Dose finding when the target dose is on a plateau of a dose-response curve: comparison of fully sequential designs

Consider the problem of estimating a dose with a certain response rate. Many multistage dose-finding designs for this problem were originally developed for oncology studies where the mean dose-response is strictly increasing in dose. In non-oncology phase II dose-finding studies, the dose-response curve often plateaus in the range of interest, and there are several doses with the mean response equal to the target. In this case, it is usually of interest to find the lowest of these doses because higher doses might have higher adverse event rates. It is often desirable to compare the response rate at the estimated target dose with a placebo and/or active control. We investigate which of the several known dose-finding methods developed for oncology phase I trials is the most suitable when the dose-response curve plateaus. Some of the designs tend to spread the allocation among the doses on the plateau. Others, such as the continual reassessment method and the t-statistic design, concentrate allocation at one of the doses with the t-statistic design selecting the lowest dose on the plateau more frequently.