Hypothesis: CKKS-style homomorphic encryption of sufficient statistics will allow multicenter external validation of MRONJ prediction models with calibration and discrimination comparable to centralized pooled analysis, without sharing raw dental records or patient-level trajectories.

Why this matters:

• MRONJ is uncommon, so robust validation needs pooled multicenter data.
• Dental findings, medication histories, and systemic immunosuppression data are privacy-sensitive and often institutionally siloed.
• If encrypted sufficient-statistic pooling preserves calibration, decentralized validation becomes feasible for rare adverse-event models in rheumatology and osteoporosis care.

Testable design:

• Participating registries compute local summary statistics for predefined predictors: antiresorptive class, duration, extraction timing, periodontal disease, diabetes control, steroid exposure, smoking, prior exposed bone, and incident MRONJ.
• Sites encrypt sufficient statistics under CKKS and send only ciphertexts to an aggregator.
• Compare centralized logistic/Cox model performance vs encrypted pooled-fit performance on held-out external sites.
• Primary metrics: calibration intercept, calibration slope, AUROC/C-index, and Brier score.

Falsifiable prediction:

• Encrypted pooled validation will preserve AUROC within 0.02 and calibration slope within 0.05 of the non-encrypted pooled reference.

Key references:

1. Cheon JH, Kim A, Kim M, Song Y. Asiacrypt 2017. doi:10.1007/978-3-319-70694-8_15
2. Ruggiero SL, et al. J Oral Maxillofac Surg. 2022;80:920-943. doi:10.1016/j.joms.2022.02.008
3. Yarom N, et al. J Clin Oncol. 2019;37:2270-2290. doi:10.1200/JCO.19.01186