Checkerboard: a Bayesian efficacy and toxicity interval design for phase I/II dose-finding trials

Abstract

ABSTRACT The rise of targeted therapy and immunotherapy has challenged the conventional more-is-better phase I trial design paradigm that focuses on finding the MTD. In this article, we propose a novel model-assisted phase I/II design, called checkerboard design, that considers both toxicity and efficacy. As an extension of the keyboard design, the checkerboard design models the joint distribution of toxicity and efficacy, and divides toxicity and efficacy domain into a series of equal-width intervals or keys. In light of interim data, the checkerboard design continuously updates the posterior distribution of toxicity and efficacy, and adaptively determine the optimal dose for treating the next cohort of patients based on the posterior probability of toxicity and efficacy keys. As a model-assisted design, one important advantage of the checkerboard design is that its decision rule can be pretabulated, greatly simplifying its implementation. We also extend the checkerboard design to handle continuous efficacy endpoint. Simulations study shows that the checkerboard design yields competitive performance comparable to existing model-based phase I/II designs, but is simpler and easier to implement in real applications.