Inclusion of Zero-Event Trials Using Continuity Correction in Meta-Analyses of Rare Events

• Activity Number: 536
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Tianyue Zhou*+
• Companies: Sanofi
• Keywords: meta analysis ; fixed effects ; Mantel-Haenszel method ; continuity correction

Abstract

Meta-analysis has been widely used to combine data with low event rate across comparative clinical studies. When dealing with rare events, a substantial proportion of studies may not have any events of interest. In common practice, meta-analyses on a relative scale (relative risk or odds ratio [OR]) remove zero-event studies while meta-analyses using risk difference [RD] as the effect measure include them. The impact of zero-event trials on estimates of OR and RD in terms of bias, coverage and length of 95% confidence interval [CI], power of the test and type 1 error was examined through simulation. Mantel-Haenszel method is chosen to pool the estimates. Zero-event studies are included in analysis through 2 types of continuity corrections. For OR, zero-event trials had little impact on analysis of the treatment effect when 'treatment arm' continuity correction constrained to sum to 0.01 was used. However, zero-event trials should be included when the percentage of such trials was high (>=40%) to avoid huge CIs. For RD, zero-event trials should be included using constant continuity correction of 0.5 or 'treatment arm' continuity correction constrained to 1 to control type 1 error.