Clinical Data Mappings

FHIR to OMOP CDM v5.4 — identity and events

Illustrative PERSON and CONDITION_OCCURRENCE rows from synthetic FHIR-like events, with concept ids left blank on purpose.

The problem

You need EHR data in FHIR for exchange and in OMOP for analysis. Official work now includes the FHIR to OMOP Implementation Guide, which exists because sites otherwise build incompatible pipelines. The target described there is the OMOP Common Data Model; confirm the CDM version your project uses. This page uses public OMOP CDM v5.4 table names.

A working example

Official CDM v5.4: PERSON is the identity table (a person has one row; demographics live there). Clinical events such as conditions live in event tables that reference person_id. Vocabularies are resolved to standard concept_id values through the OHDSI standardized vocabularies (Athena), not by inventing integers.

Synthetic, executed in DuckDB 2.0.0 — concept ids deliberately NULL:

Illustrative PERSON

person_id person_source_value gender_source_value gender_concept_id year_of_birth month_of_birth day_of_birth
1 SYN-001 female NULL 1985 3 17
2 SYN-002 male NULL 1979 11 2

Illustrative CONDITION_OCCURRENCE

condition_occurrence_id person_id condition_source_value condition_concept_id condition_start_date condition_start_time
1 1 mild headache after first dose NULL 2026-02-14 15:30:00
2 2 upset stomach NULL 2026-02-15 NULL

The source value is retained. The standard concept is not guessed. Official ETL convention is to map through the vocabularies and keep the source code for reference. A time is stored only when the source actually had a time (subject 2 was date-only).

Drug exposure, measurements, and visits follow the same pattern: one official table, a source value, a mapped concept, and foreign keys back to PERSON (and usually a visit). See the official CDM and the FHIR-OMOP IG for field-level conventions, including what to do when a FHIR resource has no OMOP home.

The gotcha

Inventing concept_id values, or copying ids from a blog, breaks network research. Official OHDSI tooling (Athena, Data Quality Dashboard, and others listed on the CDM v5.4 page) assumes vocabulary integrity.

person_id is a CDM surrogate. person_source_value is where the FHIR Patient.id (or your enterprise MPI) belongs. Collapsing those two, then joining back to FHIR, is a usual identity bug.

A FHIR Condition and a FHIR AdverseEvent are not automatically the same OMOP table. Official IGs and ETL conventions decide. Do not funnel every narrative event into CONDITION_OCCURRENCE because the word “condition” is close enough.

When not to use this

Verification

The synthetic PERSON and CONDITION_OCCURRENCE rows were produced in DuckDB 2.0.0 with concept_id left null. Table roles come from the official OMOP CDM v5.4 specification. The existence and purpose of the FHIR-to-OMOP IG is from that IG’s introduction. No official mapping spreadsheet was copied, and no vocabulary ids were used.

Official docs