Extracting and standardizing medication information in clinical text - the MedEx-UIMA system

Area of Science:

• Natural Language Processing
• Biomedical Informatics
• Health Informatics

Background:

• Medication information extraction from electronic health records (EHRs) is crucial for research.
• Existing systems often lack standardized representation for extracted medication data.

Purpose of the Study:

• To introduce an open-source Java implementation of MedEx (MedEx-UIMA) for medication information extraction.
• To develop encoding modules for mapping extracted drug information to RxNorm concepts and normalizing frequency data to ISO standards.

Main Methods:

• Developed MedEx-UIMA based on the Unstructured Information Management Architecture (UIMA) framework.
• Implemented new encoding modules for mapping drug names, doses, and forms to RxNorm concepts.
• Translated drug frequency information to ISO standard.
• Processed 826 documents and validated performance using manually annotated test sets.

Main Results:

• MedEx-UIMA demonstrated high performance, achieving F-measures of 98.5% and 97.5% for mapping drug names to RxNorm generic ingredients.
• Achieved F-measures of 85.4% and 88.1% for mapping drug names/dose/form to specific RxNorm concepts.
• Normalized frequency information to ISO standard with an F-measure of 90.4%.

Conclusions:

• The open-source MedEx-UIMA system effectively extracts and standardizes medication information from clinical text.
• This system enhances the usability of EHR data for research by providing standardized medication representations.