Natural Language Processing for Automated Extraction of Continuous Glucose Monitoring Data
Authors' conclusion
Using an NLP approach to extract valuable glucose data from CGM PDF files is feasible and accurate, which can benefit clinical practice and diabetes research.
Study context
- Population
- Continuous glucose monitoring (CGM) reports stored as PDF files in the electronic health record at New York University Langone Health1% of documents manually reviewed
- Participants
- 226 participants
- Comparison
- Natural language processing (NLP) algorithm vs manual expert review
- Outcome
- Accuracy (correct extraction of CGM data) of the NLP algorithm vs manual review
Methodological detailssingle-centre, retrospective analysis of CGM PDF reports, Two experts independently reviewed 1% of documents manually, optical character recognition (OCR)-based pipeline: OCR of CGM reports, document-type detection by keywords, variable extraction, storage in structured database
- Single-centre
- Retrospective analysis of CGM PDF reports
- Two experts independently reviewed 1% of documents manually
- OCR-based pipeline: OCR of CGM reports, document-type detection by keywords, variable extraction, storage in structured database
Performance
As reported in the abstract
Accuracy of algorithm vs manual review
| Test or model | Accuracy |
|---|---|
| Libre | 99.87% |
| Dexcom | 100.00% |
Accuracy
Libre99.87%
Dexcom100.00%
Further findings
Document type distribution
Inter-expert agreement for Libre results
- Agreement between two experts evaluating Libre results
- 99.93%
Original publication
Natural Language Processing for Automated Extraction of Continuous Glucose Monitoring Data. Diabetes Care. Published 30 October 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →