OHDSI GIS WG

Exercise 3: OMOP integration

Work in progress. These tutorial materials are still under active development and will continue to change until the tutorial takes place on October 20, 2026. Content, links, and exercises may be incomplete or shift without notice.

Goal

Understand and query the person-place-time model, validate vocabulary roles, and calculate a temporally aligned exposure metric — here, mean PM2.5 during pregnancy.

OMOP is person/event-centric; geospatial sources are place/time/attribute-centric. The bridge between them is: PERSON/clinical event ← LOCATION_HISTORY → LOCATION ← derived EXTERNAL_EXPOSURE. LOCATION_HISTORY captures entity, location, relationship, and start/end dates (supporting residential and other relationship types, while avoiding overlap ambiguity). EXTERNAL_EXPOSURE captures what, when, where, how related, source, and the numeric/categorical value and unit. Standard concepts always win over lexical-only mappings — an existing exact Standard concept is used when available, and a provisional source concept is retained or mapped rather than invented.

Steps

1. Review the vocabulary package relevant to your exposure: OMOP GIS (geometry/geographic semantics), Exposome (toxicants and related taxonomy), and SDoH (contextual determinants and source-derived items).
2. Query the pre-populated demo data, joining PERSON, LOCATION_HISTORY, EXTERNAL_EXPOSURE, the relevant vocabulary concepts, and the clinical interval (the pregnancy episode).
3. Identify the pregnancy interval for a given person, and identify every EXTERNAL_EXPOSURE row whose interval overlaps it — including the row that only partially overlaps.
4. Calculate mean PM2.5 during pregnancy, weighting by the degree of interval overlap rather than treating any overlap as full inclusion.
5. List every semantic field your calculation depended on: the exposure concept, its unit concept, the location relationship type, and the vocabulary each concept came from.

Deliverable

Mean PM2.5 during pregnancy for the queried person(s), plus the full list of semantic fields (concepts, units, relationship types, vocabularies) used to compute it.

Check Your Work

• Overlap weighting is correct — no interval treated as fully included when it is only partial
• No post-outcome leakage (no exposure data drawn from after the interval of interest)
• Units are correct and consistent across joined rows
• Each concept is used in its correct role (e.g., a unit concept is not confused with an exposure concept)

Bridge to Session 4: validated exposure facts become covariates; they do not become causal effects automatically.

Previous: Exercise 2 | Next: Exercise 4