Medical AI & Coding¶
AI assists clinicians at a few points in the hospital workflow — and every suggestion is logged for governance. AI is always advisory; a person makes the clinical decision.
Where AI helps¶
- ICD diagnosis coding — type a free‑text diagnosis (English or Nepali) in a clinical note or insurance claim and get grounded ICD‑11/10 suggestions from your own catalogue. See EMR Diagnosis Coding.
- Claim denial prediction — a pre‑submission risk assessment on insurance claims (see Insurance & Claims).
- Drug interaction checks — background drug–drug / drug–allergy checking in pharmacy.
AI inference log¶
Medical → AI → Inferences (/medical/ai/inferences) is a read‑only governance ledger of every AI
suggestion: the capability (ICD coding / claim‑denial / drug‑interaction), the model used, a
de‑identified input digest, the suggestion, and the clinician's disposition (Accepted / Rejected /
Pending) with timestamp. Filter by capability and action.
Screenshot: AI inference log — TODO
ICD catalogue¶
Medical → Diagnosis Codes (/medical/icd-codes) maintains the per‑tenant ICD catalogue that
grounds the AI — suggestions can only come from codes you've activated.
Advisory, grounded, audited
AI never invents codes (grounded to your catalogue), never decides on its own (clinician confirms), and never sees patient identifiers (only the phrase + candidate codes). The inference log is your audit trail for SAMD/governance review.