Skip to main content
Status fields are surfaced today on responses from /v1/medications/infer (the FHIR endpoint). The PDF endpoint (/v1/med-lists/from-documents) doesn’t expose status yet — status inference for document-derived medications is in progress.
Inferred status follows FHIR’s MedicationRequest.status vocabulary. Five possible values, each with deterministic triggers.

​
The table

​
Precedence (when multiple rules fire)

If both active_order and order_stopped fire on resources that resolve to the same medication, active wins — the current state trumps historical state.

​
What raises and lowers confidence

  • Corroboration — multiple active-leaning signals firing for the same medication raise confidence; corroborating evidence is the strongest indicator of a confident classification.
  • Missing date — lowers confidence (the engine can’t reason about recency)
  • Missing RxNorm code — lowers confidence (less certainty that we deduplicated correctly)
  • Staleness — older signals carry less weight than recent ones

​
Filtering by status

Pass statuses in the request body to filter the response post-inference:
The engine still runs on all resources; the filter only trims the output list. Good for “show me just active meds” without re-invoking.

​
Special cases

If a MedicationStatement has reported=true (patient-reported, not clinician-recorded), the self_reported_active rule fires with a lower base confidence. Clinician-recorded statements carry more weight.
Grouping still works on normalized display names, but confidence is lower since we can’t confirm we’re looking at the right drug across sources.
If a MedicationRequest is status=active but the order is old and no MedicationDispense is on file, prescription_expired fires and status becomes completed. The order was never actually used.