed that have better statistical properties in specific scenarios. We will provide an overview of the new random effects methods to be implemented on RevMan Web. Specifically, we will introduce new methods to estimate the heterogeneity variance (and its confidence interval); calculate a confidence interval for the meta-analysis effect; and calculate a prediction interval. We will cover reasons for selecting a particular method; how the estimates should be interpreted; what the impact might be when updating Cochrane Reviews using the new methods; and what to write in a Cochrane protocol. We will use a combination of presentations, interactive exercises, and group work.