Application of Bayesian methodology in radiopharmaceutical therapies dose optimization.

Abstract

e15152 Background: Radiopharmaceutical therapies (RPTs) exhibit complex dose-response relationships due to inter-patient heterogeneity in tumor uptake, dosimetry, and radiation exposure. Recent FDA guidance on RPT dose optimization highlights the importance of randomized dose optimization, encourages modeling and simulation, and supports dose-response evaluation with an overall benefit-risk focus. Bayesian trial designs enable adaptive decision-making, incorporate uncertainty, and integrate historical and external data. These approaches are discussed in recent FDA guidance on Bayesian methodology and are increasingly used in oncology dose optimization, making them particularly well suited for RPT development. Methods: We applied Bayesian methods across early and late-phase oncology development: Bayesian Optimal Interval for Phase I/II Trial Design (BOIN12) was evaluated to inform dose-escalation decisions by jointly incorporating safety and preliminary efficacy information. Bayesian continuous efficacy monitoring using posterior and predictive probabilities was implemented in dose expansion or seamless phase II/III trials (NCT06726161, NCT06590857, NCT05595460, NCT07165132). Bayesian response-adaptive randomization (RAR) was assessed in phase II or III RPT dose-optimization trials, in which multiple dose regimens were initially equally randomized, with interim adaptation based on emerging efficacy and toxicity. Operating characteristics were characterized via simulations under clinically relevant dose-efficacy and dose-toxicity scenarios representative of RPT programs. Results: Simulations demonstrated that Bayesian designs reduced sample size and trial duration while enabling efficient adaptive decision-making. For phase I dose escalation, BOIN12 designs showed a higher probability of selecting doses with acceptable safety profiles compared with traditional rule-based approaches. For dose-expansion and seamless phase II/III settings, Bayesian continuous monitoring supported timely interim decision-making while preserving operating characteristics. For phase II or III dose-optimization trials, Bayesian RAR enriched enrollment to regimens with superior therapeutic indices and preserved type I error and power under clinically meaningful dose-response scenarios. Conclusions: Bayesian trial designs, including model-based dose escalation, continuous monitoring, and RAR, provide a rigorous and FDA-aligned framework consistent with recent FDA guidance on dose optimization in RPT development and Bayesian methodology. Simulation and modeling play a critical role in evaluating operating characteristics, informing design choices, and supporting dose selection under clinically relevant scenarios. These methods enhance ethical efficiency, benefit-risk characterization, and evidence-based dose optimization.