The experimental design presented here is motivated by a phase II clinical trial called PePS2, investigating the efficacy and safety of an immunotherapy called pembrolizumab in a specific subgroup of lung cancer patients. Previous trials have shown that the probability of efficacy is correlated with particular patient variables. There are clinical trial designs that investigate co-primary efficacy and toxicity outcomes in phase II, but few that incorporate covariates. We present here the approach we developed for PePS2, latterly recognised to be a special case of a more general method originally presented by Thall, Nguyen and Estey. Their method incorporates covariates to conduct a dose-finding study but has been scarcely used in trials. Dose-finding is not required in PePS2 because a candidate dose has been widely tested. Starting from the most general case, we introduce our method as a novel refinement appropriate for use in phase II, and evaluate it using a simulation study. Our method shares information across patient cohorts. Simulations show it is more efficient than analysing the cohorts separately. Using the design in PePS2 with 60 patients to test the treatment in six cohorts determined by our baseline covariates, we can expect error rates typical of those used in phase II trials. However, we demonstrate that care must be taken when specifying the models for efficacy and toxicity because more complex models require greater sample sizes for acceptable simulated performance.
Getting this paper published was a big deal for me because it was the first design idea I had had that was completely my own. Well, so I thought. An erudite peer-reviewer pointed out that the method (which I had named Bayesian Evaluation of Bivariate Outcomes and Predictive variables, or BEBOP) was actually a special case of a design by Thall, Nguyen & Estey. In response, I re-wrote the paper presenting the design as a novel refinement of their method suitable for phase II trials, called P2TNE for Phase II Thall Nguyen & Estey. Anyway, after being rejected twice from big journals, it was a big relief to get the paper published in the BAYSM2018 conference proceedings. It was especially classy of Springer to give a complementary hard cover copy of the book to each contributing author.