ObjectiveTo examine statistical performance of different rare-event meta-analyses methods.MethodsUsing Monte-Carlo simulation, we set a variety of scenarios to evaluate the performance of various rare-event meta-analysis methods. The performance measures included absolute percentage error, root mean square error and interval coverage.ResultsAcross different scenarios, the absolute percentage error and root mean square error were similar for Bayesian logistic regression model, generalized mixed linear effects model and continuity correction, but the interval coverage was higher with Bayesian logistic regression model. The statistical performances with Mantel-Haenszel method and Peto method were consistently suboptimal across different scenarios.ConclusionsBayesian logistic regression model may be recommended as a preferred approach for rare-event meta-analysis.
Objective To explore the differences between large and small studies in rare events meta-analysis. Methods Empirical data were collected from The Cochrane Systematic Review Database from January 2003 to May 2018. Meta-analyses with rare events, binary outcomes involving at least 5 studies, and at least 1 large study were screened. Peto and classical ORs were used to compare the magnitude, direction and P-value. Results A total of 214 meta-analyses were included. Among 214 pairs of ORs of large and small studies, 66 pairs (30.84%) were inconsistent in the direction of ORs based on Peto OR (Kappa =0.33), and 69 pairs (32.24%) were inconsistent in the direction of ORs based on classical OR. The Peto ORs resulted in smaller P-values compared to classic ORs in a substantial (83.18%) number of cases. Conclusion There are considerable differences between large and small studies in the results of meta-analysis of rare events.