GeoHealth US Corp produces environmental monitoring data in the United States, and aggregates that data with historical environmental data from US Government sources. Data sources include ve programs that track pollution history at the US Environmental Protection Agency, and air quality data collected by GeoHealth US Corp itself. Environmental monitoring data is combined with information from the US Department of Health & Human Services, including the US National Library of Medicine, to automatically associate environmental conditions with diseases and symptoms. Semantic Web techniques are used to perform the data integration, navigate the data for analysis, and to drive the display of contextually relevant data in a Web user interface. GeoHealth US Corp is a Virginia Bene t Corporation (or \B-corporation"), a for-pro t corporate entity that includes positive impacts on society and the environment in its core mission.
Human health is impacted by at least three classes of information: Lifestyle,
genetics, and environment. Of these, environmental conditions remain the least
poorly integrated in the US healthcare system. Environmental health
information in the US is sparse, scattered, and in forms that make it di cult to nd,
to combine, and to relate to individual patients. In some cases, it is of
questionable quality. We set out to address those problems in order to provide a
more complete picture of the environmental impacts on US public health. The
potential bene ts are signi cant given the size of the US healthcare market:
Approximately 3 trillion USD is spent annually on healthcare[
GeoHealthUS approached this problem by relating multiple US Government datasets in Linked Data formats (mostly from the US Environmental Protection Agency) with pseudo-real-time environmental air quality data collected by the company. Additional Linked Data was created to describe diseases and chemical substances related to them. Such data was previously available only in traditional (XML) formats from the US Department of Health & Human Services.
The evolving GeoHealthUS Web site is available at http://geohealth.us.
Individuals may currently view historic pollution report data while the newer
environmental data is being integrated1. The site complies with the Linked Data
Platform v1.0 speci cation[
GeoHealthUS has created mobile sensor packages called GeoHealthBoxesTM to rapidly collect air quality information. GeoHealthBoxes are typically mounted on vehicles and measurements taken. GeoHealthBoxes use a combination of Open Hardware, software written by GeoHealth US Corp and a number of proprietary environmental sensors.
Newly collected environmental data from mobile sensors was considered too voluminous for RDF modeling to make sense. Instead, RDF summaries were created to facilitate the location, and querying of such data from larger relational databases. RDF and non-RDF data is presented in a single Web user interface. Contextual subsets of data are also presented for download in various reports and formats (such as Turtle, JSON-LD, and CSV). Contextual subsets of non-RDF data are converted to RDF at download time, as requested. 3
Historic environmental data from the US Environmental Protection Agency (EPA) was available from the programs summarized in Table 1.
Historical data from government sources represents a 25-year history of environmental pollutants, and some of their e ects. All government data was either available in RDF formats or converted into RDF models for the purpose of simple composition. Vocabulary mapping of government data sets was undertaken as necessary.
The US EPA operates a Linked Open Data service in a quality assurance mode. This data service is not yet publicly available. The source data for the programs is generally available via http://data.gov, but in some cases EPA must be contacted directly to acquire information. This situation highlights the relative immaturity of US Government data sources when users desire to combine arbitrary data sets for further analysis and/or repurposing.
Table 2 list the namespaces of some of the common Semantic Web vocabularies used to represent the RDF portion of the data. Core vocabularies included the rdf, rdfs, owl, skos, and xsd namespaces. 1 A prototypical, and temporary, interface for a portion of air quality data is available at http://geohealth.us/home/pages/livedata.xhtml?view Namespace URI Purpose foaf http://xmlns.com/foaf/0.1/ Nearness, depictions geo http://www.w3.org/2003/01/geo/wgs84 pos# Locations place http://purl.org/ontology/places# Locations dbpedia http://dbpedia.org/resource/ Units of measure, companies vcard http://www.w3.org/2006/vcard/ns# Addresses
We made use of a number of RDF vocabularies speci c to the EPA datasets, which include de nitions such as the classi cation of facilities, substances, and reports. These vocabularies are under the base URI of http://opendata.epa.gov, but have not yet been formally published by the US EPA. We know of them through our prior work with the agency, and look forward to them being publicly available soon. Location information in EPA datasets was augmented by the use of a custom vocabulary created to represent US postal (ZIP) codes to facilitate the creation of maps and other geographic displays. 4
Additionally, a number of custom vocabularies were developed to represent information speci c to the GeoHealthUS application, such as the description of diseases (extracted from traditional data formats available from the US Department of Health & Human Services). This data was mapped to chemical substances via SRS identi ers and to clinical ndings using SNOMED-CT identi ers. 5
DB DB DB DB Historical Data ETL DB DB DB DB DB
SQL SPARQL RDF Data Layer
The GeoHealthUS poster illustrates three bene ts of a Semantic Web approach to data integration: 1. Conversion to RDF formats facilitated data integration across many data sets by the simple mechanism of identi er alignment. 2. Presentation to end users is via a small number of SPARQL v1.1 queries, simplifying maintenance requirements, and reducing maintenance costs over traditional approaches. 3. The approach was shown to function in a large, real-world use case with signi cant economic potential.