Bayesian Sensitivity Analysis in Survival and Longitudinal Trials with Missing Data

Abstract

AbstractSurvival and longitudinal clinical trials are commonly conducted to evaluate experimental drug, biologic, and vaccine products. Conventional methods such as the log-rank test and the Cox proportional hazards model assume non-informative censoring for time-to-event data, and mixed model analysis assumes missing-at-random (MAR) in longitudinal trials. Although such assumptions play a critical role in influencing the outcome of the analysis, there are no formal methods to validate such assumptions from the data at hand. Sensitivity analyses are often recommended to test the robustness of an analysis that depends on such assumptions. In this chapter, we discuss how to perform practical sensitivity analysis, in a Bayesian modeling setting, to handle missing and censored data in clinical trials. Specifically, we focus on the delta-adjusted and control-based imputation strategies under informative censoring or missing-not-at-random (MNAR) mechanisms. Applications to real clinical trials are presented to demonstrate these methods.