An investigation of alternative statistical methods for stability data, including but not limited to the use of defensible Bayesian priors

摘要

Accurate shelf life estimation is very important to a variety of applications. This work specifically focuses on the pharmaceutical industry, where inaccurate estimation can lead to desirable consequences. Overestimation of shelf life could lead to consumption of drugs that are no longer stable and effective, while underestimation can cause drug development to be terminated or the consumer to discard good product prematurely. Thus accurate estimation of shelf life is essential to both consumers and producers. The purpose of this dissertation is to explore alternative statistical methodology that could be shown to offer improvement relative to current practices. In this dissertation we develop the alternative shelf life estimation methods that address the shortcomings of currently mandated procedures for shelf life estimation in the pharmaceutical industry. Specifically, we will first describe the current approach that is outlined in the International Conference on Harmonization (ICH) guideline, which is supported by the Food and Drug Administration (FDA), and its possible drawbacks. Then we will focus on characterizing different alternative approaches presented in the literature to date and summarize the findings. In attempt to develop a viable shelf life estimation procedure that has a chance of getting industry and regulatory acceptance we present three approaches to shelf life estimation: a Bayesian Augmented Mixed Model method, Direct Bayesian method and Posterior Predictive Distribution method and discuss their potential for shelf life estimation.