Monitoring adverse events through Bayesian nonparametric clustering across studies

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Pharmacovigilance

Background:

• Accurate aggregate adverse event (AE) monitoring is crucial for drug safety.
• Existing methods may lack flexibility in integrating diverse data sources and covariate information.
• Real-time safety monitoring under blinding presents unique analytical challenges.

Purpose of the Study:

• To develop a Bayesian nonparametric inference framework for aggregate AE monitoring across multiple studies.
• To enable the integration of external historical trial data for establishing background AE rates.
• To support real-time, blinded safety surveillance with a pathway to unblinding.

Main Methods:

• A covariate-dependent product partition model (PPMx) for Bayesian nonparametric inference.
• Integration of external data for background rate definition.
• A novel pairwise similarity measure to group experimental units with similar covariate profiles.
• Random partitioning of units based on similarity to enhance AE rate estimation precision.

Main Results:

• The proposed model effectively integrates historical data and handles varying covariate granularity.
• Grouping similar experimental units significantly improves the precision of AE rate estimation.
• The framework demonstrated capability in detecting safety signals and assessing risk in simulated and case study scenarios.
• Successful implementation of real-time, blinded safety monitoring with a seamless transition to unblinded analysis.

Conclusions:

• The Bayesian nonparametric approach offers a robust and flexible method for aggregate AE monitoring.
• The model enhances the precision and reliability of safety signal detection in clinical trials.
• This framework supports adaptive and efficient safety surveillance throughout the trial lifecycle.