Integrating rule-based NLP and large language models for statin information extraction from clinical notes

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Informatics
• Natural Language Processing

Background:

• Identifying patient-specific barriers to statin therapy is crucial for cardiovascular care.
• Automating the extraction of these barriers from clinical notes can enhance clinical decision support (CDS).

Purpose of the Study:

• To develop and evaluate a novel hybrid artificial intelligence (AI) framework for extracting statin therapy barriers from clinical notes.
• To accurately and efficiently process large volumes of clinical data.

Main Methods:

• A hybrid AI framework combining rule-based NLP and LLM-based filters and classifiers was developed.
• The framework was trained on 2000 notes and validated on 197,761 notes from 47,192 patients.
• Performance was assessed against manual chart review for statin intolerance, contraindications, and patient deferral.

Main Results:

• The AI framework demonstrated high efficiency, with an initial filter removing over 77% of irrelevant notes.
• High F1 scores were achieved for classifying intolerance (0.99), contraindications (0.81), and patient deferral (0.86).
• The study identified specific percentages of patients experiencing these barriers: 6.4% intolerance, 0.7% contraindications, and 2.9% deferred therapy.

Conclusions:

• The hybrid AI framework offers an efficient, scalable, and reliable method for analyzing clinical notes.
• This AI solution can significantly enhance clinical decision support systems, improve guideline adherence, and reduce provider workload.
• Future work should focus on integrating these AI-driven insights into CDS tools to optimize statin therapy and patient outcomes.