Bayesian Adaptive Designs and Versatile Software Platform For Early Phase Clinical Trials

<p>This research is dedicated to improve the efficiency of Bayesian adaptive designs for early phase clinical trials. Phase I clinical trials can be conducted using algorithm-based (also known as rule-based), model-based, or model-assisted designs. Numerous studies have shown that model-assisted designs have the simplicity of algorithm-based designs while possessing the great performance of the model-based designs. Despite the desirable properties of current model-assisted designs, their use is still limited. More importantly, they can fall short of tackling some emerging challenges brought by the development of novel therapies. Thus, there is a pressing need for model-assisted designs that can complement the current designs. In this work, we first clarify some mis-conceptions between algorithm-based designs and the model-assisted designs with the purpose to eliminate the confusion caused by the designs' similarity in their appearance. Second, we develop a class of novel model-assisted designs that aim to accommodate the urgent need to utilize readily available historical data or real-word evidence to further improve the efficiency of the Phase I model-assisted designs. Third, we construct a seamless Phase I/II design that addresses the challenges emerging along with the vast development of immunotherapy and targeted therapy. Fourth, we develop a versatile software platform to provide user-friendly web-based applications to facilitate the use of a series of well-performed model-assisted designs that are built on sound statistical foundations and have superior operating characteristics (e.g., have high probability of identify the MTD and treat a large number of patients on the MTD). In addition to the important issues addressed for Phase I model-assisted designs, we also include the examination of a critical topic in Phase II clinical trials: sequential monitoring. We thoroughly study the connections between different sequential monitoring approaches theoretically. Furthermore, we conduct extensive simulations to examine the impact of different types of prior distributions on the false positive rate and power to test the efficacy of a treatment, and provide practical recommendations for Phase II sequential monitoring. Our research will greatly advance drug development as it not only provides a wide range of innovative designs, but also creates user-friendly versatile software platforms to facilitate the implementation of the novel designs.</p>

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Dissertation (PhD)

Year dc:date.available

2020

Authors dc:creator

• Zhou, Yanhong
• <p>0000-0003-2744-2100</p>

Contributors dc:contributor

• J. Jack Lee
• Ying Yuan
• Xuelin Huang

• Phase I
• Phase II
• model-assisted design
• Bayesian adaptive design
• Sequential monitoring
• Bayes factor
• Nonlocal prior
• Shiny application
• Medicine and Health Sciences

Repository record dc:identifier

https://digitalcommons.library.tmc.edu/utgsbs_dissertations/996

OAI identifier oai:identifier

oai:digitalcommons.library.tmc.edu:utgsbs_dissertations-2046