Claim

For lupus nephritis outcome modeling, homomorphically encrypted pooling of sufficient statistics (for example, longitudinal proteinuria, eGFR, complement, and treatment-exposure summaries) will preserve clinically useful external validation performance for response-trajectory models across decentralized centers, with decision-curve net benefit close to conventional pooled analysis and better privacy protection than raw-data sharing.

Why this is plausible

The main bottleneck in lupus nephritis model validation is not only modeling technique but data-sharing friction across hospitals, countries, and governance regimes. Many useful validation tasks rely on sums, cross-products, gradients, or site-level updates rather than full patient-level export. Approximate-number homomorphic encryption makes those operations feasible while preserving confidentiality. If the encrypted workflow retains calibration, discrimination, and net benefit close to pooled validation, it could unlock broader DeSci-style multicenter validation without centralizing raw patient records.

Testable prediction

In a multi-institution lupus nephritis federation:

1. encrypted sufficient-statistic validation will produce AUROC, calibration slope, and Brier score within a small pre-specified margin of conventional pooled analysis,
2. decision-curve analysis will show similar net benefit across clinically relevant treatment-escalation thresholds,
3. site participation and refresh frequency will exceed what is feasible under raw-data transfer agreements alone.

Suggested study design

• Data: center-local longitudinal lupus nephritis datasets with repeated proteinuria and kidney-function measures
• Model: mixed-effects or Bayesian trajectory model for renal response/nonresponse
• Comparator: raw pooled analysis vs homomorphically encrypted sufficient-statistic pooling
• Outcomes: discrimination, calibration, decision-curve net benefit, and operational feasibility metrics
• DeSci extension: federated governance where sites keep raw data locally and share only encrypted aggregates or parameter updates

Falsification

This hypothesis is weakened if encryption-induced approximation materially degrades calibration or if operational overhead eliminates the practical participation advantage.

References

1. Cheon JH, Kim A, Kim M, Song Y. Homomorphic Encryption for Arithmetic of Approximate Numbers. DOI: 10.1007/978-3-319-70694-8_15
2. Moons KGM, Altman DG, Reitsma JB, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD). DOI: 10.1136/bmj.g7594
3. Vickers AJ, Elkin EB. Decision Curve Analysis: A Novel Method for Evaluating Prediction Models. DOI: 10.1177/0272989X06295361