Background: Meta-analysis of rare events represents a challenging issue, particularly in the presence of studies having no events in both arms. However, it remains unclear how such studies were handled in the published meta-analyses; a systematic understanding may improve future practice.
Methods: We searched Cochrane systematic reviews published from January 2003 to May 2018 for meta-analyses that had at least one study with zero event in both arms. We extracted original data (in .rm5 format) from each eligible meta-analysis, including significance of the pooled effect, between study heterogeneity, effect measures, and how studies with zero event in both arms were handled. All these data were used only for academic research purposes.
Results: We identified 831 meta-analyses that had one or more studies with zero event in both arms. Of these, 206 (24.79%) were statistically significant (p < 0.05) in the pooled estimates and 49 (5.9%) trended to be significant (0.05 < p < 0.1); 75 (9.03%) chose odds ratio as an effect measure, 410 (49.34%) risk ratio, 343 (41.28%) risk difference, and three (0.36%) hazard ratio. Of the 831 meta-analyses, 458 (55.11%) removed studies with zero event in both arms, and 343 (41.28%) used the continuity correction to deal with such studies. No other methods (e.g. Bayesian, GLMM) were used.
Conclusion: The most common approaches to handling studies with no events in Cochrane systematic reviews were the exclusion of such studies or continuity correction. Other advanced methods were not used.