Observation

Epic equivalent: Social history / Flowsheet entries / Questionnaires / Structured note templates

The observation table is the catch-all for clinical facts that don't fit into conditions, drugs, or measurements. Smoking status, social determinants of health, patient-reported outcomes, screening responses, advance directives, and reasons for medication non-use all land here.

In Epic, this data comes from social history tabs, flowsheet entries, questionnaires, and structured note templates. The OHDSI community's rule: if the data can go in a more specific table, it should. If not, it goes here.

Epic-to-OMOP Field Mapping

Field reference (click to expand)

OMOP Field	Epic Equivalent	What It Captures
observation_id	Observation record ID	Unique identifier
person_id	Patient ID / MRN	Links to the patient
observation_concept_id	Structured concept	Standardized concept (e.g., "Former smoker", "Lives alone")
observation_date	Observation date	When recorded; observation_datetime has time precision
observation_type_concept_id	Source type	Origin: EHR, survey, registry, etc.
value_as_number	Numeric value	For scaled/scored responses (e.g., pack-years, pain score)
value_as_string	Free-text value	Unstructured text response (not for standardized analysis)
value_as_concept_id	Coded value	Categorical response as a concept (e.g., "Yes", "No", "Former")
qualifier_concept_id	Modifier / context	Optional severity, status, or other qualifier
unit_concept_id	Unit	Unit for numeric observations
provider_id	Recording clinician	Who recorded the observation
visit_occurrence_id	Linked encounter	Visit context
observation_source_value	Local field name	Original EHR label (e.g., "TOB_STATUS")

What to Watch For

Common pitfalls

Domain overlap

If it looks like a diagnosis, check condition_occurrence first. If it has a numeric result with units, check measurement. observation is the last resort.

Prefer concept_id over string

value_as_string is not analyzable at scale. Use value_as_concept_id for categorical responses.

High site-specific variation

Different institutions load very different things into observation. Always check what's actually in your dataset before assuming coverage.

Research Patterns

Question	Tables Involved
Current smoking prevalence	observation_concept_id (smoking status) + value_as_concept_id
Food insecurity screening results	observation_concept_id (SDoH) + value_as_concept_id (positive)
Patient-reported pain scores > 6	observation_concept_id (pain score) + value_as_number filter
Patients who declined statin therapy	observation (medication non-use) + condition_occurrence
Advance directive documentation in hospice patients	observation_concept_id (advance directive) + visit_occurrence