BE-BOIN: A Dose Optimization Design Accommodating Backfill and Late-Onset Toxicity

2025-09-14

AI Summary

Traditional dose-finding trials overly rely on the maximum tolerated dose (MTD), neglecting efficacy and late-onset cumulative toxicity. To address this, we propose a novel Bayesian design targeting the optimal biological dose (OBD). Methodologically, we extend the BOIN design by integrating time-to-event toxicity modeling and a dynamic cohort re-assignment mechanism, enabling concurrent multi-dose enrollment, delayed toxicity correction, and quantification of late cumulative toxicity risk. Our key contribution is the first unified framework jointly optimizing OBD identification, late-toxicity modeling, and real-time adaptive cohort re-assignment. Simulation studies demonstrate that the proposed design improves OBD selection accuracy by 22%, reduces median trial duration by 31%, and significantly enhances completeness of toxicity assessment—thereby achieving a more balanced trade-off between safety and efficacy. This paradigm offers a robust, efficient approach to precision dose selection in clinical oncology trials.

Technology Category

• Search and Optimization: Mixed Discrete/Continuous Search
• Reasoning under Uncertainty: Stochastic Optimization
• Machine Learning: Bayesian Learning

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

The US Food and Drug Administration (FDA) launched Project Optimus and issued guidance to reform dose-finding and selection trials, shifting the paradigm from identifying the maximum tolerable dose (MTD) to determining the optimal biological dose (OBD), which optimizes the risk and benefit of treatments. The FDA's guidance emphasizes the importance of collecting sufficient toxicity and efficacy data across multiple doses and considering late-onset cumulative toxicity that often results in tolerability issues. To address these challenges, we propose the BE-BOIN (Backfill time-to-Event Bayesian Optimal INterval) design, which allows backfilling patients into safe and effective doses during dose escalation and accommodates late-onset toxicities. BE-BOIN enables the collection of additional safety and efficacy data to enhance the accuracy and reliability of OBD selection and supports real-time dose decisions for new patients. Our simulation studies show that BE-BOIN accurately identifies the MTD and OBD while significantly reducing trial duration.

Problem

• Accommodating backfill and late-onset toxicity in trials
• Shifting from maximum tolerable to optimal biological dose
• Enhancing accuracy and reliability of dose selection

Innovation

• Backfill patients during dose escalation
• Accommodates late-onset toxicities design
• Enhances OBD selection accuracy