A Nonparametric Bayesian Local-Global Model for Enhanced Adverse Event Signal Detection in Spontaneous Reporting...

Area of Science:

• Pharmacovigilance
• Biostatistics
• Real-world evidence (RWE)

Background:

• Spontaneous reporting system databases are vital for post-marketing surveillance and real-world evidence (RWE) generation.
• Current adverse event (AE) signal detection methods struggle with rare AEs or new drugs due to low statistical power.
• Existing methods often overlook complex drug-drug associations, limiting signal detection capabilities.

Purpose of the Study:

• To develop novel nonparametric Bayesian models for enhanced AE signal detection.
• To capture complex between-drug associations for principled information sharing.
• To improve statistical power and sensitivity in identifying drug-specific AEs.

Main Methods:

• Utilized local-global mixture Dirichlet process (DP) prior-based nonparametric Bayesian models.
• Developed efficient Markov chain Monte Carlo algorithms for model implementation.
• Employed a false discovery rate (FDR)-controlled, false negative rate (FNR)-optimized hypothesis testing framework.

Main Results:

• Simulations demonstrated superior sensitivity, often exceeding existing methods by twofold or more.
• The proposed methods strictly controlled the false discovery rate (FDR).
• Application to FDA FAERS data confirmed effectiveness in real-world AE signal detection.

Conclusions:

• Novel Bayesian models effectively capture drug-drug associations for improved AE signal detection.
• The methods enhance statistical power, particularly for rare adverse events and newer drugs.
• This approach offers a significant advancement in pharmacovigilance and drug safety surveillance.