DOI
https://doi.org/10.25772/CQ2F-ZX94
Author ORCID Identifier
0000-0002-9891-0971
Defense Date
2023
Document Type
Dissertation
Degree Name
Doctor of Philosophy
Department
Biostatistics
First Advisor
Dr. Roy T. Sabo
Second Advisor
Dr. Edward Boone
Third Advisor
Dr. Nitai Mukhopadhyay
Fourth Advisor
Dr. Robert A. Perera
Fifth Advisor
Dr. Andrew S. Poklepovic
Abstract
In Phase II clinical trials, Thall and Simon’s Bayesian posterior probability design is commonly implemented to allow for an early termination to determine whether a new treatment warrants further investigation in a larger-scale Phase III trial; this in turn requires a pre-selected prior distribution based on known clinical opinion or historical information. Moreover, this Bayesian approach can result in an issue of inflating type I error rate by monitoring interim data to inform early termination decisions. Alternatively, a Bayesian approach with the decreasingly informative prior (DIP), which is an informative yet skeptical prior, can be implemented to overcome the contentious prior selection process and help reduce erroneous termination in early stages of a trial. In addition, the DIP gradually shifts effective sample size from the skeptical prior to the observed information in the likelihood function as more subjects are accrued, consequently controlling the type I error rate while permitting early termination.
We applied the Bayesian DIP approach to one-parameter and two-parameter models of Phase II clinical trials, and aimed to estimate the smallest sample size needed and stopping decision cutoffs to achieve admissible power and significance levels. For implementing the Bayesian DIP approach to two-parameter models, we extended the expected local-information-ratio approach for single-parameter in multivariate cases to help functionalize the DIP on planned and observed sample sizes. Simulations comparing the performance of the standard Bayesian approach and DIP approach in one-parameter showed that the DIP approach requires fewer patients when admissible designs are achieved; otherwise, the DIP approach controls the type I error and type II error rates with comparable or fewer sample size. Comparisons between the DIP and the standard approach were mixed in two-parameter cases. We also built an R package, BayesDIP, that accommodates the admissible designs for both standard Bayesian approach and DIP approach for one-parameter and two-parameter models of Phase II clinical trials.
Rights
© The Author
Is Part Of
VCU University Archives
Is Part Of
VCU Theses and Dissertations
Date of Submission
7-17-2023