NDH payer endpoint and organization coverage
The NDH is discussed in the CMS provider-directory community as the place where payer API endpoints and organization identifiers will live, so a consumer could discover a payer’s public Provider Directory API from one federal index instead of hunting developer portals payer by payer. H49 measures how far the shipped release is from that. Result at 2026-08-20, and half of it changed: CMS added payer organizations. 27 now carry a "pay" type coding and own 233 health plans, where the 2026-05-08 release had no payer type at all across 1,999,818 typed resources. Payer endpoints did not follow. 0 of those 27 organizations carry an endpoint, and 1 of 110,973 FHIR REST endpoints self-labels as a payer provider directory. The endpoint table is overwhelmingly an EHR-vendor patient-data index: athenahealth, eClinicalWorks and Office Ally alone account for 65,338 of the endpoints, and 58 of 2,962 distinct hosts are payer-operated. The 92 payer-host endpoints that are carried are largely unusable as an index anyway: only 7 have a managingOrganization, and 30 are marked status=error. A control directory verified live, public and unauthenticated under CMS-9115-F is absent from the index entirely.
Headline
CMS added payer organizations in the 2026-08-20 release: 27 carry a 'pay' type coding and own 233 Medicare Advantage plans, where the previous release had no payer type at all. Payer endpoints did not follow. 0 of the 27 payer organizations carry an endpoint, 1 of 110,973 FHIR REST endpoints is a payer provider directory, and a live CMS-9115-F directory is still absent from the index.
1 / 111.0K = 0.00%
unit: count
What this means
CMS publishing the data
Three concrete gaps, all additive rather than breaking. Organization.type has no payer coding, so payer entities cannot be represented or selected today. Of the 92 payer-host endpoints that are carried, only 7 have a managingOrganization, so the endpoint cannot be resolved to an organization for the other 85. And a directory that is live, public and mandated is absent entirely, which means the gap is not "payers have not built these" but "the index does not yet carry them".
Startups + integrators
Do not plan on discovering payer directory endpoints from the NDH in this release. They are not there. Endpoint resolution from the NDH gets you EHR patient-data URLs, which answer a different question than "where is this payer’s network directory".
Researchers
Classifying payers by organization name does not work here. Every Pennsylvania organization matching payer-like name patterns is a provider, such as KEYSTONE RURAL HEALTH CENTER. Host-based classification of the endpoint URL is the workable route, and the per-endpoint CSV ships alongside this finding.
Methodology readers
The provider-directory count is a floor, not a census: it counts servers that self-label in their URL path, and a directory at an unlabelled path would be missed. That is why the control probe exists. Without it, a low count could equally mean payers had built nothing, which would not be the index’s fault.
Null hypothesis
The NDH carries payer organizations and their public Provider Directory API endpoints, so it can serve as a discovery index for payer directories. Rejected on both halves at 2026-05-08. At 2026-08-20 the first half holds: 27 payer organizations arrived owning 233 health plans. The second half still fails: 0 of the 27 carry an endpoint and 1 of 110,973 FHIR REST endpoints is a payer directory. Identity arrived, reachability did not.
Denominator
The 110,973 Endpoint resources with connectionType.code = hl7-fhir-rest in the pinned 2026-08-20 NDH release (the Direct Trust HISP addresses, 1,017,196 of them, are messaging endpoints and are excluded per H28). The organization-type denominator is the Organization resources that carry any type coding; roughly half the Organization file carries no type coding at all and is not counted either way.
Data source
Three capped BigQuery scans of cms_npd.endpoint plus one of cms_npd.organization (reading the resource JSON column to enumerate every type coding rather than the flattened _org_type, so a payer type could not be hidden by the extractor). Plus one control probe by curl of a payer Provider Directory API verified live and unauthenticated. Compute script: analysis/h49_ndh_payer_endpoints.py. About $0.03 per run.
Notes
Re-measured against the 2026-08-20 release by streaming the published NDJSON directly, not through BigQuery, so the check could run the day the release landed rather than after a 45 GB ingest. Classification constants are imported from analysis/h49_ndh_payer_endpoints.py rather than copied, so this cannot drift from the finding it updates. Half of the original result is resolved: a payer organization type now exists. The other half stands: no payer organization carries an endpoint, and the control directory (Capital BlueCross) answers 200 to an unauthenticated request while remaining absent from the index, which separates 'nothing to index' from 'not indexed yet'. Reproduce with analysis/h49_recheck_release.py --dir <release dir>.
Get the next finding in your inbox. One email per release, no filler.