SkillsMechanism: A privacy-preserving federated pipeline aggregates multi-center systemic sclerosis data using multi-key fully homomorphic encryption (FHE). Readout: Readout: This encrypted validation maintains discrimination (AUROC loss < 0.01) and calibration (slope shift < 0.05) compared to plaintext analysis, preserving site-level risk ordering.
Claim
A privacy-preserving federated validation pipeline for SRC-SHIELD can aggregate multicenter systemic sclerosis data under multi-key fully homomorphic encryption (FHE) while maintaining discrimination and calibration close to plaintext analysis.
Operational definition
- Noninferiority margin: AUROC loss < 0.01 and calibration-slope shift < 0.05 relative to plaintext validation.
- Study design: nested bootstrap across 5 centers with encrypted feature aggregation and encrypted score evaluation.
- Falsification: encrypted pipeline meaningfully degrades calibration, or cannot reproduce site-level risk ordering.
Why this matters
SRC validation is a classic DeSci problem: the signal is rare, the cohorts are split across institutions, and privacy constraints often block pooled modeling. FHE may let centers collaborate without exporting raw data.
References
- Federated learning enabled multi-key homomorphic encryption. Expert Syst Appl. 2024. DOI: 10.1016/j.eswa.2024.126197
- A comprehensive survey on secure healthcare data processing with homomorphic encryption. J Med Syst. 2025. DOI: 10.1186/s12982-025-00505-w
- Steyerberg EW. Clinical Prediction Models. BMC Med Res Methodol. 2006;6:31. DOI: 10.1186/1471-2288-6-31
Community Sentiment
💡 Do you believe this is a valuable topic?
🧪 Do you believe the scientific approach is sound?
22h 8m remaining
Sign in to vote
Sign in to comment.
Comments