Time-to-Event Meta-Analysis for Rare Safety Outcomes

• Activity Number: 167
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Xiao Ding*+ and Mat Soukup and Bradley McEvoy
• Companies: Gilead Sciences and FDA/CDER and FDA/CDER
• Keywords: meta-analysis ; time to event ; Cox regression ; log rank test ; rare outcome ; safety analysis

Abstract

Recent research has shown that similar results should be expected from three popular methods for time-to-event meta-analysis: stratified Cox regression, stratified log-rank, and Inverse-variance weighted Cox model. In the regulatory setting, meta-analysis approach is needed for safety evaluation when the outcome is somewhat rare (incidence rate often no more than 1%~2% per year). We found that the results from the three methods might differ dramatically for rare outcome. We also explore other alternatives to improve the performance of time-to-event such as the use of penalized MLE (Firth's procedure) and the use of sample size weighting to replace inverse-variance weighting. Statistical properties, including power, type 1 error rate, bias of the point estimate, and coverage of the confidence interval, are compared among all the methods. Under different setting, different method is recommended.