Extracting medications and associated adverse drug events using a natural language processing system combining...

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Biomedical Data Science

Background:

• Accurate detection of adverse drug events (ADEs) and medication information from clinical narratives is crucial for healthcare and research.
• Existing clinical NLP systems require enhancement for specific tasks like ADE detection.

Purpose of the Study:

• To develop and evaluate a hybrid clinical NLP system for automated extraction of medical concepts and relations related to ADEs and medications.
• To assess the system's performance in a national challenge focused on adverse drug events and medication extraction.

Main Methods:

• A hybrid approach combining a knowledge-based general clinical NLP system for concept extraction and a deep learning system (attention-based bidirectional long short-term memory networks) for relation identification.
• Utilizing a task-specific deep learning model for enhanced accuracy in identifying relationships between medications and adverse events.

Main Results:

• The hybrid NLP system achieved an F-measure of 0.9442 for relation identification in the 2018 National NLP Clinical Challenges.
• The system ranked fifth in the challenge, demonstrating performance close to the top-ranked system (<2% difference).
• Error analysis was performed to identify areas for future system improvement.

Conclusions:

• The developed hybrid NLP system effectively extracts ADE and medication-related information from clinical narratives.
• This approach demonstrates the successful integration of general clinical NLP with deep learning for task-specific applications.
• The system holds potential for real-world applications in supporting ADE research and clinical decision-making.