Bayesian Phase 1-2 Designs with Adaptive Rules for Staggering Patient Entry

For first-in-human dose-finding trials, to protect patient safety, regulatory agencies may enforce strict within-cohort staggering rules that require delaying treatment of each patient in the first cohort at an untried dose until dose-limiting toxicities (DLTs) of all previously treated patients have been evaluated. Consequently, many new patients may face therapy delays, which reduces their probability of achieving a response due to disease progression, or be treated off-protocol, which may significantly extend trial duration. To address this, we propose a Bayesian phase 1-2 design, Adaptive Stagger, that reduces delays while protecting patients by making adaptive within-cohort staggering decisions. Adaptive Stagger exploits the relationship between the number of low-grade toxicities and DLT, and accounts for the risk of disease progression. A utility function is used to quantify the tradeoff between DLT and response, with a patient’s treatment delayed only if it has greater expected utility than immediate treatment at the current recommended dose, or the current dose fails a safety requirement. Otherwise, the patient is treated without delay at the current dose. Simulations show that, compared to a design with strict within-cohort staggering rules, Adaptive Stagger improves safety slightly, increases the optimal dose selection rate, and substantially shortens trial duration. The design is illustrated by a trial of CD70 natural killer cells for treating hematologic malignancies.