PetriDish MedSafePre-prescription risk stratification · v1.0Guideline implemented

Catch medication risk before
the prescription reaches the patient.

One auditable report combining genotype, laboratory values, patient factors and concurrent medicines. Deterministic guideline rules — no LLM in the disposition path. Every reason cites its source (CPIC or approved product information); missing critical data blocks the disposition rather than producing false reassurance.

Never outputs
  • "Safe" or "unsafe" verdicts
  • A dose the prescriber can adopt without review
  • A drug–gene relationship not in the encoded rule set
Always outputs
  • Neutral 5-state disposition
  • Every reason cites CPIC or approved product information
  • Missing-critical-data list (blocks disposition when populated)
  • rule_version + trace_id + clinician-review flag
Structured input · JSON
Deterministic rules · sub-second · zero LLM in disposition path
Phase-1 scope

Start narrow. Validate deeply.

Selected NSAIDs
Ibuprofen, naproxen, diclofenac, celecoxib, aspirin. Rules: eGFR, GI-bleed history, concurrent anticoagulant/antiplatelet, heart failure, age × duration, pregnancy trimester, CYP2C9 (ibuprofen, celecoxib).
Codeine · tramadol
CYP2D6 phenotype (CPIC), age contraindications, concurrent CNS depressants, elderly caution.
Paracetamol
Daily-dose ceiling, chronic alcohol / liver disease adjustment, duplicate-ingredient detection across Indian combination products.
Validation state ladder
Guideline implementedRule encoded and unit-tested
Not yet retrospectively or prospectively evaluated in an Indian cohort. The next state — retrospectively evaluated — requires 200–500 de-identified cases + one lab or hospital partner + independent clinician adjudication.
→ PetriDish Autoimmune · AI co-scientist for lab reports→ Start a validation pilot→ Methods & rule versioning