US Healthcare Provider Data: Every Public Source, Ranked (2026)

Published August 2026. Every source below was checked against its own publisher on 2026-08-28, and every row count was produced by downloading the file rather than quoting a summary.

You need a national file of US healthcare providers. You find the NPI registry, download the monthly file, and it lands with 9.7 million rows in it. That feels like a solved problem for about an hour, which is roughly how long it takes to notice that the United States has fewer than 1.1 million licensed physicians.

Both numbers are correct. They are counting different things, and the distance between them is where most provider-data projects quietly go wrong.

That gap is the whole subject of this guide. We downloaded and counted the major public sources rather than repeating the figures other people publish, and the honest summary is that there is no single file of American healthcare providers, no federal source carries licence status, and the sources that look most authoritative are often the stalest. Below is every public source worth knowing, ranked, with what each one actually counts.

US Physician Count: Why Five Sources Give Five Different Numbers

Start here, because almost every provider-data mistake we see traces back to picking the wrong number and then defending it. These are not five estimates of one quantity. They are five different objects.

Source	Figure	As of	What it actually counts
NPPES, all NPI records	9,726,865	Aug 2026 file	Every identifier ever issued and not purged, including 351,911 deactivated shells
NPPES, individual (Type 1)	7,415,294	Aug 2026 file	Live individual identifiers across every profession, not just physicians
NPPES, organisational (Type 2)	1,959,660	Aug 2026 file	Organisations, including every separately enumerated subpart and location
CMS Doctors and Clinicians	1,622,454 clinicians	Jul 2026	Clinicians with Medicare enrolment and claims in a 12-month lookback. Only 701,303 are MD or DO
FSMB census	1,082,187	End 2024	Distinct physicians holding an active, full, unrestricted licence, deduplicated across states
AAMC	1,032,365 active	2024	Licensed physicians who finished training and work more than 20 hours a week
BLS employment	804,140	May 2025	Wage and salary jobs at establishments. Excludes the self-employed entirely

Read down that table and the spread stops looking like error. NPPES is an append-only registry of administrative identifiers, issued on request to anyone who bills for care. It never deletes. Organisational subparts each get their own number, and one person can hold more than one. A country with roughly 6,000 hospitals shows nearly two million organisational identifiers for exactly that reason.

The others narrow as the gate tightens. FSMB counts licences held, which sweeps in physicians who are retired but still carry one. AAMC adds a work test of more than 20 hours and drops about fifty thousand people. BLS surveys employers rather than people, so it misses the self-employed altogether. You can see the size of that exclusion directly: the same occupations come to 862,800 in the BLS series that does include self-employment, against 804,140 in the one that does not.

Which Provider Data Source to Use for Which Job

• Identity resolution and joins: NPPES. It is the only free national file with a stable per-provider key, addresses and specialty. Filter to individuals, drop the deactivated shells, and never treat the row count as a headcount.
• “How many physicians are there” in a report: FSMB for licensed, AAMC for working. Say which one and why.
• Workforce supply and per-capita ratios: AAMC.
• Wages and metro labour markets: BLS, with the self-employed caveat stated out loud.
• Medicare-facing targeting and group affiliation: the CMS Doctors and Clinicians file.
• Licence status or disciplinary history: none of the above. No federal file carries it.

US Healthcare Provider Data Sources, Ranked

Twenty national sources, scored out of 20 on the same six criteria we used for our audit of all 51 US business registries: bulk availability, API, cost, field depth, freshness and licence terms. Higher is more usable, not more authoritative. Several low scorers are excellent for the job they were built for.

#	Source	Score	Access	Refresh	Carries NPI	What it is, and the catch
1	CMS Doctors and Clinicians (Care Compare) CMS Provider Data Catalog	19	Free CSV and API	Monthly	Yes	Medicare-enrolled clinicians, with group and facility affiliation. The only free CMS file that maps a clinician to a facility
2	HRSA Health Center Service Delivery Sites HRSA	18	Free CSV and XLSX	Daily	Yes	Site-level FQHC data, and the only non-CMS source here with a native NPI column
3	Medicare Order and Referring / Revalidation CMS	18	Free CSV and API	Twice weekly	Yes	An eligibility flag file. No address and no specialty, so it is not a directory
4	Provider of Services files CMS	17	Free CSV and API	Quarterly	No	Medicare-certified institutions. Now split into three datasets with different schemas
5	Hospital General Information CMS Care Compare	17	Free CSV and API	Quarterly	No	5,419 hospitals keyed on certification number, not NPI
6	NPPES data dissemination CMS	16	Free bulk download	Monthly full, weekly delta	Yes	The spine of everything. Self-reported and never verified
7	Open Payments CMS	16	Free bulk and API	Annual	Yes	Industry payments. Only covers clinicians who received one
8	Medicare Part D prescribers CMS	16	Free CSV and API	Annual	Yes	Prescribing volume. Addresses are a frozen snapshot from a year after the data year
9	Medicare physician and other practitioners CMS	16	Free CSV and API	Annual	Yes	Service volume. Excludes Medicare Advantage entirely
10	Medicare FFS public provider enrolment CMS	16	Free CSV and API	Quarterly	Yes	Enrolment records with type and state. The file most people want when they reach for Order and Referring
11	HCRIS cost reports CMS	16	Free ZIP, API on derived extract	Quarterly	No	Institutional finances. Raw files ship with no column headings
12	NPI Registry API CMS	15	Free API, no key	Daily	Yes	The freshest provider source in the country. Query only, hard paging cap
13	HRSA shortage areas (HPSA and MUA) HRSA	15	Free CSV and shapefiles	Daily	No	Designations, not headcounts. Withdrawn designations stay in the file
14	BLS occupational employment and wages BLS	15	Free flat files and API	Annual	No	Wages and geography. Excludes the self-employed, and 45% of physicians land in one catch-all category
15	SAMHSA behavioural health facilities SAMHSA	12	Free, portal navigation	Annual	No	Facility level. The public locator and the downloadable survey are two different things
16	HRSA Area Health Resources Files HRSA	11	Free ZIP	Annual	No	County-level aggregates. 6,000 variables with different reference years in one edition
17	CDC NCHS provider surveys CDC NCHS	11	Free FTP	Irregular	No	De-identified visit microdata. The office-based physician series broke after 2022
18	NUCC taxonomy code set NUCC (private body)	10	Free CSV	Twice yearly	No	The specialty vocabulary NPPES points at. Commercial embedding needs a licence
19	AAMC physician workforce data AAMC (private body)	6	Dashboard only	Annual	No	Aggregates only. The file underneath is licensed and cannot be passed on
20	FSMB physician census FSMB (private body)	6	Free PDF	Biennial	No	The only count sourced from the regulators themselves. Published as a journal article

Carries NPI matters more than it looks. It is the difference between a source you can join and a source you can only look at.

Five Problems With US Provider Data Sources

Each of these is invisible until it has already corrupted something, which is the expensive order to meet them in.

1. Deactivated NPI Records Arrive as Empty Rows

In both the monthly and weekly NPPES files, a deactivated provider’s row carries the identifier and the deactivation date. Every other column is blank. We counted 846 such rows in a single recent weekly file of 35,299.

Run a naive upsert on that and you have not marked a provider inactive. You have erased their name, address and specialty from your database and replaced them with nulls. The blast radius is every downstream table that trusted those fields.

2. CMS Medicare Files Do Not Cover the Whole Profession

The CMS Doctors and Clinicians file is the best-structured provider file the federal government publishes, and it is a Medicare artefact. A clinician appears only with an approved enrolment and claims inside a 12-month lookback, which quietly excludes paediatricians, Medicaid-only and cash-pay practices, and VA physicians.

It is also not one row per person. The file is unique at clinician, enrolment, group and address, so 3,388,151 rows resolve to 1,622,454 distinct clinicians. Anyone counting rows is overstating the population by more than double, and a physician practising at four sites under two enrolments occupies eight rows on their own.

3. No Federal Source Carries Physician Licence Status

This one surprises people, because NPPES does carry a licence number field. That field is self-reported, never verified, and populated on roughly 69% of individual records. It proves nothing. CMS says so plainly on the download page: issuing an identifier does not validate that the provider is licensed or credentialed.

Licence status and discipline live with the state boards, and they are as fragmented as anything we have audited. More on that below.

4. CMS Facility Files Do Not Carry an NPI

Three of the main CMS facility sources carry no NPI. None. The Provider of Services files, Hospital General Information and the HCRIS cost reports are keyed on the CMS certification number instead. We checked the headers directly: there is no NPI column in any of them.

There is exactly one free bridge, and it is easy to miss. The Facility Affiliation file that ships alongside Doctors and Clinicians maps NPI to certification number across 2,253,582 rows. Without it, the individual half of the CMS estate and the facility half do not meet.

5. Federal Health Data APIs Fail Silently

This is the finding we did not expect and the one most likely to be quietly wrong in a production system today. Four separate federal APIs cap their results, and none of them raises an error when you exceed the cap.

API	Cap	What happens when you exceed it
Open Payments	500 rows	Returns HTTP 200 with an empty results array. Looks like an empty dataset
Provider Data Catalog	1,000 rows	Returns HTTP 200 with an empty array
data.cms.gov	6,500 rows	Returns 6,500 rows no matter how many you ask for
NPI Registry	skip capped at 1,000	Re-serves the same page forever, so a crawler loops on duplicates

Every one of them returns a success code. A paginator that trusts its own limit parameter will either conclude the dataset is empty or ingest the same rows indefinitely, and in both cases the job reports success.

State Medical Board Data: Where Physician Licence Status Lives

Since no federal source carries licence status, anyone who needs it ends up at the state boards. We sampled eight to establish the pattern, using the same scoring as above.

State	Score	Bulk route	Cost	Refresh	The catch
California	19	Free, direct public folder	Free	Weekly	Two boards, since osteopathic physicians are licensed separately. The address of record is explicitly not a practice location
Washington	17	Free open-data portal with a real API	Free	Daily	No address and no specialty in the file. The credential number format changed in March 2026, breaking older joins
Florida	16	Free, behind free registration	No fee published	Daily	The richest fields of the eight, including practice and mailing addresses separately. Self-reported and not verified by the department
Ohio	15	Statewide portal, not the board	No fee for the CSV	Daily	Published by a different agency than the board, covering 24 boards at once. Access gate unresolved from outside
Texas	11	Paid, portal with checkout	Not published anywhere public	Monthly	Single board covering both physician types, which is unusually clean. No price list exists outside the login
New York	10	Records request only	Varies	Real time	Licensure, discipline and profile sit in three separate bodies. Only discipline is openly published
North Carolina	9	Paid request form	$150 per copy, monthly	Monthly	Redistribution is banned outright in the purchase terms. That alone makes a resold national file unsound if it is included
Massachusetts	9	None published	Unverified	Unclear	Excellent per-record detail and no bulk route at all. The profile database is marked no-index

The mean is 13.25 out of 20. That average hides the real finding. Six of the eight will hand over a bulk extract. Only four do it free, only one offers a genuine API, and only one is a direct download with no form, no login and no fee.

The obstacles are rarely technical. Each state solved the same problem in a different building. Licensure and discipline are split across separate agencies in New York and Washington. Physician types sit in separate boards in California. The file covering Ohio’s medical board is published by a different department entirely. Schemas share no keys, almost none carries an NPI, and freshness ranges from daily to a file that does not exist.

So a national licence-status file is not a scrape. It is fifty-odd acquisition paths, several of them contractual rather than technical, on fifty different clocks, normalised into a schema none of the sources shares, with a fuzzy name-and-address match to attach identity. Anyone quoting national licence coverage from a single pipeline is either buying it from FSMB commercially or overstating it.

How We Scored Each Healthcare Provider Data Source

Each source was checked against its own publisher, and where a row count appears in this guide we produced it by downloading and counting the file rather than quoting a published summary. The NPPES monthly file is 1.15 GB compressed and 11.6 GB as CSV, which is part of why these numbers are usually repeated rather than checked.

Criterion	How it is scored
Bulk (0 to 5)	Free full direct download 5. Free via portal or registration 4. Paid bulk 3. Request only 2. Query or search only 1. None 0.
API (0 to 3)	Documented free public API 3. Key or payment required 2. Portal export only 1. None 0.
Cost (0 to 3)	Free 3. Nominal one-off 2. Hundreds per year 1. Thousands per year or per record 0.
Fields (0 to 3)	Identity, location, specialty and affiliation all present 3. Identity, location and specialty 2. Minimal or aggregate only 1.
Freshness (0 to 3)	Daily or weekly 3. Monthly 2. Quarterly 1. Annual, slower or unclear 0.
Licence (0 to 3)	No restriction 3. Attribution or minor terms 2. Restricted commercial use 1. Redistribution barred 0.

A low score is not a criticism. The FSMB census scores 6 and is the only physician count sourced from the regulators themselves. It scores low because it is a journal article rather than a dataset, which is exactly what it was meant to be.

Limitations of This Provider Data Audit

Stated plainly so you can weight it yourself.

• The state board sample is eight, not fifty. It establishes a pattern. It does not establish national coverage, and we have not priced the other forty-two.
• Several state figures are unverified because the sites blocked automated access during checking. Texas publishes no bulk price outside its login, Massachusetts published nothing we could read directly, and Ohio’s access gate could not be resolved from outside.
• Two commonly cited figures are stale and we have left them out. A widely repeated FSMB coverage number traces to a 2015 release, and the AAMC specialty report people cite was retired in 2023 and folded into a dashboard.
• The NPPES and CMS clinician counts are ours, not official statistics. They are reproducible from the files on the dates given, and CMS does not publish them as headline numbers.
• One publisher is being reorganised. SAMHSA is being folded into a new agency, so its URLs and publisher name are a live risk on any reference page including this one.

Frequently Asked Questions

Is there a national database of US healthcare providers?

There is a single national database of provider identifiers, and it is not the same thing. NPPES covers everyone who has ever been issued an NPI, but it is self-reported, never verified, includes deactivated records, and carries no licence status. For anything beyond identity you are combining sources.

How many physicians are there in the US?

Either 1,082,187 or 1,032,365, depending on the question. The first is FSMB counting physicians who held an active unrestricted licence at the end of 2024. The second is AAMC, counting those who finished training and work more than 20 hours a week. Quote either. Just say which one you used, because the two answer different questions.

What is the NPI database?

NPPES, the National Plan and Provider Enumeration System, is the federal registry behind every NPI. CMS publishes it as a free monthly file with weekly updates, and a free query API refreshed daily. It holds names, addresses, specialty codes and a self-reported licence number for every provider ever enumerated. It does not confirm that anyone is currently licensed, and it never deletes a record.

Which healthcare provider data source is updated most often?

The NPI Registry API, which CMS refreshes daily. The freshest bulk file is the Medicare Order and Referring list at twice weekly, and the freshest file with substantial fields is the weekly NPPES incremental. At the other end, the derived cost report extract runs roughly two years behind.

Where can I find physician licence status and disciplinary history?

No. It is held by the state boards, and assembled nationally only by FSMB as a commercial product. Some states publish discipline openly while withholding the licensee roster, and at least one bans redistribution of the roster it sells.

Working With US Healthcare Provider Data

None of this is a complaint about public data, and it should not read as one. The federal provider estate is genuinely generous by international standards: most of it is free, much of it has an API, and the licence terms are mostly permissive. The problem is not access.

The problem is that each source was built to answer one question for one programme, and the questions do not overlap. Medicare enrolment answers who can bill, the identifier registry answers who exists, and the employment survey answers what work pays. Stack them without reconciling what each one counts and you get a file that is confidently wrong, which is worse than an obviously incomplete one.

So pick the source that matches the question, say which one you used, and reconcile before you aggregate. We maintain pipelines across sources that behave like these, including the ones that go quiet. If you would rather not maintain twenty of them plus fifty state boards, talk to our team. And if you find a source here that has moved, tell us. Several will have, and the table is easier to keep honest with more eyes on it.