Environmental Data Standards Landscape¶
Deliverable 2.1.
Preliminary
This survey is a working document synthesized from deep-research reports. Findings have not been fully vetted by all stakeholders and may be revised.
Executive Summary¶
This report surveys standards, tools, and projects linking environmental and geospatial data to clinical health outcomes.
Key findings:
-
No ratified OMOP table for environmental data. The OHDSI GIS Working Group has proposed the
external_exposuretable, but it is not in the official OMOP CDM spec. Every site uses ad-hoc workarounds. The GIS WG opened gaiaCatalog Issue #19 in July 2025 to finalize the design. -
Limited OMOP vocabulary coverage. The Exposome Vocabulary has ~79,000 toxin-target relationships, but most environmental exposure variables used in studies (air pollutants, water contaminants, noise, greenspace, climate variables) lack OMOP concept IDs. This is the single biggest blocker.
-
Epic Healthy Planet captures SDoH but not environmental exposures. Epic supports structured SDoH screening (PRAPARE, AHC-HRSN) at the point of care, encoded with LOINC; the values are subsequently mapped to OMOP's Observation table via downstream ETL. Environmental exposures have no clinical workflow equivalent -- they must always be linked post-hoc via geocoding.
-
The Monarch/EHS-Data-Standards ecosystem provides schema infrastructure.
linkml-microschema-profiledefines a composable CDE pattern EnVar should adopt.exposome-schemaprovides reusable exposure concepts but lacks geospatial metadata.somamodels lab-assay outcomes but not population-level geospatial exposures. -
Most environmental epidemiology does not use OMOP. European exposome projects (EHEN/EXPANSE, HELIX, UK Biobank), US cohort studies (ECHO, MESA Air), and government platforms (CDC Tracking, OpenSAFELY) use custom data models.
-
Transformation metadata is the critical gap. Across 42+ datasets, the spatial interpolation parameters, temporal aggregation windows, and exposure assignment methods that convert raw measurements into exposure estimates are systematically underreported.
OHDSI GIS Working Group¶
The external_exposure Table¶
Proposed CDM extension linking persons to place-based health determinants via location_history. Key fields: person_id, location_id, exposure_concept_id, exposure_start/end_date, value_as_number, unit_concept_id.
What it cannot represent:
| Limitation | Example |
|---|---|
| Distributions / uncertainty | PM2.5 = 12.3 +/- 2.1 ug/m3 |
| Multi-component exposures | Air toxics risk = f(benzene, formaldehyde, ...) |
| Time series | Daily PM2.5 over a 90-day window |
| Spatial resolution metadata | "12km CMAQ grid cell" vs "1km LUR estimate" |
| Temporal aggregation method | "Annual mean" vs "98th percentile of daily values" |
| Exposure assessment method | CMAQ vs satellite-derived vs monitor interpolation |
| Data provenance | URL/DOI of source dataset, model version |
These limitations are what EnVar's micro-schemas and OMOP proposals should address.
GAIA Toolchain¶
Four components: gaiaDB (PostGIS + OMOP integration), gaiaCore (R/Python/REST API), gaiaCatalog (Schema.org JSON-LD metadata), gaiaDocker (deployment). Architecture is sound; needs more content and vocabulary.
Key gap: GAIA uses Schema.org JSON-LD; EnVar uses LinkML. No bridge exists today, but LinkML can generate JSON-LD, making a bridge feasible.
OMOP Vocabulary Coverage¶
| Vocabulary | Coverage |
|---|---|
| OMOP GIS Vocabulary | Geographic and geospatial concepts |
| OMOP Exposome Vocabulary | ~79,000 toxin-target relationships (T3DB-sourced); thin on common epi variables |
| OMOP SDoH Vocabulary | SVI, ADI, EJI, COI; 6,738 concept associations (per OHDSI GIS WG); uses "Phenotypic Feature" domain |
Critical gap: Many commonly used environmental variables (PM2.5 at various aggregation levels, NO2, noise metrics, NDVI, heat indices) lack OMOP concept IDs.
Epic Healthy Planet / SDoH¶
Epic's Healthy Planet supports structured SDoH screening at point of care. Environmental exposures differ fundamentally: they are area-based (not individual), have no clinical workflow (always retrospective linkage), and lack standard vocabulary.
Area-level SDoH is the structural precedent. Indicators like SVI and ADI face the same challenges as environmental exposures: values belong to geographies, not persons. They use the same geocoding -> spatial join -> external_exposure pipeline. Environmental data should follow this pattern.
Monarch/EHS-Data-Standards Schemas¶
soma¶
Models biological assays for airway biology (~40+ classes). ExposureCondition captures lab exposure (agent + concentration + duration) but not population-level geospatial exposures. The QuantityValue pattern is reusable; the AOP framework is conceptually relevant.
Recommendation: Don't import directly. Useful background, not a schema dependency.
exposome-schema¶
ExposureEvent abstract class has relevant concepts (exposure_route, exposure_medium, exposure_duration) but inadequate types (bare strings, bare floats, no units). No support for spatial resolution, temporal aggregation, model type, or data source.
Recommendation: Reuse concepts but re-type them as proper value micro-schemas.
linkml-microschema-profile¶
The template EnVar micro-schemas should follow. Defines composable, identifier-free CDEs:
| Class | Purpose |
|---|---|
Quantity |
Numeric value with unit (UCUM/UO/QUDT) |
Timepoint |
A point in time |
TimeInterval |
A period with start, end, duration |
CodedValue |
A term from a controlled vocabulary |
The ClinicalMeasurementRecord pattern provides the basis for an EnvironmentalExposureRecord. OMOP is already in the prefix map.
Federal Programs¶
| Program | Focus | Status |
|---|---|---|
| HEW Data Accelerator | EnVar's parent program; GB-EDoH standardization | Active |
| ORNL C-HER | 30+ spatially indexed exposomic datasets | Active; test datasets for EnVar |
| CHORDS | Climate and health data infrastructure (wildfires) | Active; 3-year, $4M NIEHS project |
| CAFE RCC | Climate-health research coordination (BU/Harvard) | Active; $6.7M NIEHS grant |
European Exposome Projects¶
The largest coordinated exposure assessment efforts globally, but none use OMOP:
| Project | Scale | Focus |
|---|---|---|
| EHEN | 9 sub-projects, 126 research groups, 24 countries; €100M+ (Horizon 2020) | Air pollution, noise, greenspace, chemicals. Sub-projects include EXPANSE, ATHLETE, EPHOR, EQUAL-LIFE, EXIMIOUS, HEDIMED, HEAP, LongITools, REMEDIA. |
| EXPANSE (EHEN sub-project) | Urban settings | LUR models becoming de facto standards |
| ATHLETE (EHEN sub-project) | 18 birth cohorts | Multi-omics + external exposome |
| HELIX | 32K mother-child pairs (exposure modeling); 1.2K subset (biomarkers) | 200+ exposures; rexposome R package |
| HBM4EU | 28 countries; ended June 2022 | Chemical biomonitoring (EUR 74M) |
| PARC | ~200 partners across 28 countries; 2022–2029; EUR 400M (50% EU / 50% Member States co-funded) | Chemical risk assessment |
| UK Biobank | 500K participants | Geocoded environmental linkages via NHS-linked addresses (addresses not released to researchers) |
Clinical Data Integration¶
- All of Us / CLAD / CHEL -- CHEL annotates participants with H3 hex IDs and provides geospatial datasets. Jim Phuong's geocoding pipeline inspired the Geodata 4 Health collaboration.
- FHIR PIT (UNC Chapel Hill) -- Integrates EHR data (FHIR format) with EPA CMAQ, roadway, and Census ACS data. Validated on ~160K patients with asthma or related pulmonary conditions (Xu et al. 2022, PMC9015759). Feeds into ICEES (NCATS Biomedical Data Translator).
Dataset Landscape¶
42+ environmental datasets routinely used in epidemiology, organized by domain. See the priority variables survey for the specific variables derived from these datasets.
Most-Used Dataset Families¶
| Family | Example | Why it dominates |
|---|---|---|
| Regulatory monitoring | EPA AQS | Authoritative, validated, long records |
| Satellite/reanalysis surfaces | MODIS AOD, ERA5 | Spatial completeness |
| National administrative systems | SDWIS, Superfund NPL | Policy-relevant |
| ML fusion exposure models | Di et al. PM2.5, CACES | Gap-filled, high-resolution |
Cross-Cutting Methodological Challenges¶
- Spatial misalignment: 43--68% reduction in risk ratio estimates for primary pollutants (Goldman et al. 2010, Atlanta time-series, Environ Sci Technol; PMC2948846)
- MAUP: NO2-COVID-19 associations changed from positive to negative to null depending on aggregation strategy [VERIFY CLAIM]
- Geocoding error: 74.4% urban vs 10.5% rural address-level precision (Goin et al. 2017, French E3N cohort, Environ Health; PMC5324215)
- Residential mobility: 55% of Texas children with leukemia moved between birth and diagnosis (Janitz et al. 2019, J Expo Sci Environ Epidemiol; PMC11465071)
The Gap: What No Existing Schema Captures¶
The transformation metadata layer -- spatial interpolation, temporal aggregation, exposure assignment, data fusion -- is not captured by any existing schema.
Critical New Slots for Environmental Micro-Schemas¶
| Slot | Type | Critical? |
|---|---|---|
spatial_resolution |
CodedValue | Yes |
temporal_aggregation_method |
CodedValue | Yes |
exposure_model_type |
CodedValue | Yes |
data_source |
uri | Yes |
temporal_coverage |
TimeInterval | Yes |
buffer_distance |
Quantity | Maybe |
model_uncertainty |
Quantity | Maybe |
Standards Silos¶
| Tradition | Standards | Blind Spots |
|---|---|---|
| Geospatial/climate | CF Conventions, ISO 19115 | Nothing about epidemiological transformations |
| Biomedical | OMOP, FHIR, SNOMED, LOINC | No environmental exposure concepts |
| Ontological | ECTO, ENVO, ExO | No transformation metadata |
No crosswalk exists from CF standard_names to OMOP concepts. EnVar should produce SSSOM mappings from ECTO/ENVO/CHEBI to OMOP.
Alignment Priorities¶
| Priority | Action | Impact |
|---|---|---|
| 1 | Expand OMOP Exposome Vocabulary with environmental variable concepts | Highest -- biggest blocker |
| 2 | Support external_exposure table ratification via OHDSI governance |
High -- structural prerequisite |
| 3 | Build on linkml-microschema-profile for environmental CDEs |
Right pattern, team involved |
| 4 | Coordinate with OHDSI GIS WG on all proposals | All OMOP changes go through this group |
| 5 | Create SSSOM mappings from ECTO/ENVO/CHEBI to OMOP | Bridge the vocabulary silo |
| 6 | Design micro-schemas with 5--6 critical geospatial metadata slots | Keep it lean; iterate from use cases |
| 7 | Build GAIA <-> LinkML bridge via JSON-LD generation | Connect dataset and variable metadata |