Claim

A homomorphically encrypted federated validation framework using pooled sufficient statistics from CYP2C9- and VKORC1-aware warfarin induction models will preserve clinically useful calibration for antiphospholipid syndrome (APS) anticoagulation cohorts while avoiding transfer of raw genotype data or patient-level INR trajectories.

Rationale

APS remains a warfarin-dominant disease in many high-risk phenotypes, but multicenter validation of genotype-informed dosing and early INR stability models is hard because pharmacogenomic data and longitudinal INR records are privacy-sensitive. If encrypted pooled sufficient statistics can reproduce conventional validation results, DeSci-style collaboration becomes more realistic for rare and high-risk autoimmune thrombosis cohorts.

Testable design

• Participating centers fit a prespecified early-warfarin-control model using CYP2C9, VKORC1, age, interacting drugs, baseline INR, and APS phenotype
• Sites encrypt score vectors or sufficient statistics under a shared homomorphic-encryption scheme
• Central aggregation estimates pooled calibration slope, Brier score, time-in-range prediction error, and external discrimination without raw genotype exchange
• Benchmark encrypted aggregation against trusted-analyst pooled validation

Falsification criterion

The hypothesis is weakened if encrypted aggregation materially degrades calibration or if compute and communication burden prevents feasible multicenter deployment.

Why this matters

If true, decentralized autoimmune pharmacogenomics could validate clinically important warfarin-support tools without compromising patient-level genetic privacy.

References

1. Johnson JA, Caudle KE, Gong L, et al. Clinical Pharmacogenetics Implementation Consortium guideline for pharmacogenetics-guided warfarin dosing: 2017 update. Clin Pharmacol Ther. 2017;102(3):397-404. DOI: 10.1002/cpt.668
2. Froelicher D, Müller P, De Mestral C, 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
3. Pengo V, Denas G, Zoppellaro G, et al. Rivaroxaban vs warfarin in high-risk patients with antiphospholipid syndrome. Blood. 2018;132(13):1365-1371. DOI: 10.1182/blood-2018-04-848333

Claim

A federated validation framework based on homomorphically encrypted pooled survival sufficient statistics can preserve clinically useful discrimination and calibration for denosumab-interruption rebound-fracture models while avoiding transfer of patient-level DXA values, vertebral imaging results, or medication timelines between institutions.

Rationale

Denosumab interruption is a clinically important but relatively uncommon event at single-center scale. External validation therefore needs multicenter pooling, yet bone-health and imaging data are difficult to share across institutions. Survival-model sufficient statistics and calibration summaries may be aggregated under homomorphic encryption with less privacy leakage than raw patient-level exchange.

Testable design

• Sites fit a prespecified rebound-fracture time-to-event model using a shared variable dictionary
• Each site encrypts score vectors or sufficient-statistic summaries under a common homomorphic scheme
• Central aggregation derives pooled hazard-ratio estimates, calibration slope, integrated Brier score, and time-dependent C-index
• Compare encrypted pooled validation with conventional trusted-analyst pooled validation as reference standard

Falsification criterion

The hypothesis is weakened if encrypted aggregation materially degrades calibration or discrimination relative to conventional pooling, or if communication/computation cost makes routine multicenter validation impractical.

Why this matters

If true, this would create a DeSci-ready path for multicenter validation of clinically relevant but privacy-sensitive bone-safety models.

References

1. Tseng E, Wei L, Saeed M, et al. Secure federated survival analysis with homomorphic encryption. AMIA Annu Symp Proc. 2022:997-1006.
2. Froelicher D, Müller P, De Mestral C, 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
3. Cummings SR, Ferrari S, Eastell R, et al. Vertebral fractures after discontinuation of denosumab. J Bone Miner Res. 2018;33(2):190-198. DOI: 10.1002/jbmr.3337

Claim

For rare autoimmune neurovascular complications such as lupus-associated PRES or giant-cell-arteritis visual ischemia, Bayesian dynamic borrowing with prespecified conflict-robust priors will achieve better calibration and narrower uncertainty than fixed historical-control validation, while limiting bias when new cohorts differ materially from prior data.

Why this may be true

Rare-event clinical validation is chronically underpowered. Conventional historical-control approaches either ignore prior information and waste signal, or overuse it and become fragile when case-mix shifts. Dynamic borrowing methods—such as commensurate priors or power priors with robust mixture components—can shrink toward historical evidence when compatible and back off when incompatible.

Testable design

• Assemble decentralized retrospective cohorts of autoimmune neurovascular events across rheumatology, nephrology, neurology, and ophthalmology services
• Split data into historical and prospective validation eras
• Validate one or more transparent clinical tools under three frameworks: no borrowing, fixed borrowing, and conflict-robust dynamic borrowing
• Primary metrics: calibration intercept/slope, Brier score, expected log predictive density, posterior interval width, and false-positive escalation rate under simulated prior-data conflict

Falsifiable prediction

Dynamic borrowing will reduce posterior interval width by at least 15% versus no borrowing while preserving calibration slope between 0.9 and 1.1 in both concordant and conflict simulation scenarios; fixed borrowing will miscalibrate more often under conflict.

References

1. Hobbs BP, Carlin BP, Mandrekar SJ, Sargent DJ. Biometrics. 2011;67(3):1047-1056. DOI: 10.1111/j.1541-0420.2011.01564.x
2. Neuenschwander B, et al. Biometrics. 2016;72(4):1021-1031. DOI: 10.1111/biom.12555
3. Viele K, et al. Pharm Stat. 2014;13(1):41-54. DOI: 10.1002/pst.1589