Simulation study shows safety lead-in phase II designs yield high rates of unsafe recommendations in oncology trials, highlighting the necessity of continuous toxicity monitoring.
Key Points
To evaluate the operating characteristics and safety of phase II safety lead-in designs compared to established dose-finding methods and toxicity monitoring alternatives.
Simulated one to three dose levels comparing Simon two-stage designs with six-patient safety lead-ins against an extended model incorporating toxicity monitoring at each stage.
Compared multi-dose safety lead-in strategies to phase I dose-finding methods (Continual Reassessment Method, Bayesian Optimal Interval, and 3+3) combined with Simon phase II designs, evaluating the effects of toxicity monitoring and patient rollover.
Adding toxicity monitoring to a single-dose Simon safety lead-in trial decreased recommendations of toxic but efficacious doses from 24% to 7%.
For two or three dose levels with acceptable doses, safety lead-in approaches yielded lower overall success than the Continual Reassessment Method (up to 19% lower) and Bayesian Optimal Interval (up to 23% lower), with unsafe recommendations exceeding 30% without phase II toxicity monitoring versus 6% with it.
Algorithm-based approaches (safety lead-in and 3+3) terminated trials incorrectly for toxicity more often when acceptable doses existed, but had up to 36% higher success when all candidate doses exceeded toxicity thresholds.