Inferring Rheumatoid Arthritis Disease Activity Status From the Electronic Health Records Across Health Systems
Authors' conclusion
RA disease activity can be inferred at scale from within-institution EHR data, though cross-institution performance is limited. The inferred disease activity replicated known associations with MACEs
Study context
- Population
- Patients with rheumatoid arthritis (RA) at Mass General Brigham (MGB) and the Veterans Affairs (VA), EHR data linked with RA registries prospectively collecting DAS281,105 MGB and 2,631 VA
- Comparison
- Machine learning (ML) models with structured EHR data vs structured plus NLP data, within and across institutions
- Outcome
- Performance (AUC) of ML models inferring RA disease activity (registry-collected DAS28) from EHR data
Methodological detailsmulticentre, machine learning, within- and cross-institution validation
- Multicentre
- Machine learning
- Within- and cross-institution validation
Performance
As reported in the abstract
AUC for inferring RA disease activity, within institution
| Test or model | AUC | AUC (MGB) | AUC (VA) |
|---|---|---|---|
| Structured data | 0.68 to 0.70 | — not stated in the abstract | — not stated in the abstract |
| Structured + NLP | — not stated in the abstract | 0.843 | 0.833 |
AUC
Structured data0.68 to 0.70
Structured + NLP— not stated in the abstract
AUC (MGB)
Structured data— not stated in the abstract
Structured + NLP0.843
AUC (VA)
Structured data— not stated in the abstract
Structured + NLP0.833
— not stated in the abstract
Further findings
Cross-institution validation (transportability)
- AUC, MGB→VA
- 0.679
- AUC, VA→MGB
- 0.718
- Limited transportability of algorithms across sites
Association of inferred disease activity with incident MACE (face validity), within institution
- HR, MGB
- 1.12
- HR, VA
- 1.14
- Stratified Cox models
- Significantly associated with increased risk
Original publication
Inferring Rheumatoid Arthritis Disease Activity Status From the Electronic Health Records Across Health Systems. Arthritis Rheumatol. Published 13 April 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →