Semantic processing to identify adverse drug event information from black box warnings

Area of Science:

• Pharmacovigilance
• Medical Informatics
• Natural Language Processing

Background:

• Adverse drug events (ADEs) cause millions of injuries and deaths annually.
• Limited availability of comprehensive, current, and free drug information hinders safety.
• Clinical decision support systems (CDSS) can reduce ADEs but require structured data.

Purpose of the Study:

• To develop and evaluate a semantic natural language processing (NLP) approach.
• To extract key safety information from FDA drug labels.
• To assess the feasibility of supporting CDSS with extracted data.

Main Methods:

• Utilized a semantic NLP approach to process drug warning labels from the DailyMed website.
• Extracted data on adverse drug events, at-risk conditions, and susceptible populations.
• Quantified extraction performance using precision, recall, and F-score metrics.

Main Results:

• Achieved high precision (90%) and moderate recall (51%) overall.
• Specific performance metrics: 94% precision, 52% recall for ADEs; 80% precision, 53% recall for conditions; 95% precision, 44% recall for populations.
• Demonstrated successful extraction of structured safety information.

Conclusions:

• The NLP approach effectively extracts critical drug safety information from FDA labels.
• Extracted data can be structured to support clinical decision support systems.
• This method offers a pathway to improve drug safety monitoring and patient care.