We consider dose-finding in phase I/II clinical trials by accounting for the preferences of both patients and investigators. We propose to search for the dose that maximizes the difference in the probability of response relative to the probability of toxicity, to reflect the patient's preference to be treated at the best dose. We introduce a penalty term in assessing doses to reflect the principal investigator's preference to examine the efficacy and toxicity of various doses and therefore select the best dose for treating future patients. The combined criterion is called the penalized maximum difference dose (pMDD). We develop a flexible probability model that relaxes the usual monotonicity assumption that the probability of toxicity increases with dose. However, in the proposed dose-assignment rules, we restrict escalation to untried doses if the highest tried dose is toxic. Under the new probability model and the new dose-assignment rules, we obtain improved simulation results compared to currently available methods. In addition, we demonstrate that our new method can be directly applied to two different types of trials: one in which the probability increases with dose and the other in which toxicity does not increase with dose.
Ji, Yuan; Li, Yisheng; and Bekele, Benjamin, "Penalized Maximum Difference Dose - A New Criterion for Phase I/II Dose-Finding Trials" (December 2005). UT MD Anderson Cancer Center Department of Biostatistics Working Paper Series. Working Paper 18.