Handling Both-Armed Zero-Events Studies in Meta-Analysis: A Simulation Study

• Activity Number: 692
• Type: Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ji Cheng*+ and Eleanor Pullenayegum and John K. Marshall and Lehana Thabane
• Companies: St. Joseph's Healthcare Hamilton /McMaster University and McMaster University and McMaster University, Department of Medicine and St Joseph's Healthcare Hamilton
• Keywords: zero-event ; meta-analysis ; simulation

Abstract

Background There is no consensus on whether studies with no observed events in both the treatment and control arms should be included in a meta-analysis. Currently, they are handled differently depending on the choice of effect measure and analysis approach: in classical meta-analysis, they are included for risk difference, but excluded for odds ratio or relative risk, while Bayesian approach includes all. We conducted a simulation study to evaluate the impact of excluding and including both-armed zero-event studies for both classical and Bayesian meta-analyses. Method We simulated 1000 datasets for different scenarios varying the parameters of baseline event rate, treatment effect and number of studies. We evaluated the performance of commonly used pooling methods in classical and Bayesian approaches comparing bias, mean squared error, coverage and empirical statistical power. Result We believe including both-armed zero-event studies in meta-analysis reduces bias, increases precision and makes results comparable across methods. Conclusion All studies with relevant data should be included in primary meta-analysis with sensitivity analyses to assess the impact of their exclusion.