Confidence Intervals for Heterogeneity Measures in Meta-analysis

Abstract

Two methods of quantifying heterogeneity between studies in meta-analysis were studied. One method quantified the proportion of the total variance of the effect estimate due to variation between studies (R-I), and the other calibrated the variance between studies to the size of the effect itself through a between-study coefficient of variation (CVB). Bootstrap and asymptotic confidence intervals for R-I and CVB were derived and evaluated in an extensive simulation study that covered a wide range of scenarios likely to be encountered in practice. The best performance was given by asymptotic Wald confidence intervals developed for R-I and CVB. The use of these heterogeneity measures together with their confidence intervals was illustrated in 5 typical meta-analyses. A new user-friendly SAS macro (SAS Institute, Inc., Cary, North Carolina) is provided to implement these methods for routine use and can be downloaded at the last author's website.