Evaluation of clinical trial designs for the development of targeted agents
With the increasing need for targeted agents in cancer therapies, efficient clinical trials for the development of targeted agents are more and more important. Proper study design and statistical analysis play a crucial role in the success of clinical trials for the development of targeted agents. In this study, we evaluate the Bayesian adaptive randomization design (BAR) and the equally randomized (ER) marker stratified design, for the development of targeted agents. This study contains two parts. In the first part, we evaluate and decide the appropriate analysis approach for design evaluation in the second part. To achieve this goal, we simulate one-arm phase II clinical trials for the screening of targeted agents. We compare the frequentist score test and Bayesian analysis approaches with two different decision rules, i.e. controlling the type I error and minimizing the loss. Type I error, power and loss are computed to make the comparison. The Bayesian analysis approach controlling the type I error provides the same results as the frequentist score test. However, under the framework of drug development, the ultimate goal is to minimize the loss, not to control the type I error. Upon the proper specification of the utility functions of rejecting a good drug and accepting a bad drug, the Bayesian analysis approach can minimize the overall loss. In contrast to the frequentist framework, the Bayesian paradigm provides more flexibility in identifying the optimal treatment, for the inference is free of the preset sampling scheme. Thus, we use the two Bayesian analysis approaches for the design evaluation in the second part. In addition, we evaluate the equally randomized marker stratified design and the Bayesian adaptive randomization design for the testing of targeted agents. ER and BAR are simulated in two-arm phase II clinical trials with one marker. Type I error, power and loss are computed to make the comparison. By real time learning of the accumulated information, BAR is more flexible and efficient for the development of targeted agents, resulting in higher overall response rate, and shows comparable type I error, power and loss to ER. Thus BAR is suitable for the development of multiple targeted agents with multiple biomarkers. However, it takes more effort on the trial design, infrastructure setting up, trial analysis and reporting. To resemble what happens in the real world, one important new feature of this study is that the response rate of the drugs follows a statistical distribution rather than being a fixed value when we simulated the data. This better mimics the uncertainty and variability of the drugs, and provides us more accurate assessment of the performance of different clinical trial designs. Consequently, by improving the statistical properties of the clinical trial design, our study can benefit more patients participating in the clinical trial studies, and has important implications for improving public health in the general population.
Zhao, Yang, "Evaluation of clinical trial designs for the development of targeted agents" (2013). Texas Medical Center Dissertations (via ProQuest). AAI1549847.