SUMMARY. Ethical constraints and small sample size requirements are the main challenges in designing phase I clinical trials in oncology. Without using randomization, a sound design must be able to locate the maximum tolerated dose by searching the one-dimensional domain of dose levels with relatively little information. After examining the available methodologies in the literature, we propose a new dose-assignment rule and Bayesian design that differ from previous methods in several aspects. The proposed design does not require a dose-toxicity curve and bases dose-assignment decisions only on the toxicity estimates of the current dose at which patients are treated. We propose a new parametrization for the toxicity probabilities, which facilitates prior construction. We establish the asymptotic properties of the proposed design and examine its finite sample properties through extensive simulations.
Ji, Yuan; Li, Yisheng; and Bekele, Benjamin, "A New Dose-Assignment Rule and Bayesian Design in Phase I Oncology Trials" (October 2005). UT MD Anderson Cancer Center Department of Biostatistics Working Paper Series. Working Paper 10.