SkillsHypothesis: Homomorphically encrypted ECG-lab sufficient statistics can preserve external calibration of hydroxychloroquine QT-safety models across decentralized autoimmune cohorts
Claim A homomorphically encrypted validation pipeline using pooled sufficient statistics from baseline QTc, potassium, magnesium, eGFR, concomitant QT-active drugs, and hydroxychloroquine exposure will preserve clinically useful external calibration of hydroxychloroquine QT-safety models across decentralized autoimmune cohorts without sharing patient-level ECG or laboratory data.
Rationale Hydroxychloroquine is broadly used in lupus and other autoimmune disease, but arrhythmic risk is highly context dependent and many single-center datasets are too small for robust external validation. Multicenter aggregation is difficult because ECG-linked lab data are privacy sensitive. If calibration-relevant summary statistics can be combined under homomorphic encryption, external validation could become feasible without exposing raw traces or patient-level records.
Testable design
- Sites independently fit the same prespecified logistic or survival model for short-horizon QT prolongation or clinically relevant monitoring thresholds.
- Each site exports encrypted sufficient statistics or encrypted gradient/Hessian terms rather than patient-level data.
- Central aggregation reconstructs calibration slope, intercept, Brier score, and AUROC under encrypted or privacy-preserving computation.
- Compare encrypted pooled validation against conventional secure in-house pooled analysis as a reference standard.
- Success criterion: calibration slope/intercept and AUROC differ by less than a clinically negligible margin from raw-data pooled validation.
Falsification The hypothesis fails if encrypted aggregation materially degrades calibration estimates, requires impractical computational overhead, or cannot handle heterogeneous ECG-lab timing across sites.
Why it matters This would offer a concrete path to privacy-preserving pharmacovigilance for cardiovascular drug-safety questions in autoimmune medicine and could generalize to other ECG-linked toxicities.
References
- Jankelson L, Karam G, Becker ML, Chinitz LA, Tsai MC. QT prolongation, torsades de pointes, and sudden death with short courses of chloroquine or hydroxychloroquine. Circulation. 2020;142(1):3-5. DOI: 10.1161/CIRCULATIONAHA.120.047521
- Mercuro NJ, Yen CF, Shim DJ, et al. Risk of QT interval prolongation associated with hydroxychloroquine with or without concomitant azithromycin. JAMA Cardiol. 2020;5(9):1036-1041. DOI: 10.1001/jamacardio.2020.1834
- Froelicher D, Troncoso-Pastoriza JR, Sousa JS, et al. Truly privacy-preserving federated analytics for precision medicine with multiparty homomorphic encryption. Nat Commun. 2021;12:5910. DOI: 10.1038/s41467-021-25972-y
Mechanism: Decentralized hospitals encrypt statistical summaries of patient ECG and lab data using homomorphic encryption, enabling privacy-preserving aggregation at a central server. Readout: Readout: This process allows robust external validation of drug safety models, yielding calibration slope and AUROC scores negligibly different from unencrypted pooled analysis.