Systemic autoimmune diseases such as lupus, rheumatoid arthritis, vasculitis, systemic sclerosis, idiopathic inflammatory myopathies, Sjogren's disease, and antiphospholipid syndrome may share a short-lived pre-organ-threatening transition state that is visible as coordinated drift across symptoms, labs, and AI uncertainty before the diagnosis-specific activity index changes materially.

Hypothesis: a federated, FHE-preserved clinical score that combines symptom-biomarker discordance, serial change in routine laboratory panels, and calibrated model uncertainty will outperform any single disease-specific index for detecting the onset of this transition state.

Testable predictions:

1. In longitudinal cohorts, the composite score will rise 1-4 encounters before organ-threatening events such as nephritis, vasculitic ischemia, interstitial lung disease acceleration, inflammatory myopathy worsening, or thrombotic APS complications.
2. The score will add incremental discrimination beyond disease-specific indices, with gain persisting after adjustment for treatment changes, baseline severity, and center effects.
3. Sites that exchange only encrypted intermediate representations will retain calibration within a small margin of central training, showing that privacy-preserving deployment does not erase predictive signal.
4. Cases with the largest symptom-biomarker discordance will show the highest false-negative rate for traditional indices, but the lowest net reclassification gain from the new score.

Clinical significance: if validated, the model could support earlier escalation, safer tapering, and cross-site deployment without exposing patient-level data, which is especially relevant for rare vasculitis, myositis, and APS populations.

Limitations: this is a retrospective-to-prospective hypothesis, not a proven rule; performance may degrade under heavy treatment selection bias, sparse sampling, or center-specific ordering patterns; and FHE/federated deployment adds operational complexity that may outweigh benefit in low-volume settings.

LES AI • DeSci Rheumatology