The Randomized CRM: An Approach to Overcoming the Long-Memory Property of the CRM

摘要

The primary object of a phase I clinical trial is to determine the maximumtolerated dose (MTD). Typically, the MTD is identified using a dose-escalationstudy, where initial subjects are treated at the lowest dose level andsubsequent subjects are treated at progressively higher dose levels until theMTD is identified. The continual reassessment method (CRM) is a popularmodel-based dose-escalation design, which utilizes a formal model for therelationship between dose and toxicity to guide dose-finding. Recently, it wasshown that the CRM has a tendency to get "stuck" on a dose-level, with littleescalation or de-escalation in the late stages of the trial, due to thelong-memory property of the CRM. We propose the randomized CRM (rCRM), whichintroduces random escalation and de-escalation into the standard CRMdose-finding algorithm, as an approach to overcoming the long-memory propertyof the CRM. We discuss two approaches to random escalation and de-escalationand compare the operating characteristics of the rCRM to the standard CRM bysimulation. Our simulation results show that the rCRM identifies the true MTDat a similar rate and results in a similar number of DLTs compared to thestandard CRM, while reducing the trial-to-trial variability in the number ofcohorts treated at the true MTD.