Referential integrity
Dangling `Practitioner` / `Organization` references in `PractitionerRole`, unresolvable `managingOrganization` in Location, and Organization-to-HealthcareService coverage.
Headline
Referential integrity is clean but coverage is sparse. 0.002% of 33.4M declared cross-resource references actually dangle (target missing). But only 9.6% of Endpoints carry a managingOrganization (108,584 of 1,128,169) and only 61.8% of Locations do (1,567,973 of 2,535,686). H8: the NPD bulk export now ships HealthcareService. CMS added it in the 2026-08-20 release, alongside InsurancePlan, taking the export from 6 resources to 8.
724 / 33.4M = 0.00%
unit: percent
What this means
Payer data teams
97% of NDH Endpoints have no managingOrganization back-reference. You can’t reliably traverse Endpoint → Organization. Work around via NPI-based secondary joins (NPPES, CAQH).
Provider data teams
If your organization has Endpoints registered, audit whether they declare managingOrganization pointing back to you. The 97% coverage gap is in this exact pointer.
FHIR implementers
Integrity of DECLARED references is 100%: zero dangling refs across 17M edges. The defect pattern is under-population, not broken pointers. Trust your resolver, but expect 4 of the 10 NDH IG resources (HealthcareService, InsurancePlan, Network, Verification) to be absent entirely.
Regulators
The NDH bulk export omits 4 of the 10 NDH-IG resources (HealthcareService + InsurancePlan + Network + Verification). Any rule citing those resources cannot be measured from the current public-use artifact.
Null hypothesis
Cross-resource references resolve at ≥99% inside the bulk export.
Denominator
All reference fields across `PractitionerRole.practitioner`, `PractitionerRole.organization`, `Location.managingOrganization`, and `HealthcareService.providedBy`.
Data source
Edge tuples extracted from the NPD bulk export in a single streaming pass, queried in DuckDB.
Notes
Integrity (dangling rate among declared references): H6a PR→Practitioner 0.0000%, H6b PR→Organization 0.0000%, H7 Location→Org 0.0000%, Endpoint→Org 0.6668%. All near zero — when a reference is declared, it resolves. Coverage (share of rows with the optional reference populated): PR→Practitioner 100.00% (required), PR→Organization 91.89%, Location→managingOrganization 61.84%, Endpoint→managingOrganization 9.62%. The Endpoint→Organization gap pairs with H5 (98.69% of Orgs have no Endpoint referencing them) — the Endpoint↔Organization link is sparse in both directions. H8 requires HealthcareService. Which NDH IG resources the export carries is read from the published manifest on every run rather than hardcoded, because it changed: HealthcareService and InsurancePlan were absent through 2026-05-08 and present from 2026-08-20. Resources shipped in this release: Endpoint, HealthcareService, InsurancePlan, Location, Organization, OrganizationAffiliation, Practitioner, PractitionerRole.
Get the next finding in your inbox. One email per release, no filler.