Claim

A time-updated Bayesian model that jointly tracks current mycophenolate exposure, washout interval, contraception reliability, conception intent, and disease stability will predict unsafe conception or early-pregnancy exposure events better than binary compliance checks based only on whether counseling or contraception was documented.

Why this is plausible

Binary compliance logic collapses a dynamic process into a yes/no label. In real autoimmune care, risk changes over time as adherence shifts, conception intent changes, washout progresses, and disease activity forces therapy adjustments. A Bayesian longitudinal model should preserve calibration and discrimination because it updates risk as the clinical state changes rather than freezing it at one visit.

Testable prediction

Across decentralized rheumatology and nephrology registries, a time-updated model will show:

1. higher AUROC and better calibration for unsafe conception exposure,
2. earlier warning before pregnancy-test positivity,
3. lower false reassurance than REMS-style checklist completion alone.

Suggested study design

• Data: visit-level registry data from autoimmune clinics with pregnancy planning documentation
• Outcome: unsafe conception or first-trimester exposure while mycophenolate is active or washout is incomplete
• Comparator: binary documented-compliance rule
• Analysis: hierarchical Bayesian transport model with site-level random effects; external validation across centers
• DeSci extension: privacy-preserving federation using encrypted or summary-level parameter updates so no raw patient-level reproductive data leaves the local site

Falsification

This hypothesis is weakened if the Bayesian model loses calibration on external validation or if checklist completion alone performs equivalently after accounting for site differences.

References

1. Sammaritano LR, et al. 2020 ACR Reproductive Health Guideline. DOI: 10.1002/acr.24130
2. Götestam Skorpen C, et al. EULAR antirheumatic drugs before/during pregnancy. DOI: 10.1136/annrheumdis-2015-208840
3. Perez-Aytes A, et al. Prenatal Exposure to Mycophenolate Mofetil: An Updated Estimate. DOI: 10.1016/S1701-2163(16)34622-9