medExtractR

medExtractR extracts medication dose and timing information from clinical notes in electronic health records to produce medication-specific datasets for pharmacological research.

Key Features:

• Implementation: Implemented in the R programming language.
• Extraction approach: Uses lexicon dictionaries and regular expression patterns to identify drug entities, including dose and timing, from clinical text.
• Targeted extraction: Focuses on extracting information for specified medications rather than all medication mentions in a note.
• Development dataset: Developed and refined using clinical notes from Vanderbilt University's Synthetic Derivative.
• Comparative evaluation: Evaluated against MedEx, MedXN, and CLAMP.
• Performance metrics: Demonstrated F-measures exceeding 0.95 overall and outperformed existing systems in most entity-level extraction tasks, with lower accuracy noted for dose amount extraction in lamotrigine and allopurinol.
• Tested medications: Initially tested on tacrolimus and lamotrigine and assessed for generalizability using allopurinol.

Scientific Applications:

• Medication-specific dataset generation: Produces high-quality, medication-focused datasets for secondary analysis of EHR data.
• Pharmacological studies: Enables analyses of dosing patterns, timing, and adherence in pharmacokinetic and pharmacoepidemiologic research.
• Entity-level medication extraction evaluation: Serves as a benchmark for entity-level extraction of dose and timing compared to other NLP systems.

Methodology:

Implemented in R and employs lexicon dictionaries and regular expression patterns to identify drug entities (dose and timing) in clinical notes; developed using notes from Vanderbilt University's Synthetic Derivative.