Some topics in multi-regional clinical trials and meta-analysis using Bayesian models

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] The dissertation consists of two distinct research topics. One is about sample size determination in Multi-Regional Clinical Trials (MRCTs), the other one is about the Bayesian meta-analysis using only summary statistics. For MRCTs, one key issue is how to assess the consistency of treatment effect among concerned regions and how to determine the sample sizes. Two frequentist consistency criteria defined by the guideline released by the Ministry of Health, Labour and Welfare in Japan (MHLW) are commonly used. In the dissertation, we propose six Bayesian consistency criteria. By assigning different values of parameters, the criteria can be easily tailored to fit the requirements of regulatory agencies. Considering normal endpoints, we derive the closed-form formulas or show the numerical methods to calculate the regional sample size based on those criteria. We also show the Bayesian approach to calculate the overall sample size. We use numerical examples to illustrate the Bayesian criteria and compare the performance of Bayesian criteria with the frequentist criteria in MHLW guideline. For meta-analysis problem, we consider the case when only summary statistics are available. We use linear combination estimate to combine those statistics and find the optimal weight via minimizing Bayes risk. The closed-form formulas for optimal weight based on the squared error loss and the weighted squared error loss are given. We use several examples to show the asymptotic properties of the optimal weight. Focusing on linear regression problems, we use simulation studies to illustrate the performance of our method.