Traditionally, phase I clinical trial designs determine a maximum tolerated dose of an experimental cytotoxic agent based on a fixed schedule, usually one course consisting of multiple administrations, while varying the dose per administration between patients. However, in actual medical practice patients often receive several courses of treatment, and some patients may receive one or more dose reductions due to low-grade (non-dose limiting) toxicity in previous courses. As a result, the overall risk of toxicity for each patient is a function of both the schedule and the dose used at each adminstration. We propose a new paradigm for Phase I clinical trials that allows both the dose per administration and the schedule to vary, making treatment two-dimensional. We provide an outcome-adaptive Bayesian design that simultaneously optimizes both dose and schedule in terms of the overall risk of toxicity, based on time-to-toxicity outcomes. The method is illustrated with a trial of an agent hypothesized to prolong cancer remission after allogeneic bone marrow transplantation, and a simulation study in the context of this trial is presented.
Available at: http://works.bepress.com/peter_thall/1/