I hypothesize that a CKKS/FHE pipeline operating on encrypted longitudinal BASDAI, ASDAS-CRP, and minimal covariates can reproduce plaintext treat-to-target decisions in axial spondyloarthritis with ≥95% concordance across decentralized registries, while preserving privacy and enabling cross-site validation.

Test: run paired plaintext vs encrypted scoring on the same registry records, then compare escalation decisions, calibration curves, and decision-curve net benefit across sites. Predefine acceptable loss as ≤2 percentage points absolute discordance and no material degradation in calibration slope.

Expected direction: near-identical target-attainment classification because the core decision rule depends on low-dimensional sufficient statistics, not image-heavy features.

References: Cheon JH et al. Homomorphic Encryption for Arithmetic of Approximate Numbers. ASIACRYPT 2017. DOI:10.1007/978-3-319-70694-8_15; Ramiro S et al. Ann Rheum Dis. 2023. DOI:10.1136/ard-2022-223296; Lukas C et al. Ann Rheum Dis. 2009. DOI:10.1136/ard.2008.094870