Date of Award

2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Public Health Sciences

College

College of Graduate Studies

First Advisor

Caitlyn Meinzer

Second Advisor

Viswanathan Ramakrishnan

Third Advisor

Renee' Hebert Martin

Fourth Advisor

Brian Neelon

Fifth Advisor

Byron Gajewski

Sixth Advisor

Lee Schwamm

Abstract

Two opposing paradigms, analyses via frequentist or Bayesian methods, dominate the statistical literature. Most commonly, frequentist approaches have been used to design and analyze clinical trials, though Bayesian techniques are becoming increasingly popular. However, these two paradigms can generate divergent results even in analyses of the same trial data, which may harm the scientific interpretability of the trial. Therefore, it is crucial to harmonize analyses under each approach. In this dissertation, novel unified approaches for one-sided frequentist and Bayesian hypothesis testing problems comparing two proportions in fixed-sample and group-sequential clinical trials are proposed. When a frequentist design with desired type I and II error rates are given, the unification is achieved by deriving specific Bayesian decision thresholds and sample sizes. Similarly, when a Bayesian design is given, the unification is achieved by deriving corresponding frequentist characteristics. In addition, theoretical methods to determine the Bayesian decision threshold, sample size and power are provided. Numerical results show that the unified approach can yield the same type I and II error rates for frequentist and Bayesian hypothesis tests through a numerical study. Further, detailed evaluations suggest that Bayesian priors specifications, allocation ratios, number of analyses can affect the resulting Bayesian sample sizes and decision thresholds. Overall, the unified approach can be adopted into the current clinical trial setting and is helpful to make trial results translatable between frequentist and Bayesian methods.

Rights

All rights reserved. Copyright is held by the author.

Download

Share

COinS