Résumé : It is well recognized that standard methods of meta-analysis break down with rare binary events. Not only are effect sizes and within study variances badly estimated, but heterogeneity is generally not identifiable or strongly underestimated, and the overall summary index is biased.
Indeed, with commonly used sample sizes in randomized controlled trials (RCT), the number of events observed in each study is too low (e.g. 0-5 events when event rate is comprised within 1‰-2% and 50-250 patients are included per trial arm) to allow reliable estimates by standard meta-analysis methods and parameter estimation is plagued by bias.
Despite considerable efforts in the development of new methods to try to improve the estimates in sparse data meta-analysis, none of the methods developed so far have really provided a convincing solution, particularly in the random effects setting.
Therefore, in this presentation we will investigate by simulations the performance of standard meta-analysis methods (inverse variance and Der Simonian & Laird) and the improvement provided by some more sophisticated estimators.