Three-dose-cohort designs in cancer phase I trials.

Traditional designs for phase I clinical trials assign the same dose to patients in the same cohort. In this paper, we present a new class of designs for cancer phase I trials which initially rapidly escalate by allowing multiple doses (usually 3) to be assigned to each cohort of patients. The class of designs, called the LMH-CRM (an extension of the continual reassessment method (CRM) by administering different percentiles of the maximum tolerated dose (MTD), denoted 'low', 'medium', 'high'), is proven to be consistent and coherent (a commonsense property of phase I trials for dose escalation and de-escalation). Three designs (slow, moderate and fast) are derived based on different dose-escalation restrictions. Simulation results show that moderate and fast LMH-CRM combine the advantages of the CRM with one patient per cohort and three patients per cohort: it accurately estimates the MTD; controls overall toxicity rates; and is time efficient.