A Continual Reassessment Method With Cohort Size Adaptation Based on Bayesian Posterior Probabilities in Phase I Dose-Finding Studies

Abstract

In phase I cancer studies, the maximum tolerated dose (MTD) is estimated by gradually increasing dose levels while accumulating safety information. Recently, Bayesian dose-finding methods such as the continual reassessment method (CRM) have gained popularity. Due to the lack of safety information, phase I studies on new drugs must start at doses low enough that efficacy is not expected but safety is certain up to an acceptable level. To reach the MTD with fewer patients, a 2-stage method has been proposed that enrolls only a single patient at each dose level until the first dose-limiting toxicity is observed. If the study drug is less toxic, it may require many cohorts to complete the study and thus may lead to a longer study period. In this paper, the authors propose a new CRM with cohort size adaptation to reduce the number of cohorts without reducing the accuracy of MTD selection. The cohort size is determined based on the Bayesian posterior probabilities computed during a study. Simulation studies show that the proposed method reduced the number of cohorts compared with the 2-stage method while still yielding a comparable probability of selecting the MTD correctly.