A Bayesian Treatment Selection Design for Phase II Randomised Cancer Clinical Trials

2025-05-14

AI Summary

In phase II oncology trials, the lack of a clearly defined primary endpoint and the difficulty in balancing treatment selection efficiency with practical feasibility hinder optimal decision-making. Method: This paper proposes a Bayesian adaptive treatment selection design based on posterior interval probabilities. It introduces a novel posterior interval decision criterion grounded in joint-distribution integration—overcoming the limitations of conventional single-point thresholds—and develops two Bayesian frameworks for adaptive sample-size determination. An interactive R Shiny application is also provided to support real-time clinical decision-making. Contribution/Results: Simulation studies and empirical validation demonstrate that the design significantly improves accuracy in identifying the optimal treatment and enhances decision transparency under small-sample settings. It effectively addresses the inflexibility of frequentist approaches and the rigidity of fixed thresholds, offering a new paradigm for early-phase oncology trials that reconciles statistical rigor with practical implementability.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Reasoning under Uncertainty: Stochastic Optimization
• Machine Learning: Bayesian Learning

Application Category

• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

It is crucial to design Phase II cancer clinical trials that balance the efficiency of treatment selection with clinical practicality. Sargent and Goldberg proposed a frequentist design that allow decision-making even when the primary endpoint is ambiguous. However, frequentist approaches rely on fixed thresholds and long-run frequency properties, which can limit flexibility in practical applications. In contrast, the Bayesian decision rule, based on posterior probabilities, enables transparent decision-making by incorporating prior knowledge and updating beliefs with new data, addressing some of the inherent limitations of frequentist designs. In this study, we propose a novel Bayesian design, allowing selection of a best-performing treatment. Specifically, concerning phase II clinical trials with a binary outcome, our decision rule employs posterior interval probability by integrating the joint distribution over all values, for which the 'success rate' of the bester-performing treatment is greater than that of the other(s). This design can then determine which a treatment should proceed to the next phase, given predefined decision thresholds. Furthermore, we propose two sample size determination methods to empower such treatment selection designs implemented in a Bayesian framework. Through simulation studies and real-data applications, we demonstrate how this approach can overcome challenges related to sample size constraints in randomised trials. In addition, we present a user-friendly R Shiny application, enabling clinicians to Bayesian designs. Both our methodology and the software application can advance the design and analysis of clinical trials for evaluating cancer treatments.

Problem

• Balancing efficiency and practicality in Phase II cancer trials
• Overcoming limitations of frequentist designs with Bayesian methods
• Enabling transparent treatment selection using posterior probabilities

Innovation

• Bayesian design for treatment selection
• Posterior interval probability decision rule
• User-friendly R Shiny application