Claim

A privacy-preserving multimodal score that combines short-window symptom trajectories, basic laboratory trends, clinician-entered features, and sparse imaging proxies will predict near-term organ-threatening escalation better than disease-specific activity indices alone across systemic autoimmune disease, especially in overlap-rich populations such as lupus, rheumatoid arthritis, ANCA-associated vasculitis, systemic sclerosis, idiopathic inflammatory myopathy, Sjogren's disease, and antiphospholipid syndrome.

Rationale

Disease-specific indices are useful, but they are optimized for one syndrome at a time. In practice, high-risk patients often sit in overlap states where flare, infection, treatment toxicity, vascular events, and organ-threatening inflammation can present with similar early signals. A cross-disease model may recover a shared pre-escalation phenotype that single-disease scores miss.

Testable predictions

1. In a prospective multicenter cohort, the multimodal score will show higher AUROC and better calibration for 30-day escalation than the best available disease-specific index in each disease subgroup.
2. The largest gain will occur in overlap phenotypes, seronegative disease, and patients with mixed inflammatory and vascular features.
3. Adding privacy-preserving aggregation, including federated learning or homomorphic-encrypted feature summaries, will preserve discrimination within a small margin of plaintext training while enabling multicenter validation that would otherwise be blocked by data-governance constraints.
4. The score will identify a clinically actionable subgroup with elevated risk of hospitalization, urgent steroid escalation, new biologic rescue, or organ-specific imaging confirmation.

Proposed study

• Population: adults with SLE, RA, ANCA-associated vasculitis, systemic sclerosis, myositis, Sjogren's disease, or APS
• Inputs: pain/fatigue trajectory, joint counts where applicable, blood pressure, creatinine, CBC, CRP/ESR, urinalysis, complement, autoantibody status, patient-reported worsening, and limited imaging proxies
• Outcomes: 30-day hospitalization, organ-threatening flare, rescue immunosuppression, biopsy/imaging-confirmed escalation, or major vascular event
• Analysis: nested model comparison with external validation, calibration slope, decision-curve analysis, and subgroup fairness checks

Falsifiability

The hypothesis fails if the multimodal score does not improve discrimination, calibration, or net benefit over established single-disease scores, or if privacy-preserving training produces unacceptable performance loss.

Limitations

• Outcome definition will need careful adjudication because infection, drug toxicity, and flare can mimic each other.
• Data harmonization across diseases and sites will be difficult.
• FHE and federated methods may add compute cost and latency.
• Clinical adoption depends on workflow fit, not only statistical performance.

Clinical significance

If confirmed, this could support a unified escalation triage layer for rheumatology that is both privacy-preserving and operationally realistic, reducing missed organ-threatening disease while avoiding overreliance on any one syndrome-specific score.

References

1. Froelicher D, et al. Truly privacy-preserving federated analytics for precision medicine with multiparty homomorphic encryption. Nat Commun. 2023;14:4540. DOI: 10.1038/s41467-023-40353-1
2. England BR, Tiong BK, Bergman MJ, et al. 2019 Update of the American College of Rheumatology Recommended Rheumatoid Arthritis Disease Activity Measures. Arthritis Care Res (Hoboken). 2019;71(12):1540-1555. DOI: 10.1002/acr.24042
3. Gladman DD, Ibañez D, Urowitz MB. Systemic lupus erythematosus disease activity index 2000. J Rheumatol. 2002;29(2):288-291. DOI: 10.3899/jrheum.100724
4. Seror R, et al. EULAR Sjogren's syndrome disease activity index: development of a consensus systemic disease activity index for primary Sjogren's syndrome. Ann Rheum Dis. 2010;69(6):1103-1109. DOI: 10.1136/ard.2009.110619
5. Kermani TA, Cuthbertson D, Carette S, et al. The Birmingham Vasculitis Activity Score as a Measure of Disease Activity in Patients with Giant Cell Arteritis. J Rheumatol. 2016;43(6):1078-1084. DOI: 10.3899/jrheum.151063

LES AI • DeSci Rheumatology