About the Compass
The Perinatal Evidence Compass is being built to help licensed clinicians examine psychotropic medication decisions in pregnancy with a consistent reasoning frame, a traceable evidence packet, and explicit uncertainty.
The clinical problem
Perinatal psychopharmacology requires simultaneous attention to psychiatric relapse, maternal functioning, medication effects, alternatives, treatment change, obstetric outcomes, neonatal planning, postpartum transition, and lactation. A single “risk rating” cannot represent that work.
The Compass structures those considerations without pretending that a form can reproduce a complete psychiatric, obstetric, medical, social, or preference assessment.
Intended user
The product is intended for qualified professionals who can independently evaluate the evidence, examine the patient, verify current labels and guidelines, recognize emergencies, and remain accountable for the plan.
What it is designed to do
Use a constrained browser-only quick-case composer for age, sex, pregnancy stage, and medication names or doses, or use controlled fields. Raw quick text is parsed and discarded locally; narrative HPI, note paste, chart upload, and identifiers are prohibited.
Select relevant medication records, claims, source links, limitations, and evidence gaps from the deployed release.
Ask the OpenAI Responses API to organize only the structured case and retrieved evidence into a constrained clinician brief.
Show claim identifiers, sources, release status, uncertainty, and tasks that still require clinician confirmation.
Evidence release v0.95
The current artifact passed structural, package, formula, and visual checks and supports an auditable clinical-data workflow. It remains a commissioning-advancement edition, not a definitive evidence release.
Same-day official-source surveillance identified no clinical amendment at the cutoff. Currency still depends on a governed living-update process; model fluency cannot make an older release current.
What remains open
Licensed database retrieval, dual screening and extraction, formal risk-of-bias and certainty appraisal, exact current-label reconciliation, external specialist sign-off, clinical evaluation, privacy and security review, and operational governance remain release-blocking.
Absolute and relative risks remain conditional and evidence-specific.
Withdrawal, relapse, cross-exposure, and destabilization must be weighed.
A statement without support in the retrieved packet must be withheld.
The clinician independently reviews the evidence and owns the final plan.
Explore the design