Dose Finding Methods in Oncology: From the Maximum Tolerated Dose to the Recommended Phase II Dose

Phase I oncology clinical trials are designed to identify the optimal dose that will be recommended for phase II trials. This dose is typically defined as the dose associated with a certain probability of dose limiting toxicity (DLT) during the first cycle of treatment, although toxicity is repeatedly measured over cycles on an ordinal scale. We present the main dose finding methods developed in the era of cytotoxic agents. We illustrate their properties and limitations in different scenarios. We also explore different implementations of these methods that have been proposed in a Bayesian or likelihood framework or that can rely on several dose-toxicity models. We highlight the fact that the binary nature of the primary outcome (DLT or no DLT) drastically limits the performances of any methods. We then present adaptive dose-finding designs that use toxicity measurements at all cycles of treatment and not only the first one; some authors have proposed to consider the DLT as a time to event variable while others have analyzed the longitudinal measurements of toxic side events. This however raises the delicate issue of the definition of the optimal dose. These approaches are illustrated on two dose finding phase I trials; data are reanalysed and results are compared and discussed. Integration of richer information appears appealing in phase I dose-finding trials, as it gives more accurate estimates of the risk of toxicity and increases the ability of selecting the correct dose. Use of longitudinal data in addition allows for detecting cumulative or delayed effects of strong magnitude. Model-based methods give a flexible framework for using more complete data.