An Efficient Approach to Design Bayesian Platform Trials

2025-07-16

AI Summary

Evaluating operating characteristics of Bayesian platform trials is computationally prohibitive due to high Monte Carlo simulation costs and challenges in exploring large design spaces. Method: We propose an efficient method that models the joint posterior predictive distribution for multiple endpoints and stages using only two simulation runs—each conducted at a distinct sample size. Contribution/Results: We provide the first rigorous theoretical proof that this two-point strategy enables accurate extrapolation of operating characteristics—including statistical power and type I error rate—across the entire design space. The method supports mid-trial adaptive decisions, dynamic arm addition/deletion, and incorporation of external evidence, while respecting practical constraints and regulatory requirements. Compared with conventional dense-grid simulation, computational overhead is substantially reduced. The approach was successfully deployed in the SSTARLET tuberculosis prevention platform trial, enabling high-frequency validation required for regulatory submissions.

Technology Category

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

Application Category

• Economics, Online Markets and Human Computation: Economics and fairness of platforms and recommendation systems
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Platform trials evaluate multiple experimental treatments against a common control group (and/or against each other), which often reduces the trial duration and sample size. Bayesian platform designs offer several practical advantages, including the flexible addition or removal of experimental arms using posterior probabilities and the incorporation of prior/external information. Regulatory agencies require that the operating characteristics of Bayesian designs are assessed by estimating the sampling distribution of posterior probabilities via Monte Carlo simulation. It is computationally intensive to repeat this simulation process for all design configurations considered, particularly for platform trials with complex interim decision procedures. In this paper, we propose an efficient method to assess operating characteristics and determine sample sizes as well as other design parameters for Bayesian platform trials. We prove theoretical results that allow us to model the joint sampling distribution of posterior probabilities across multiple endpoints and trial stages using simulations conducted at only two sample sizes. This work is motivated by design complexities in the SSTARLET trial, an ongoing Bayesian adaptive platform trial for tuberculosis preventive therapies (ClinicalTrials.gov ID: NCT06498414). Our proposed design method is not only computationally efficient but also capable of accommodating intricate, real-world trial constraints like those encountered in SSTARLET.

Problem

• Efficiently assess Bayesian platform trial operating characteristics
• Determine optimal sample sizes and design parameters
• Accommodate complex interim decisions and real-world constraints

Innovation

• Bayesian platform trials with flexible arm adjustments
• Efficient Monte Carlo simulation for design assessment
• Joint sampling distribution modeling at two sizes