What does the CMS National Provider Directory actually tell you — and where does it fall short?
NPD today is an NPPES pass-through enriched with PECOS reassignment data and CEHRT vendor submissions. CAQH is not in the pipeline. That means the same stale, self-attested data that plagued NPPES is inherited intact — in a cleaner FHIR wrapper. Here's what we can verify by looking at 27.2M ingested records.
Enter an org name to search NPD. Supply your own reference numbers to see the variance.
Working through the NDH release specs against what's actually visible in the 2026-04-09 NPD release:
| FHIR field | Primary source | Update cadence | Enforcement |
|---|---|---|---|
| Practitioner.identifier[NPI] | NPPES | Daily feed | CMS-issued, never changes |
| Practitioner.name | NPPES | Self-attestation | None |
| Practitioner.address[] | NPPES + PECOS (Medicare subset) | On self-update / enrollment | PECOS validates for Medicare only |
| Practitioner.telecom | NPPES | Self-attested | None |
| Practitioner.qualification (taxonomy) | NPPES + state licensure (where wired) | Uneven | State boards vary |
| Practitioner.active | NPPES | Effectively never flips to false | None |
| Organization.identifier[NPI] | NPPES Type 2 | Daily | CMS-issued |
| Organization.address | NPPES + PECOS | Uneven | PECOS only for Medicare |
| PractitionerRole.organization + .practitioner | CEHRT vendor bulk submissions (HTI-1) + PECOS reassignment | Vendor-dependent | HTI-1 publication mandate, not accuracy |
| PractitionerRole.specialty | PECOS Physician Specialty codes | Medicare enrollment events | PECOS validates |
| Endpoint.address + .connectionType | CEHRT vendor reporting (HTI-1) + payer submissions (CMS-9115-F) | Vendor self-report | HTI-1 publication |
| Endpoint.managingOrganization | CEHRT vendor assertion | — | None verified |
Bottom line: NPD Practitioner + Organization is ~90% NPPES with PECOS enrichment for the Medicare subset. CAQH is not in the pipeline. Each field inherits the accuracy guarantees of its upstream — for NPPES fields that means self-attestation with no enforcement.
active flag
96.7%
marked active nationally. Reality is closer to 85-90% actively practicing. The flag almost never flips.
meta.lastUpdated
100%
of practitioners carry the same 2026-04-09 release timestamp. There is no per-provider attestation date.
qualification.period
Mostly empty
License entries typically have no start/end dates — you cannot distinguish a current license from one that lapsed.
Best available proxy: a Practitioner being referenced by a recent PractitionerRole that came from a payer directory submission (CMS-9115-F) implies current network participation. This channel is thin in 2026 and scheduled to ramp through CY2027 for Medicare Advantage. Absent CAQH-style 120-day re-attestation pressure, there is no mechanism in the current NPD pipeline to flip a provider from active to retired.
From the HTE release specifications repo (ftrotter-gov/HTE_data_release_specifications), published specs cover:
None mention credentialing, re-attestation, or CAQH ingestion. Scope is endpoints, affiliations, and connectivity health metrics. The NDH profile library definesVerificationResult withattestation-who andprimarysource-who extensions, but these are aspirational — no one is populating them in the 2026-04-09 release.
A CAQH-derived feed could theoretically enter NPD indirectly via the payer submission channel: a payer who credentials through CAQH and then submits its provider directory API. But: (1) you lose CAQH field provenance once reshaped into FHIR Provider Directory format, (2) nothing in NPD will say "this came from CAQH re-attestation on date X," and (3)it's scoped to network participants only — not a general credentialing layer. For a single source of truth that tells you who's actively practicing right now, the current NPD architecture does not deliver it.
Every NPD number here is subject to (a) a measured 0.015% ingestion error rate against the CMS source manifest (see Data Quality), (b) deduplication choices (ROW_NUMBER by _id), and (c) FHIR reference parsing. Name-pattern matching (e.g., "UPMC") may catch unrelated orgs like "Arthritis and Internal Medicine Associates-UPMC in San Diego" which is unlikely to be UPMC Pittsburgh. Published organizational numbers (UPMC marketing pages, Health Plan network pages) are themselves marketing-curated and may not reflect current credentialed counts. CAQH counts are not publicly available at the org level.
Anyone making a business or clinical decision off these numbers should verify against primary sources. The core finding stands regardless: NPD today is a structural improvement over searching NPPES by hand — not yet a provenance improvement.