|
Abstract:
|
Meta-analysis is a powerful tool for drug safety assessment by synthesizing findings from independent clinical trials. However, a large number of published clinical studies may not report rare adverse events intentionally. To derive exact inference and robust estimates for the missing not at random data, we propose a Bayesian multilevel regression model to accommodate censored sparse binomial event data with a stochastic data-coarsening mechanism. Under the assumption of coarsened at random in coarsened data framework, the coarsening mechanism can be ignored for likelihood based inference. A sensitivity analysis is also suggested to assess whether it is appropriate to ignore the stochastic nature of the coarsening. The proposed approach is illustrated using data from a recent meta-analysis of 125 clinical trials in oncology involving PD-1/PD-L1 inhibitors with respect to their toxicity profiles. We demonstrate that if the censored information is ignored, the incidence rate of adverse event could be significantly overestimated.
|
