Bayesian Meta-analysis of Rare Events with Non-ignorable Missing Data
Meta-analysis is a powerful tool for drug safety assessment by synthesizing treatment-related toxicological findings from independent clinical trials. However, published clinical studies may or may not report all adverse events (AEs) if the observed number of AEs were fewer than a pre-specified study-dependent cutoff. Subsequently, with censored information ignored, the estimated incidence rate of AEs could be significantly biased. To address this non-ignorable missing data problem in meta-analysis, we propose a Bayesian multilevel regression model to accommodate the censored rare event data. The performance of the proposed Bayesian model of censored data compared to other existing methods is demonstrated through simulation studies under various censoring scenarios. Finally, the proposed approach is illustrated using data from a recent meta-analysis of 125 clinical trials involving PD-1/PD-L1 inhibitors with respect to their toxicity profiles.
READ FULL TEXT