Keywords: Bivariate normal dynamic linear model; dose-finding study; efficacy and safety; patient population; utility function
Phase II clinical trials are conducted to determine the optimal dose of the study drug for use in Phase III clinical trials while also balancing efficacy and safety. In conducting these trials, it may be important to consider subpopulations of patients grouped by background factors such as drug metabolism and kidney and liver function. Determining the optimal dose, as well as maximizing the effectiveness of the study drug by analyzing patient subpopulations, requires a complex decision-making process. We propose a Bayesian utility approach (BUART) to randomized Phase II clinical trials which uses a first-order bivariate normal dynamic linear model for efficacy and safety in order to determine the optimal dose and study population in a subsequent Phase III clinical trial. We carried out a simulation study under a range of clinical scenarios to evaluate the performance of the proposed method in comparison with a conventional method separately analyzing efficacy and safety in each patient population. We showed the proposed method is more useful for decision making between a phase II dose finding trial and a phase III trial.