Keyboard design for phase I drug-combination trials

The Keyboard design is a novel model-assisted dose-finding method to find the maximum tolerated dose (MTD) for single-agent trials. The Keyboard design is easy to implement and has superior performance in comparison to more complicated model-assisted designs. In this article, we extend the Keyboard design to dual-agent dose-finding trials. The proposed Keyboard combination trial design maintains the simplicity of the original single-agent Keyboard design, and its dose escalation and deescalation rules can be pre-tabulated before conducting the trial. We show that the Keyboard combination design has desirable theoretical properties, including the optimality of its decision rules, coherence in dose transition, and convergence to the target dose. Extensive simulations are conducted to evaluate the performance of the proposed Keyboard combination design using a novel, random two-dimensional dose-toxicity scenario generating algorithm. The simulation results confirm the desirable and competitive operating characteristics of the Keyboard design. An R Shiny application is available at www.trialdesign.org to implement the Keyboard combination design.