An optimal dose-finding technique based on bayesian decision theory for oncology phase I/II clinical trials
Loke, Yee Chong
Date of Issue2008
School of Mechanical and Aerospace Engineering
A phase I dose-finding clinical trial has always been a challenge to find a safe dosage, given a very limited sample size and shorter timeframe. This is particularly important for cancer trials as new experimental drugs could be cytotoxic and life threatening if too high a dose is given or an inaccurate estimated dose level is thought to be safe and continues to be used for subsequent Phase II and III trials. In this research work, a new approach is proposed that help to improve on the current Phase I design in obtaining a better reliability of finding a safe dosage and using fewer patients to obtain the results. The proposed design utilizes a Bayesian decision theory to look at either single event of toxicity or a combined event of both toxicity and efficacy to obtain an optimal dose for a Phase VII clinical trial study. The method assumes binary outcomes for all of these events which give rise to a multinomial likelihood. A decision table is defined with utility values attached to each possible decision. One of the unique attributes of this design is that it has a relationship between these utility values and the target probability associated with a toxicity and efficacy outcomes. By eliciting this target value from the clinical investigator, along with their beliefs on the relative importance of efficacy and toxicity in their trial, appropriate trial specific utility values may be obtained. Due to this interaction between the utility and the target value, the process of finding the optimal dose can be escalated. In addition, the design proposes a stopping criterion based on Bayesian inference to bring the trial to an early completion with a minimum number of patients. The results show promising performance by the proposed design and it has a competitive edge over some of the well-established phase I designs where the comparison is based on simulation run using various clinical scenarios. The design is also tested using actual clinical study data to obtain the optimal dose at the correct dose level using fewer cohorts of patients.