A patient-intake company prototyped an ontology: 105,000 conditions, one answer per question

Ask how many patients have diabetes and every system gives a different answer, because each EMR sends the disease under a different code. The fix is an ontology the company owns, built on free standard vocabularies, run as its own service that the product, the warehouse and the AI features all read from.

The problem

One diabetic patient might arrive as E11.9 from one practice, 250.00 from a legacy system, a SNOMED code from another EMR, or as 'diabetes, type 2' typed into a form. Teams cope with spreadsheets that go stale, semantic layers that cannot express that type 2 is a kind of diabetes, or terminology subscriptions that cost six figures a year and keep patient data in a vendor's API.

What we built

Concepts plus is-a relationships, loaded from the free OMOP vocabularies (SNOMED CT, ICD-10-CM, ICD-9-CM). Every incoming code resolves to one concept; queries walk the hierarchy. In a nightly feed of 16 records from three EMRs, 14 mapped automatically and 2 went to review. The service refuses to guess on ambiguous input: an old abbreviation that matches its clinical opposite goes to a person with suggestions attached.

An intake form where patients answer plain-language questions and 'not sure' is a valid answer, stored against the parent concept and still counted. A graph view with live patient counts from the same query the reports use.