Monitoring Adverse Events Through Bayesian Nonparametric Clustering Across Studies

2025-09-08

AI Summary

This study addresses two key limitations in cross-trial adverse event (AE) surveillance: (1) non-robust background rate estimation and (2) oversimplified, single-granularity integration of covariates. To overcome these, we propose a Bayesian nonparametric aggregation monitoring framework. Methodologically, we develop a covariate-dependent product partition model (PPMx) that dynamically clusters trial units by jointly leveraging patient-level and study-level historical data; we further introduce a pairwise similarity metric to enable adaptive grouping, supporting real-time signal detection under blinded conditions and seamless transition to unblinded analysis. Our key contribution is the first integration of PPMx with dynamic clustering for AE background rate modeling—achieving both flexibility and interpretability. Simulation studies and real-world case analyses demonstrate substantial improvements in signal detection performance: +18.3% sensitivity and +12.7% specificity. The framework exhibits strong robustness and practical utility in multicenter, heterogeneous clinical trial settings.

Technology Category

• Multiagent Systems: Adversarial Agents
• Machine Learning: Bayesian Learning
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Large-scale security measurements

Abstract

We introduce a Bayesian nonparametric inference approach for aggregate adverse event (AE) monitoring across studies. The proposed model seamlessly integrates external data from historical trials to define a relevant background rate and accommodates varying levels of covariate granularity (ranging from patient-level details to study-level aggregated summary data). Inference is based on a covariate-dependent product partition model (PPMx). A central element of the model is the ability to group experimental units with similar profiles. We introduce a pairwise similarity measure, with which we set up a random partition of experimental units with comparable covariate profiles, thereby improving the precision of AE rate estimation. Importantly, the proposed framework supports real-time safety monitoring under blinding with a seamless transition to unblinded analyses when indicated. Using one case study and simulation studies, we demonstrate the model's ability to detect safety signals and assess risk under diverse trial scenarios.

Problem

• Monitoring adverse events across multiple clinical studies
• Integrating external historical trial data for background rates
• Supporting real-time safety monitoring with blinded and unblinded analyses

Innovation

• Bayesian nonparametric clustering for AE monitoring
• Integrates external historical data for background rates
• Real-time safety monitoring with blinded transition capability