|
Hal Li1, William Wang1
1Merck Research Laboratories, 351 N Sumneytown Pike, North Wales, PA 19454
Abstract
A Bayesian approach designed for learning and adaptive decision making is a natural methodology for safety evaluations. In this presentation, we discuss how to identify the sentinel safety events that trigger a formal monitoring plan on those events. A Bayesian proactive approach for monitoring safety profile using a false discovery rate related q-value is introduced. The similarity and relationship of the p-value and this Bayesian q-value is discussed with the impact of the probability of a null hypothesis as a factor. We also assessed both the impact of the number of simultaneous tests and the impact of the number of tests in sequence on the q-value.
For the sequential safety assessment, we assessed the impact of error spending functions on the Bayesian q-value, emphasizing the exponential decay error spending function. The implications of the Bayesian approach are presented including Bayesian stopping rules as well as current FDA guidance on Bayesian analysis. Simulation studies and graphical methods are used to illustrate the statistical properties.
|
