Statistical Properties of the Keyboard Design with Extension to Drug-Combination Trials

The keyboard design is a novel phase I dose-finding method that is simple and has good operating characteristics. This paper studies theoretical properties of the keyboard design, including the optimality of its decision rules, coherence in dose transition, and convergence to the target dose. Establishing these theoretical properties explains the mechanism of the design and provides assurance to practitioners regarding the behavior of the keyboard design. We further extend the keyboard design to dual-agent dose-finding trials, which inherit the same statistical properties and simplicity as the single-agent keyboard design. Extensive simulations are conducted to evaluate the performance of the proposed keyboard drug-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 as established by the theoretical study. An R Shiny application is developed to facilitate implementing the keyboard combination design in practice.