Development of a confidence-based Natural Language Processing tool to identify inflammatory arthritis from non-inflammatory conditions in Rheumatology medical notes
Authors' conclusion
The model showed high confidence and accuracy, with low risk of missing patients with inflammatory disease.
Study context
- Population
- Anonymized electronic health records (EHRs) from a single Italian rheumatology center
- Participants
- 1,372 anonymized EHRs
- Exposure
- Natural Language Processing (NLP)-based classifier discriminating rheumatoid (RA) and psoriatic arthritis (PsA) vs non-inflammatory rheumatic diseases (OA and FM) from rheumatology notes
- RAn = 468
- PsAn = 732
- Non-inflammatory rheumatic diseases (NIRD)n = 172
- Outcome
- Classification of RA, PsA and NIRD from rheumatology notes
Methodological detailssingle-centre, NLP-based classifier, confidence-based decision system, explainability assessed with LIME and SHAP
- Single-centre
- NLP-based classifier
- Confidence-based decision system
- Explainability assessed with LIME and SHAP
Main finding
As reported in the abstract
Overall accuracy in distinguishing the three diagnostic groups
85%
Cases classified with high certainty
74.5%
Accuracy in high-certainty cases
94.6%
Cases considered uncertain
1.5%
Observational data: associations, not effects.
Further findings
Interpretability (LIME and SHAP)
- Partially meaningful insights, but results inconsistent, especially with complex or negated language
Original publication
Development of a confidence-based Natural Language Processing tool to identify inflammatory arthritis from non-inflammatory conditions in Rheumatology medical notes. Arthritis Res Ther. Published 10 January 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →