Clinician commissioning preview
Medication decision intelligence · release v0.95
Perinatal Evidence Compass
Psychopharmacology guidance for the decision in front of you.
Compare untreated illness, medication exposure, and treatment change across pregnancy, postpartum, and lactation—grounded in a versioned clinical library, never model memory.
- Quick text stays in-browser
- Identifiers prohibited
- No HPI or note paste
- Source-linked synthesis
Clinical decision framePEC · v0.95
Second trimesterBipolar disorderContinuation
1
Untreated illnessRelapse, functioning, suicide, mania, psychosis, prenatal care
2
Current regimenMaternal, obstetric, congenital, neonatal, developmental evidence
3
AlternativesPrior response, comparative evidence, feasibility, patient goals
4
Treatment changeWithdrawal, cross-exposure, destabilization, monitoring burden
One versioned evidence system
The manual became a data product.
Every displayed recommendation must be traceable to the same medication inventory, claim ledger, source map, workbench, appraisal queue, and release controls—not to model memory.
Bounded AI workflow
The model writes the brief. The library sets the limits.
Structure the case
Use the constrained quick-case composer for age, sex, pregnancy stage, and medications, then review the normalized fields. Narrative HPI, note paste, files, and identifiers are prohibited.
Retrieve the release
A deterministic server-side search selects medication records, claims, source links, certainty notes, and known evidence gaps.
Synthesize with boundaries
OpenAI produces a structured clinician brief from that evidence packet only. Unsupported claims and uncited recommendations are rejected.
Review as clinician
The brief surfaces limitations and verification tasks. It never signs orders, replaces assessment, or removes clinician accountability.
Five different questions
A pregnancy “risk rating” is not a clinical plan.
The tool separates the decision being made, because continuing an effective regimen is not the same question as starting, switching, or responding to an inadvertent exposure.
Read the decision framework- 01Continuation
- 02Initiation
- 03Switching
- 04Unplanned exposure
- 05Delivery, postpartum & lactation
Designed to fail closed
Clinical usefulness starts with restraint.
Emergency signals bypass generative guidance. Identifier-like input is rejected before any API call. Missing evidence is rendered as a gap, never filled with a plausible sentence.
No “safe / unsafe” labels
Absolute and relative risks remain contextual and qualified.
No abrupt-stop recommendations
Maternal destabilization, withdrawal, and cross-exposure are explicit.
No silent evidence extrapolation
Class evidence cannot stand in for a new molecule without a warning.
No autonomous medical decision
A licensed clinician must review evidence, patient context, and the final plan.
Internal evaluation
Test the clinical workflow with a de-identified case.
The quick-case box accepts only age, sex, trimester or gestational week, and medication names or doses. Do not enter names, dates of birth, record numbers, contact details, facilities, addresses, quotations, narrative HPI, or note text.