Use of a Novel Natural Language Processing Utility to Extract Structured Data From Free-Text Medical Notes

Area of Science:

• Health Informatics
• Natural Language Processing (NLP)
• Clinical Data Management

Background:

• Electronic medical records (EMRs) contain vast research data, but much is unstructured free text.
• Extracting this unstructured data manually is time-consuming and presents a barrier to research.
• Traditional rule-based NLP and newer machine learning models are used for data extraction.

Purpose of the Study:

• To develop and evaluate Note Language Processing (NoteLP), a software utility for extracting research data from free-text clinical notes.
• To compare the efficiency and accuracy of NoteLP against manual data search methods in EMRs.
• To assess the performance of NoteLP in retrieving data for research projects.

Main Methods:

• Developed NoteLP software to convert free-text notes into structured datasets.
• Conducted manual data searches in the Epic EMR to establish a time-based reference standard.
• Designed 22 sample research projects, comparing NoteLP's data extraction time and accuracy against manual methods and ICD-9 codes.

Main Results:

• NoteLP reduced data retrieval time significantly: 37.5 seconds per patient for ICD-9 comparisons and 16.6 seconds for extensive searches, versus 63.4 seconds manually.
• NoteLP achieved a mean sensitivity of 0.98, significantly higher than ICD-9's 0.65.
• NoteLP demonstrated a mean specificity of 0.94, comparable to ICD-9's 0.93.

Conclusions:

• NoteLP is a more efficient tool for extracting research data from EMRs than manual searching.
• NoteLP provides superior sensitivity and equivalent specificity compared to manual coding and ICD-9 codes.
• Rule-based NLP utilities like NoteLP are valuable for efficiently accessing unstructured medical data for research.