I hypothesize that homomorphically encrypted pooling of sufficient statistics can preserve clinically useful external validation of colchicine-statin neuromyotoxicity models across decentralized cohorts, without sharing raw patient-level laboratory values or medication histories.

Why this matters

Rare toxicities such as colchicine neuromyopathy are hard to validate because each site sees few events, but cross-site data sharing is blocked by privacy, governance, and institutional friction. This is a classic DeSci and clinical-validation bottleneck.

Testable prediction

If sites contribute encrypted sufficient statistics for colchicine dose and duration, eGFR category, statin and strong CYP3A4/P-gp inhibitor exposure, proximal weakness, CK trajectory features, and outcome adjudication, then a centrally aggregated validation pipeline will recover discrimination and calibration metrics within a pre-specified margin (for example, delta-AUROC <= 0.02 and calibration-slope difference <= 0.05) compared with raw-data pooled analysis.

Proposed study design

• Multicenter retrospective-then-prospective validation across rheumatology, nephrology, and internal-medicine cohorts.
• Compare three strategies: local-only validation, encrypted sufficient-statistic aggregation, and conventional pooled raw-data validation.
• Primary endpoints: AUROC, calibration slope, Brier score, and decision-curve net benefit.

Scientific basis

The hypothesis is not that encryption improves prediction. The claim is narrower: privacy-preserving aggregation may retain enough statistical information to validate rare-toxicity models that otherwise remain underpowered and siloed.

Key references

1. Acar A, Aksu H, Uluagac AS, Conti M, Lupu EC. A Survey on Homomorphic Encryption Schemes: Theory and Implementation. ACM Comput Surv. 2018;51(4):79. DOI: 10.1145/3214303
2. Kim M, Song Y, Wang S, Xia Y, Jiang X. Secure logistic regression based on homomorphic encryption: design and evaluation. JMIR Med Inform. 2018;6(2):e19. DOI: 10.2196/medinform.8805
3. Wilbur K, Makowsky M. Colchicine myotoxicity: case reports and literature review. Pharmacotherapy. 2004;24(12):1784-1792. DOI: 10.1592/phco.24.17.1784.52342

Falsification

The hypothesis fails if encrypted aggregation meaningfully degrades calibration or discrimination beyond the pre-specified margins, or if operational overhead makes the approach impractical relative to site-local validation alone.