Shallow medication extraction from hospital patient records

Area of Science:

• Medical Informatics
• Natural Language Processing
• Computational Linguistics

Background:

• Hospital patient records (PRs) contain crucial medication information in free text zones.
• Extracting this data automatically is challenging due to language complexities and variations.
• The Patient Safety through Intelligent Procedures in medication (PSIP) project aims to improve medication safety.

Purpose of the Study:

• To develop and evaluate methods for shallow Information Extraction (IE) of medication details from Bulgarian patient records.
• To automatically extract drug names, dosage, modes, and frequency of administration.
• To assign the appropriate ATC (Anatomical Therapeutic Chemical) code to each extracted medication event.

Main Methods:

• Utilized rule-based text analysis modules for symbolic computations in information extraction.
• Developed components to handle negative statements, elliptical constructions, conjunctive phrases, and temporal inferences.
• Implemented an algorithm to assign drug ATC codes to extracted medication events.
• Trained and tested the system on a substantial corpus of Bulgarian patient records (1,300 training, 6,200 test).

Main Results:

• Achieved high extraction accuracy for drug names (f-score: 98.42%).
• Demonstrated strong performance in extracting dosage information (f-score: 93.85%).
• Successfully assigned ATC codes to extracted medication events, enhancing data standardization.

Conclusions:

• The developed shallow Information Extraction system is effective for processing Bulgarian patient records.
• The methods employed successfully address linguistic complexities in medication data extraction.
• The system contributes to improved patient safety by enabling automated extraction and coding of medication information.