Healthcare data has a geography that is easy to overlook. Every encounter, referral and claim happens somewhere: in a clinic, a ward, a lab or a patient’s home. When systems exchange health data through HL7 FHIR, that “somewhere” is represented by a small number of resources, and handling them well is a spatial data problem as much as a clinical one. This article is written for geospatial practitioners who meet healthcare projects, and for health IT teams who meet maps. It covers how FHIR models places and organizations, how provider directories are structured, and what geocoding and mapping of a care network involves.
How FHIR represents places and organizations
Two resources carry most of the weight.
The Location resource carries details and position information for a physical place where services are provided and resources and participants may be stored, found, contained or accommodated. As the FHIR Location definition shows, it can be as coarse as a building or as fine as a room or a bed. Several elements matter for spatial work:
- address, an optional postal address;
- position, geographic coordinates on the WGS84 datum, with latitude and longitude required when the element is present and altitude optional;
- partOf, a reference to a containing Location, which allows hierarchies such as room, wing and building;
- type, one or more codes that describe the function of the place;
- managingOrganization, a reference to the Organization responsible for it;
- hoursOfOperation, with days, times and a free-text field for exceptions.
The Organization resource is the legal or administrative body: a hospital system, a practice group, a payer or a department. It has its own addresses, identifiers and hierarchy, and one organization usually manages many locations.
Practitioners are connected to places indirectly. A PractitionerRole links a person to an organization, to a specialty and to one or more locations, which is how a directory can answer “which cardiologists work at this site”.
For anyone used to GIS data models, the pattern is familiar: points with attributes, containment hierarchies, and a managing entity. The differences are in the details. Coordinates are optional, so many Location records have only an address. The hierarchy is expressed through references, not through geometry. And the same physical building may appear several times, once per organization that operates in it.
Provider directories
A provider directory is the main place where these resources become a product. The HL7 Da Vinci Plan-Net implementation guide defines a FHIR interface to a health plan’s insurance plans, their networks and the organizations and providers that participate in them. In the Plan-Net guide, Location, Organization and Practitioner are profiled on top of the US Core guide, and supporting resources such as OrganizationAffiliation, PractitionerRole, HealthcareService and Endpoint are included. Publishing this data through a standard FHIR API lets third-party apps query a plan’s network and present it to consumers.
The standard fixes the shape of the data. It does not fix its quality. Directories are known for stale entries: practitioners who have moved, locations that closed, phone numbers that no longer work. Anyone building a map on directory data should expect to spend effort on cleaning, deduplication and provenance, and to display the date of last update where the source provides it.
Geocoding, with caution
Many Location records contain addresses without coordinates, so a map-based product usually has to geocode. Several points deserve attention.
Geocoder accuracy varies by place. Urban addresses with rooftop-level matches behave very differently from rural routes, campuses with several buildings and facilities that share a street address. A match score belongs next to every coordinate, and low-confidence results should be flagged and reviewed instead of silently plotted.
A single point rarely describes a healthcare site. A hospital has an administrative address, an emergency entrance, a parking structure and several clinics under one roof. Decide which of these the user needs, and model it explicitly, using the Location hierarchy where the source data supports it.
Coordinates that arrive in the data should not be trusted blindly. Check the datum, the order of latitude and longitude and the plausibility of the point against its stated address and region.
Retain the original address as supplied. Normalization and geocoding change over time, and a record of the input makes later corrections possible.
Mapping a care network
Once locations are reliable, typical questions appear: which sites are reachable within a given travel time, where are the gaps between sites, how do the networks of two plans overlap?
Straight-line distance is a weak proxy for access. Travel time on the road network, with public transit where relevant, is closer to a patient’s experience, and the choice of routing engine and assumptions should be stated on the map. Service availability matters too: a site open on weekday mornings is not equivalent to one open every day, which is why hoursOfOperation and service-level data are worth carrying into the analysis.
Aggregations deserve care. Choropleths by postal area or county hide variation inside the unit, and small-number areas can suggest patterns that are noise. When health data is joined with place, such as patient counts by region, small cells may identify individuals, so the product needs rules for suppression and for the level of detail shown. Questions about what is permitted belong to privacy and legal specialists in the organization.
Engineering notes
A few practices help a spatial pipeline fed by FHIR data.
Store the FHIR resources as received, with a derived spatial table beside them, so that a change in the geocoding approach does not require a new data pull. Index the derived table with a spatial index and keep a stable key back to the FHIR resource identifier.
Treat updates as events. Directories change, and a pipeline that refreshes on a schedule needs a way to detect closed locations and to remove or archive them.
Test with messy data, not only with sample bundles: missing coordinates, several addresses on one Location, hierarchies that point to missing parents and organizations with conflicting names.
Keep the interface contract clear. If the product consumes data through a vendor or payer API, record which resources, search parameters and profiles are actually available, since these vary between servers.
When integration help is needed
Connecting these data sources to a mapping product involves FHIR clients, authorization, mapping logic and geodata handling. Teams without that combination of skills sometimes look for outside engineering support, and for those teams, learn more about JetRuby’s EHR integration engineering offer: software work, without clinical or insurance expertise. The same question applies to any provider: what will the engineers build, and which decisions stay with the product owner?
Whichever route a team takes, the principle is the same. Treat the places in healthcare data as data with provenance, uncertainty and history, and the maps built on them will be worth trusting.