Abstract

What is the best design for a First in Human trial? The answer to this question depends on the aims and constraints of an individual trial.

First in Human trials are conducted sequentially, allocating a dose to one cohort at a time. After observing results from one cohort, trial teams must decide which dose to give the next cohort. We have developed a framework to make these decisions optimally. The emphasis is on fully specifying the aims of the trial up front: if you tell us what you want the trial to do, we can find the optimal design for your specific trial.

We obtain optimal designs via dynamic programming. This requires a set of calculations to be performed for every possible data set at each stage of the trial. Even with a small sample size this state space is large, and prohibitively large when it comes to considering phase I trials with a safety and an efficacy endpoint. We consider reformulating the state space as the space of posterior density functions for the dose-response model parameter and adapting the dynamic programming algorithm to a sample of this space. This produces a design that is a good approximation to the optimal rule produced by performing dynamic programming on the space of all possible data sets. We use this approximate version of the algorithm to find an optimal design for a First in Human trial with both a binary efficacy endpoint and a binary safety endpoint.

Different sets of aims lead to different optimal rules. This highlights the importance of clearly defining the trial aims up front and choosing a design that meets those aims. This framework enables objective comparison of phase I trial designs to find the optimal design for a specific trial.