Incorporating uncertainty of betweenstudy variance estimates increases the reliability of biased randomeffects metaanalysis
*Kristian Thorlund, Copenhagen trial
Keywords: metaanalysis, betweenstudy variability, randomeffects model
Results of clinical trials are often combined in a metaanalysis using the randomeffects model. Despite its theoretical advantages the model is often criticised for being overly sensitive to various sources of bias. From a modelling perspective such shortcomings may be fixed by incorporating uncertainty of the betweenstudy variance estimate. This technique is, however, has not yet been tested rigorously. We conduct a comprehensive simulation study to compare metaanalytic inferences from the randomeffects model with and without incorporating uncertainty of the betweenstudy variance estimate. Our results strongly suggest that incorporating uncertainty of the betweenstudy variance estimate in the randomeffects model will reduce bias of pooled treatment effects, reduce type I error, and improve the coverage of confidence intervals in biased metaanalyses.
