Lupus & antiphospholipid syndrome
Identification of clinical phenotypes and disease trajectories in systemic lupus erythematosus (SLE) using artificial intelligence (AI) through a natural language processing framework
Authors' conclusion
This novel framework, based on real-world data, enables longitudinal phenotype characterization of SLE patients.
Study context
- Population
- SLE patients from electronic health records (EHRs) with ≥1 hospitalization, ≥1 outpatient visit and follow-up ≥1.5 years; 88% female, median age 43 years, median follow-up 6 yearsof 1,000 extracted
- Participants
- 262 patients
- Exposure
- High- vs medium- vs low-complexity phenotype at first contact
- Outcome
- Clinical phenotypes (clinical domains, disease complexity) and disease trajectories (flares) identified by NLP
Methodological detailsNatural language processing (NLP)-based pipeline, combined AI and human intelligence framework, ontology-based definitions
- NLP-based pipeline
- Combined AI and human intelligence framework
- Ontology-based definitions
Main finding
As reported in the abstract
| Group | Value | Flares over time |
|---|---|---|
| High complexity | 43% | 5 |
| Medium complexity | 35% | 3 |
| Low complexity | 22% | 3 |
Value
High complexity43%
Medium complexity35%
Low complexity22%
Flares over time
High complexity5
Medium complexity3
Low complexity3
- High complexity vs Medium complexity and Low complexity: P < 0.001 — 5 vs 3 and 3
- Clinical domains at first contact: hematological
- 64%
- Clinical domains at first contact: articular
- 47%
- Clinical domains at first contact: cutaneous
- 59%
- Clinical domains at first contact: renal
- 58%
Observational data: associations, not effects.
Further findings
New clinical domains and use of conventional immunosuppressants, biologics and GC during follow-up
- Low and medium-complexity phenotypes showed a higher increase
Original publication
Identification of clinical phenotypes and disease trajectories in SLE using AI through a natural language processing framework. Rheumatology (Oxford). Published 17 February 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →