Abstract:
|
In Phase I cancer trials, the target is to identify the maximum-tolerated dose (MTD), which is defined as the probability of dose-limiting toxicities (DLT) is closest to and lower than a targeted probability. The most widely used methods include the 3+3 design and Bayesian model-based designs. The 3+3 design has been widely used over the years because it's simple to be understood and implemented. The Bayesian model-based approach has gradually drawn more attention because it efficiently uses all observed data and also can include prior information. The objective of this work is to investigate the impact of mis-specified prior on Bayesian model-based design, which is motivated by the fact that in many clinical trials, the prior information is available only among animal studies. The species difference between animal and human could lead to a mis-specified prior. A comparison between 3+3 and Bayesian model-based dose finding designs will be provided in terms of operation characteristics. We will propose some guidelines in the choices of designs and priors based on the likelihood that the prior clinical knowledge would be consistent to the current dose finding trial.
|