Privacy-preserving validation studies in reproductive rheumatology are usually framed as a simple raw-data sharing versus no-sharing problem. I hypothesize that a two-stage validation pipeline using multiparty homomorphic encryption over site-level sufficient statistics will preserve clinically useful random-effects calibration for decentralized teratogenic-exposure models, including between-site heterogeneity, with less than 10 percent error relative to pooled raw-data meta-analysis. Proposed test: simulate and then prospectively evaluate multicenter leflunomide and mycophenolate pregnancy-transition cohorts across sites with different baseline disease severity and documentation quality. Estimate calibration slope, intercept, and tau-squared using raw-data pooling versus encrypted sufficient-statistic aggregation. Primary endpoint: absolute error in calibration slope and heterogeneity variance. This matters because reproductive drug-safety cohorts are small, distributed, and governance-sensitive. References: Sav S et al. Information Fusion. 2024;102004. DOI:10.1016/j.inffus.2023.102004. Kieffer V et al. HEALTHINF 2024. DOI:10.5220/0012753900003767. Chambers CD et al. Arthritis Rheum. 2010;62(5):1494-1503. DOI:10.1002/art.27358.