| Abstract： |
In phase I trials, the maximum tolerated dose (MTD) is determined under a specific target toxicity probability (TTP) for the dose limiting toxicity (DLT). In fact, the major purpose of phase I trials is to estimate the MTD under a control of an inappropriate dose assignment for patients. Note that the continual reassessment method (CRM)(O'Quigley et al., 1990) is probably the most popular model-based Bayesian dose-finding design. However, the CRM tends to overdose patients. Therefore, Babb et al. (1998) suggested the escalation with overdose control (EWOC) method that, unfortunately, underestimates the MTD. In this talk, a new Bayesian design is introduced for phase I trials where a Bayesian test for the acceptable toxicity, toxicity probability less than the TTP, is adaptively used for dose escalation and the MTD is estimated to be the mode of the posterior distribution of MTD. The Bayesian test-based design is also modified for the late onset DLT that often occurs for anti-cancer treatments. The results of a simulation study investigation the probability of overdose, MTD estimation and required trial time under a variety of dose-toxicity models are finally reported and discussed. |