Phase I Clinical Trial Design Optimization

Abstract

Phase I oncology clinical trials face the dual imperative of identifying the maximum tolerated dose (MTD) or recommended phase II dose (RP2D) while minimising patient exposure to sub-therapeutic or excessively toxic doses. This study conducted a comprehensive simulation-based evaluation of six dose escalation designs -- traditional 3+3, modified Fibonacci 3+3, continual reassessment method (CRM), escalation with overdose control (EWOC), Bayesian optimal interval (BOIN), and i3+3 -- across 2,880 simulation scenarios spanning varying true MTD positions, dose-toxicity relationship steepness, sample size constraints, and dose spacing parameters. Each design was evaluated on five metrics: MTD selection accuracy, percentage of patients treated at or above the MTD (overdosing), percentage treated below the therapeutic range (underdosing), operating characteristics under model misspecification, and trial duration. BOIN achieved the highest MTD selection accuracy (76.4% correct selection rate across all scenarios) while maintaining an overdosing rate of 14.8% -- significantly outperforming 3+3 (correct selection 51.2%; overdosing 9.6%) and CRM (correct selection 74.1%; overdosing 21.4%). The i3+3 design balanced simplicity with performance (correct selection 68.4%, overdosing 13.2%), offering a practical alternative requiring no model-fitting software. Under model misspecification scenarios (true dose-toxicity relationship deviating from assumed logistic), BOIN showed the smallest performance degradation (delta accuracy -4.8%) compared to CRM (delta -18.4%), demonstrating superior robustness. These simulation results provide a quantitative evidence base for Phase I design selection and establish BOIN as the preferred model-assisted design for oncology dose escalation when regulatory and patient safety are co-primary objectives.