Endpoint-to-organization linkage
An endpoint is only actionable if you can say whose it is. "There is a FHIR server at this URL" answers nothing on its own; "this URL belongs to the organization with NPI 1234567890" is what an integrator or a payer doing network validation needs. H50 measures how often Endpoint.managingOrganization is present and resolvable, and publishes the crosswalk it produces. Result at 2026-08-20: 16,262 of 110,973 FHIR REST endpoints (14.7%) resolve to a managing organization, leaving 94,711 URLs with no owner in the directory. It was 19,334 of 114,071 (16.9%) at 2026-05-08, so attribution moved backwards; the endpoint count fell too, meaning some of the change is removal rather than lost names. 88 references are present but point at an organization the file does not contain. The crosswalk this yields is published as a CSV alongside this finding.
Headline
Of 110,973 FHIR REST endpoints in the 2026-08-20 NDH release, 16,262 (14.7%) resolve to a managing organization. The remaining 94,711 are URLs with no owner in the directory. Every endpoint that does resolve reaches an organization carrying an NPI, so the link yields a usable base-URL-to-NPI crosswalk for 16,262 endpoints across 15,940 organizations.
16.3K / 111.0K = 14.65%
unit: percent
What this means
Startups + integrators
The published crosswalk at /api/v1/findings/endpoint-org-crosswalk.csv resolves FHIR base URLs to an organization and its NPI. Use it rather than re-deriving the join. For the other 85%, the NDH cannot tell you whose endpoint it is, so plan a fallback rather than assuming attribution is available. H51 shows vendor files close most of that gap.
FHIR implementers
Host is not a proxy for organization. EHR vendors host thousands of tenants on one domain, so api.platform.athenahealth.com identifies the vendor and never the practice. Only managingOrganization distinguishes them, which is precisely what is missing for most endpoints.
CMS publishing the data
The shortfall is concentrated, not uniform, so it is addressable per vendor. Of the 23 hosts carrying at least 100 endpoints, 13 publish no organization link on any of them, covering 43,601 endpoints; others attribute most of theirs. H51 then found the stronger version of this: those same vendors publish the organization names themselves, against the same URLs, so 65% of the gap can be closed by ingesting files that are already public rather than by asking anyone to collect anything new.
Methodology readers
Presence and resolvability are reported separately on purpose. A reference can exist and point at nothing, and counting presence alone would score dangling references as successes. In this release 724 of the managingOrganization references dangle, against 94,711 that are simply absent. The gap is overwhelmingly absence, not breakage, and that distinction matters because absence is fixed by populating a field while breakage is fixed by repairing referential integrity.
Null hypothesis
Endpoints in the NDH carry a resolvable managingOrganization, so a FHIR base URL can be attributed to an organization. Rejected: 83.1% of FHIR REST endpoints have no organization link.
Denominator
The 110,973 Endpoint resources with connectionType.code = hl7-fhir-rest in the pinned 2026-08-20 release. Direct Trust addresses (1,017,196) are reported separately and excluded from the crosswalk, because they are messaging addresses rather than callable base URLs per H28.
Data source
Three capped joins of cms_npd.endpoint against cms_npd.organization, reconstructing the FHIR reference as CONCAT("Organization/", o._id). Presence and resolvability are counted separately so a dangling reference could not be scored as a success. Compute script: analysis/h50_endpoint_org_linkage.py. Under a cent per run.
Notes
Presence and resolvability are reported separately because a reference can exist and point at nothing. In this release they are identical: 724 of the managingOrganization references across both connection types dangle. The gap is absence, not breakage, which matters because absence is fixed by populating a field and breakage would be fixed by repairing referential integrity. Host is not a substitute for the organization link. EHR vendors host thousands of tenants on one domain, so a host identifies the vendor and never the practice. The rate is bimodal rather than uniformly low. Of the 23 hosts carrying at least 100 endpoints, 13 publish no organization link on any of them, covering 43,601 endpoints; others attribute most of theirs. That is a publishing-behaviour difference between vendors rather than one systemic cause, and it means the fix is per-vendor: a host sitting at 0% is not partially populated, it has never populated the field at all. Direct Trust addresses are excluded from the crosswalk. They are messaging addresses rather than callable base URLs (H28), though their linkage rate is reported in the chart for completeness. The crosswalk covers the FHIR REST subset only, which is the right denominator for anything that resolves a base URL.
Get the next finding in your inbox. One email per release, no filler.