Analysis Capabilities and Future Directions

The OMOP GIS extension supports analyses beyond traditional clinical observational research. This page summarizes the analysis tools available now, the extensions the working group has proposed, and the methodological and ethical needs of geospatially informed health research. The proposed items are plans, not released software.


Current analysis infrastructure

Tool Role Limits for GIS use cases
ATLAS Primary user interface for OMOP analytics: cohort construction, characterization, prediction Limited support for spatial exposure variables from external_exposure, geographic visualization, multi-level modeling, temporal exposure sequences and spatial statistics
HADES Open-source R packages that form the analytical foundation of OHDSI Used directly for GIS work today: SQL access to external_exposure, flexible feature engineering from spatial and environmental variables, causal inference and prediction, network studies across sites

GIS use cases currently work through HADES and custom scripts. The tutorial shows this end to end with CaprForExtensions and FeatureExtractionForExtensions, and gaiaCore provides exposure analytics from R.

Analysis step HADES packages
Cohort construction CohortGenerator, Capr
Descriptive statistics CohortDiagnostics, FeatureExtraction
Population-level estimation CohortMethod, SelfControlledCaseSeries
Patient-level prediction PatientLevelPrediction
Data quality DataQualityDashboard
Federated analytics Strategus


Proposed HADES extensions

Package Purpose Proposed functions
GeoFeatureExtraction Extract and engineer features from external_exposure and spatial data extractSpatialExposures(): link person-level data to exposures
aggregateTemporalExposures(): moving-window and cumulative exposures
calculateSpatialLags(): neighborhood-effect metrics
imputeMissingExposures(): kriging or nearest neighbor
SpatialCohortMethod Causal inference that accounts for spatial confounding and clustering fitSpatialPropensityModel()
adjustSpatialConfounding(): restricted spatial regression, spatial+
assessSpatialUnmeasuredConfounding()
estimateSpatialEffects(): spillover and interaction
MultiLevelEstimation Multilevel models for individuals nested in neighborhoods fitMultilevelModel(): area-level random effects
partitionVariance(): intra-class correlation
crossLevelInteractions()
spatialMediation()
GeoVisualization Privacy-preserving visualization of spatial patterns createChoroplethMap(): with small-number suppression
plotSpatialClusters(): SaTScan, DBSCAN
visualizeExposureGradients()
animateTemporalMaps()

Extensions would follow the OHDSI contribution path: prototype, community review at working group meetings and symposia, validation across use cases and datasets, documentation (vignettes, function reference, validation reports), testing to HADES standards, submission through OHDSI governance, and long-term maintenance.


Methods by research domain

Domain Analytical needs Proposed analyses HADES gaps
Environmental epidemiology Spatiotemporal exposure assessment (linking monitoring data to locations over time)
Exposure measurement error
Distributed lag models
Extreme value analysis (such as heat waves)		 Case-crossover designs for acute exposures
Spatial survival analysis
Bayesian hierarchical models for multi-pollutant exposures
Mixture analysis (quantile g-computation, weighted quantile sum)		 Distributed lag models
Time-series analysis
Spatial survival analysis
Mixture analysis
Sociodemographic epidemiology Multilevel modeling
Structural confounding adjustment
Intersectionality analysis
Longitudinal neighborhood effects		 Multilevel propensity score methods
Spatial mediation analysis
G-methods for time-varying area-level confounders
Disparity decomposition (spatial Oaxaca-Blinder)		 Multilevel modeling
Spatial mediation
Disparity analysis
Time-varying area-level exposures
Spatial epidemiology Spatial cluster detection
Spatial regression (autocorrelation, boundary effects)
Disease mapping that preserves privacy
Spatial interaction and access modeling		 Bayesian disease mapping (BYM)
Spatial scan statistics (SaTScan)
Geographically weighted regression
Spatial accessibility (two-step floating catchment area)		 Spatial autocorrelation diagnostics
Spatial regression
Small-area estimation
Accessibility analysis


Transparency and ethical analysis

Topic Challenges Transparency today Proposed enhancements
Privacy and re-identification Small areas (Census tracts) can identify people with rare conditions
Spatial patterns can reveal protected information even when aggregated
Federated analysis must protect locations while allowing spatial statistics		 The CDM documents data structure and HADES documents methods, but privacy risk assessment is left to each researcher Automated small-number suppression
Differential privacy for spatial aggregation
Re-identification risk tools
Spatial k-anonymity and l-diversity metrics
Environmental justice and equity Analyses can reinforce injustice narratives
Data quality is unequal across communities (more monitoring in affluent areas)
Boundary choices affect equity results		 Limited guidance on equity-focused analysis, no standard EJ metrics, little documentation of bias in geospatial sources EJ impact checklists
Standard metrics (cumulative burden, disparate impact ratios)
Data quality reporting stratified by neighborhood
Community-engaged validation
Structural racism and social determinants Area-level determinants reflect historical and structural racism
Causal language must not blame communities
Confounding by structural factors is pervasive		 Vocabularies document SDOH concepts but not their structural origins; HADES assumes exchangeability that may not hold; little guidance on interpreting area-level effects Structural causal models
Sensitivity analyses for structural confounding
Explicit causal assumptions in every spatial analysis
Anti-racist data science frameworks


Role of AI and LLMs in transparency

Area Possible uses
Automated documentation Human-readable summaries of analysis pipelines (provenance)
Extracting causal assumptions from code and comments
Translating HADES code into plain language for stakeholders
Bias detection and mitigation Finding patterns of missing or poor data across communities
Reviewing analytical choices for bias
Drafting robustness checks for equity-critical analyses
Stakeholder communication Community-facing summaries of complex analyses
Question-and-answer interfaces to findings
Privacy-preserving, equity-focused visualizations
Risk Safeguard
Hallucinations introduce errors into documentation Human review of all LLM-generated documentation
Automated bias detection misses context-specific equity issues Community validation of AI-assisted equity assessments
Over-reliance reduces human engagement in ethical deliberation Transparency about LLM use in analysis workflows
Tools carry algorithmic bias Regular auditing of AI tools


Future directions

Direction Examples
Community-driven development Use cases identify analytical needs; HADES package proposals come from validated use cases; community review checks scientific and ethical standards (see use case status)
Emerging methods Causal inference for spatial exposures (spatial instrumental variables, regression discontinuity)
Machine learning for exposure prediction (satellite imagery, land use)
Agent-based modeling of neighborhood effects
Precision environmental health
Cross-disciplinary collaboration Environmental science, geography and GIS, urban planning, social epidemiology, ethics and philosophy
Open science Apache 2.0 licensed extensions, vignettes on public data, published validation studies, feedback through GitHub, forums and working group meetings


Resources