MedXN: an open source medication extraction and normalization tool for clinical text.

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Pharmacovigilance

Background:

• Accurate extraction and normalization of medication information from clinical notes are crucial for pharmacovigilance and clinical research.
• Existing methods often struggle with the complexity and variability of medication descriptions in unstructured text.

Purpose of the Study:

• To develop and evaluate the Medication Extraction and Normalization (MedXN) system for comprehensive medication data extraction.
• To normalize extracted medication information to the most specific RxNorm concept unique identifier (RxCUI).

Main Methods:

• Decomposition of clinical medication descriptions into names and attributes.
• Utilized RxNorm dictionary lookup and regular expressions for extraction.
• Employed hierarchical processing, synonym expansion, and inference rules for normalization.

Main Results:

• Achieved high F-measures: 0.975 for medication name and >0.90 for attributes.
• RxCUI assignment yielded F-measures of 0.932 for names and 0.864 for full medication information.
• Identified human assumption in gold standard as a source of false negatives.

Conclusions:

• The MedXN system demonstrates high accuracy in extracting and normalizing medication information to RxCUI.
• Explicit evidence is key for accurate normalization; further inference rules can improve handling of incomplete descriptions.