Introduction to Geoinformatics
BMIN7053
Cole Brokamp
2022-09-26
Geoinformatics
Geoinformatics
Transforming place-based data into actionable knowledge
- geospatial data science
- geographic information services
- public health informatics
- place-based determinants, outcomes
Geoinformatics and Health
- Identifying study population
- Identifying potential sources and routes of exposures
- Estimating environmental levels of pollutants
- Measuring community characteristics
- Estimating personal socioeconomic characteristics
- Statistical models with spatial correlations
- Temporal exposure estimates
Geoinformatics and Health
- Direct measurement of exposure or personal characteristics often not feasible in cohort- or EHR- based studies
- Often use retrospective records, only containing home addresses
- Geomarkers are highly variable with respect to time and space
- Links outcome, exposure, and confounding data by location
- Data usually available publicly
Geomarkers
Geomarkers
- Geomarker: Any geospatial measure that influences or predicts health
- Geocoding: Converting a string of text into spatial coordinates or boundaries
place (+ time) → estimating past “exposures”
Neighborhood Characteristics
interactive deprivation index map
| Geomarkers | Space | Time | Source | Available Software Products |
|---|---|---|---|---|
| Population, socioeconomic indicators, etc | Census tract or block | Yearly | US Census American Community Survey (ACS) | |
| Built environment & location efficiency, access to jobs and workers via transit, walkability index | Census tract | 2021 | EPA Smart Location Mapping | |
| Neighborhood Deprivation Index (poverty, education, housing, and health care coverage) | Census tract or zip code | 2015, 2018 | Material Deprivation Index | dep index R code; dep index DeGAUSS Container |
| Child Opportunity Index (education, health & environment, social & economic neighborhood characteristics) | Census tract | 2010, 2015 | Child Opportunity Index 2.0 | |
| Population, racial composition, socioeconomic indicators, racial and socioeconomic Index of Concentration at the Extremes, crowding | County, zip code, or census tract | 2018 | The Public Health Disparities Geocoding Project | |
| Racial Index of Concentration at the Extremes (ICE) | Census tract or zip code | Yearly | Racial ICE | zctaDB R package |
| Relative crime risk by crime subtype* | Census block group | Average (2010 - 2017) | AGS Crime Risk | |
| Location and type of gun violence incident | Exact | Annual (2014 - current) | Gun Violence Archive (GVA) | |
| Fraction of food retailers that are “healthy” | Census tract | 2011 | Modified Retail Food Environment Index (mRFEI) | |
| Food access indicators, SNAP benefits, demographics | Census tract | 2015 | USDA Food Access Research Atlas | |
| Medically Underserved Areas | County / Census tract | Current | Health Resources and Services Administration (HRSA) | MUA App |
| Social Vulnerability Index | County / Census tract | 2000, 2010, 2014, 2016, 2018 | Agency for Toxic Substances and Disease Registry (ATSDR) | |
| Community Resilience Estimates | County / Census tract | Current | US Census | |
| Social Deprivation Index (Area level deprivation based on ACS data) | County, census tract, ZCTA and PCSA | 2015 | Robert Graham Center | |
| Neighborhood Atlas | Census block group | 2015, 2019 | University of Wisconsin School of Medicine and Public Health |
Air Pollution
| Geomarkers | Space | Time | Source | Available DeGAUSS Products |
|---|---|---|---|---|
| PM2.5 | 0.74 sq km h3 hexagonal grid | daily (2000 - 2020) | Brokamp exposure model | addPmData R package; pm DeGAUSS container |
| PM2.5, NO2, O3** | 1 sq km grid | daily (2000 - 2016) | Schwartz exposure model | schwartz DeGAUSS container |
| air toxics cancer risk, respiratory hazard index, diesel PM, PM2.5, ozone, traffic proximity and volume, lead paint indicator, proximity to RPM sites, proximity to hazardous waste facilities, proximity to NPL sites, wastewater discharge indicator | Census block groups | Annual (2015 - 2020) | EPA EJ Screen | zctaDB R package |
| Location and amount of emissions | Exact location | Yearly (2008, 2011, 2014, 2017) | National Emissions Inventory (NEI) | |
| PM2.5, PM10, O3, CO, SO2, NO2 | Census block group (Contiguous US) | Annual (1999 - 2015) | CACES LUR | |
| PM2.5, SO4, NO3, NH4, OM, BC, DUST, SS | 0.01° × 0.01° (North America) | Annual (2000 - 2018) | ACAG MAPLE | |
| PM2.5, GBD-MAPS, NO2, Inverse Visibility, OM/OC, Surface Area, NOy Deposition, Dry Deposition, AMF Code | 0.01° × 0.01° (North America) | Annual / Monthly | ACAG Washington University in St. Louis | |
| PM2.5, NO2 | Tract Centers (Contiguous US) | Monthly (1999 – 2008) | Beckerman |
Environmental Exposures
| Geomarkers | Space | Time | Source | Available DeGAUSS Products |
|---|---|---|---|---|
| Air temperature, planetary boundary height, relative humidity, precipitation, wind | North America (0.3°×0.3°) | 8-times a day (2001 – current); daily means (1979 – current) | North American Regional Reanalysis (NARR) | addNarrData R package; narr DeGAUSS container |
| Greenspace as Enhanced Vegetation Index (EVI) | 250 × 250 m grid (contiguous US) | June 10 and June 25, 2018 composite | Moderate Resolution Imaging Spectroradiometer (MODIS) | greenspace DeGAUSS container |
| Land cover classifications (21 classes) | 30 x 30 m grid | Annual (2001, 2006, 2011, 2016) | National Land Cover Database (NLCD) | addNlcdData R package; nlcd DeGAUSS container |
| Fractional imperviousness | 30 x 30 m grid | Annual (2001, 2006, 2011, 2016) | NLCD Imperviousness | addNlcdData R package; nlcd DeGAUSS container |
| Roadway type, development intensity | 30 x 30 m grid | Annual (2001, 2006, 2011, 2016) | NLCD Imperviousness Classification | addNlcdData R package; nlcd DeGAUSS container |
| Vegetation Type | 30 x 30 m grid | Annual (2011, 2016) | NLCD Tree Canopy | addNlcdData R package; nlcd DeGAUSS container |
| Greenspace as fraction of land classified as green | 30 x 30 m grid | Annual (2001, 2006, 2011, 2016) | NLCD Greenness | addNlcdData R package; nlcd DeGAUSS container |
| Highway traffic intensity | Exact location (nationwide) | Yearly | US Department of Transportation (DOT) | addAadtData R package; aadt DeGAUSS container |
| Noise level | 270 m sq grids (nationwide) | 2015 | National Park Service Geospatial Sound Modeling (GSM) |
Greater Cincinnati Area Only
| Geomarkers | Space | Time | Source | Available DeGAUSS Products |
|---|---|---|---|---|
| Parcel classification and characteristics | Exact location (Greater Cincinnati Area) | Current | Cincinnati Area Geographic Information Systems (CAGIS) | |
| Home images, value, age, other physical characteristics, owner name | Exact | Exact Inspection dates | Hamilton County Auditor | hamilton R package for parcel-based geocoding |
| Traffic Related Air Pollution | Exact location (seven county area served by CCHMC) | Monthly (2001 – current) | Air pollution exposure model | ecat R package |
| PM2.5 components: Al, Cu, Fe, K, Mn, Pb, S, Si, V, Zn | Exact location | Average daily exposure (2002-2006) | Air pollution exposure model | |
| Lead Air Pollution | Exact location (seven county area served by CCHMC) | Monthly (2001 – current) | Air pollution exposure model | airPb R package |
| Estimated travel time by car | Census tract and traffic analysis area | Hourly (2016-2017) | Uber Move | |
| Combined Sewer Overflow events | Exact location | Daily (2012 - 2016) | Metropolitan Sewer District (MSD) | |
| Areas without access to medication-assisted treatment and/or behavioral therapy for opioid addiction | Exact location (seven county area served by CCHMC) | 2019 | MAT Desert | MAT Desert shapfiles |
| Crime | Exact location (Ohio) | Annual (2013 - 2018) | Ohio Incident Based Reporting System |
Harmonizing spatiotemporal resolution and extent
Geomarker Assessment
Geocoding
- converting location information text into coordinates
- most often a postal address into latitude and longitude
- most consider it a “magic black box”, but is error prone
Street range address
Exact location
Exposure Assessment
- Containing Geography
- Census tract linkage to survey data
- Census block linked to population density
- Neighborhood linked to policies or characteristics
Exposure Assessment
- Radial Measures
- Buffer designated around location with a radius
- Length / density of predicted sources
- Calculate mean, total length, or fraction within buffer
Exposure Assessment
- Exact Location
- Proximity to predicted source
- Nearest neighbor weighting and krigging
- Prediction models (land use models, etc)
Implementation Challenges
- Large data + inefficient manual data curation
- Technical expertise and software skills
- Privacy restrictions
Privacy Approaches
Protected Health Information (PHI)
- Confidentiality of research subjects must be safeguarded
- HIPAA-defined “Safe Harbor” provision prohibits sharing of identifiers and quasi-identifiers, such as:
- time finer than a calendar year
- spatial boundary with < 20,000 residents
Sharing Data with PHI
- Consent often not obtained for unforeseen future analyses
- Retrospective consent often not feasible + consent bias
- IRB and institutional DUA approvals can be lengthy and have different requirements
- Transmission of PHI to a third party often not possible
Anonymity and Reidentification
- Anonymity can ensure small, but non-zero, chance of reidentification
- Don’t conflate re-identification of identifiers with re-identification of quasi-identifiers
- quasi-identifiers recovered by merging with extant datasets
- institutional restrictions on sharing of quasi-identifiers
Exisiting Approaches
- Anonymization: geomasking, date shifting, generalization
- Independent Geomarker Assessment: differences in methods introduce differential bias
- Existing Software Approaches: costly, not scaleable, not reproducible
Decentralized Computation
Multi-Site Study
Bringing Data to Computation
Sending Computation to Data
Getting Computation Results
DeGAUSS
DEcentralized Geomarker Assessment for mUlti Site Studies
Curated and standardized library for secure, efficient, automated, and reproducible linkage of geomarkers to protected health and geolocation data
DeGAUSS
- Container framework that reads and writes CSV files
- No extensive computational resources
- No geospatial or computing expertise required
- PHI is never exposed to a third party or the internet
DeGAUSS
- Free and open source
- Automated and continuous documentation and integration
- Metadata curation and integration
- Community supports and contributions
High Resolution Spatiotemporal
High Resolution Spatiotemporal
Examples
Elevated Blood Lead
CFPOPD
