A0639
Title: Bayesian estimation and testing in random effect meta-analysis of rare binary adverse events with flexible variability
Authors: Ming Zhang - Southern Methodist University (United States)
Jackson Barth - Southern Methodist University (United States)
Johan Lim - Seoul National University (Korea, South)
Xinlei Wang - Southern Methodist University (United States)
Johan Lim - Seoul National University (Korea, South) [presenting]
Abstract: Meta-analysis of rare binary events has become a routine in the pharmaceutical industry to access the safety of healthcare intervention since people can hardly make a quantitative and decisive conclusion from an individual study alone. Various frequentist or Bayesian methods have been proposed to attempt to report the accurate estimated treatment effect or the inter-study heterogeneity. However, almost all approaches pre-defined a direction of variance between the control and the treatment group, which might be more appropriate to be decided by data. Recently, a new flexible binomial-normal hierarchical model has been proposed by assuming no direction of variance. However, they mainly focus on comparing the current widely-used methods rather than proposing a new estimator. Therefore, we adopt a Bayesian hierarchical approach and develop our estimator (FlexB) and Bayesian hypothesis testing process using the flexible random-effects model, and we compare our method with existing frequentist and Bayesian competitors via extensive simulation. As for Bayesian calculation, we creatively incorporate the new Polya-Gamma data-augmentation technique into our sampling process, which brings some computational convenience and stability for estimation. Two data examples, updated rosiglitazone data and glutathione S-transferase P1(GSTP1) GG genotype data, are analyzed by our approach as well.
