Similarity-based profiling of pharmacovigilance data for drug safety pattern discovery

Area of Science:

• Pharmacovigilance and Drug Safety
• Network Pharmacology
• Computational Toxicology

Background:

• Spontaneous adverse event (AE) reporting systems are crucial for large-scale pharmacovigilance but often analyze drug-event pairs in isolation.
• Existing methods lack a holistic approach to understanding drug relationships based on real-world safety data.

Purpose of the Study:

• To construct a drug-drug similarity network using ingredient-level data from spontaneous AE reports.
• To identify clinically meaningful safety patterns and latent relationships across therapeutic classes.
• To provide a scalable approach for pharmacovigilance and hypothesis generation.

Main Methods:

• Utilized 968,966 AE reports from the Korea Adverse Event Reporting System (KAERS; 2020-2024).
• Identified significant ingredient-event signals using statistical methods (proportional reporting ratio, reporting odds ratio, chi-squared, information component).
• Quantified pairwise ingredient similarity based on shared significant AE terms (MedDRA PTs) and constructed a network using community detection.

Main Results:

• A network of 150 ingredients and 1,267 edges was generated from over 380,000 AE reports.
• Community detection revealed modules corresponding to known pharmacological classes (e.g., antineoplastic agents, contrast media).
• Cross-class clustering (e.g., statins with anti-infectives) suggested shared biological effects beyond primary indications.

Conclusions:

• A signal-based drug similarity network derived from AE reports can capture clinically relevant safety profiles.
• This approach reveals hidden relationships between drugs from different therapeutic classes.
• The network provides a scalable tool for enhancing pharmacovigilance and generating novel research hypotheses.