Bayesian Methods in Phase II Dose-Finding Trials

Abstract

Characterising the dose-response relationship and finding the right dose are important
but challenging tasks in the pharmaceutical drug development process. The Multiple
Comparison Procedure - Modelling (MCP-Mod) approach is an efficient statistical method
for the analysis of Phase II dose-finding trials. We propose novel Bayesian statistical
methods to address the limitation of MCP-Mod, particularly in incorporating historical
trial data. We aim to improve the accuracy and efficiency of dose selection, thereby
reducing the risk of failure in Phase III trials.
One focus of this research deals with the difficulty of incorporating historical trial
data due to the frequentist nature of MCP-Mod. We develop a Bayesian hierarchical
framework for dynamically incorporating historical trial data, allowing the amount of
borrowed information to be automatically adjusted based on the similarity between the
results of the current trial and those of historical trials. When trials exhibit a high
degree of homogeneity in treatment effects, more information is borrowed to improve
statistical power. When heterogeneity is observed, less information is borrowed to control
the type I error rate. This model is particularly useful when the effect sizes of two trials
are different.
In addition, to address the challenge of MCP-Mod that requires the pre-specification
of candidate models, we introduce the MAP-curvature method, which is a model-free
Bayesian approach. It incorporates the total curvature of the dose-response curve as a
prior parameter, avoiding the requirement for a set of pre-specified candidate models.
This method is further refined by utilising the sigmoid Emax model as the default
response function. These methods provide a flexible and robust framework for detecting
dose-response signals, estimating dose-response curves, and determining the optimal
target dose.

Date of Award	10 Dec 2024
Original language	English
Awarding Institution	University of Bristol
Supervisor	Peter Green (Supervisor) & Feng Yu (Supervisor)