Pharmacies in the National Directory

by Eugene Vestel · 2026-10-26 · proposal for the RTPBC and National Directory communities

CMS NPD already holds almost every pharmacy. Its FHIR records do not say which ones are pharmacies, and they mostly do not say where pharmacists work. Data: CMS NPD release 2026-10-07.

In plain words

Think of the National Directory as a phone book. It lists almost every pharmacy, but it does not write "pharmacy" next to the name. So you can find a pharmacy if you already know its number, but you cannot ask for "all pharmacies near me".

It also lists pharmacists, but for most of them it does not say which pharmacy they work at.

The missing facts already exist in other public sources. They are just not copied into the phone book.

What we measured

QuestionAnswer
Active pharmacy organizations in NPPES (any 3336 taxonomy in the first three slots)109,203
Of those, with an Organization record in CMS NPD108,637 (99.5%)
Organization records carrying a pharmacy taxonomy code (3336xxxxxX)0 of 2,058,500
Organization records naming any taxonomy code system (NUCC or the NDH taxonomy value set)0
Pharmacies whose record mentions an NCPDP ID1 of 108,637
Pharmacies carrying an endpoint464 of 108,637
HealthcareService records with a service name, type or specialty0 of 54,445
Pennsylvania pharmacists in CMS NPD with a PractitionerRole (where they work)812 of 11,865 (6.8%)
Of those pharmacists, in a CMS-enrolled (Medicare) billing group1

The NPPES side is the public BigQuery snapshot dated 2026-02-07, so pharmacies registered since then are not counted. The 108,637 pharmacies sit on 113,077 Organization records. 108,636 are typed prov, like every other provider organization. 4,437 are text-only ein tax records that share a pharmacy NPI, and 4 have no type. The ein records explain the gap between records and NPIs. 45% of these records (50,829 of 113,077) have "pharm" in the name, which is the only pharmacy signal in the record. Role figures come from role coverage by profession; the HealthcareService figure from new resource types.

One note on the pharmacist number. All 11,865 people in it already hold an NPI, so it says nothing about whether pharmacists need one. It shows that having an NPI does not give someone a workplace record. Across professions, CMS NPD workplace records follow Medicare billing: in Pennsylvania 74.8% of physicians have one, against single digits for nurses (6.2%) and the pharmacy category, technicians included (6.7%). Few pharmacists bill Medicare as individual providers, which fits that pattern: 1 of the 11,865 is in a CMS-enrolled group.

Try it yourself

1. The CMS NPD record for a real pharmacy

curl "https://ainpi.dev/api/npd/search?type=organization&npi=1548282270"

Returns "CVS PHARMACY 16602" with type prov. Nothing in the record marks it as a pharmacy. The state shown is Rhode Island, the corporate billing address. The CMS record also carries the store's physical address, 5125 Jonestown Rd, Harrisburg PA, marked type: physical, after the billing address. A Location for the store exists in CMS NPD, but nothing links it to this record. AINPI's search currently shows the first address listed, and we are fixing that.

2. The same NPI in NPPES

curl "https://npiregistry.cms.hhs.gov/api/?version=2.1&number=1548282270"

Returns primary taxonomy 333600000X, Pharmacy. Joining the two is the workaround available today.

3. Every Pennsylvania clinician, one row per NPI

curl -O "https://ainpi.dev/api/v1/findings/role-gap-composition-pa.csv"

Columns: npi, category, taxonomy_code, primary_switch, nppes_deactivated, ndh_role, cms_enrollment_org. Keep rows whose taxonomy_code begins 1835 to get the 11,865 pharmacists; ndh_role = Y marks a workplace record and a non-empty cms_enrollment_org a CMS-enrolled group. Filtering on category = pharmacy instead gives 12,427, because that category also holds 562 pharmacy technicians (183700000X).

What already exists

NPPES taxonomy. CMS already holds each pharmacy's NUCC code in NPPES. It does not reach the FHIR export.

Plan pharmacy directories. Medicare Advantage plans with drug coverage must already publish their pharmacy directories through their public Provider Directory API, including the type of pharmacy. 42 CFR 422.120(b)(2) requires:

the MA-PD's pharmacy directory, including the pharmacy name, address, phone number, number of pharmacies in the network, and mix (specifically the type of pharmacy, such as "retail pharmacy")

updated no later than 30 calendar days after the plan receives a change. Those directories are required to state the type of pharmacy, and nothing in CMS NPD links them to a plan.

Open measurement. AINPI re-measures role coverage and the new resource types every CMS release, with the method and code, at ainpi.dev. The pharmacy counts above were measured for this note and are not yet a published series.

Proposal: how this should work

  • CMS NPD. Carry the NUCC taxonomy CMS already holds in NPPES into the FHIR export, on the Organization and on a HealthcareService per pharmacy. In that HealthcareService, set category to pharm, specialty to the 3336 code and providedBy to the Organization, as NDH models it in its retail pharmacy example. Make it searchable. This one change lets anyone ask the API for pharmacies.
  • CMS NPD. Link each pharmacy's Location to its Organization, or at least list the physical address first. In this record the store address sits on the Organization after the corporate billing address. A Location for the store exists (5125 Jonestown Rd #16602, phone 717-671-6903), but it has no managingOrganization and no PractitionerRole points at it. Linking it lets a per-store NPI resolve to a place. We left NCPDP IDs out of this ask. Our understanding is that NCPDP licenses its pharmacy data, but we have not checked the terms, so whether CMS could republish the IDs is a question for NCPDP and CMS.
  • CMS NPD. Fill pharmacist workplace records from a source beyond Medicare billing, and link each MA-PD plan's pharmacy directory endpoint to the plan, so a pharmacy can be traced to the networks it belongs to.
  • CARIN RTPBC. The RTPBC pharmacy profile says the NCPDP Provider ID "is the preferred identifier for this purpose because it specifies the individual pharmacy location". It adds that "alternatively, an organizational NPI may be used". Phone and address then help tell locations apart: they "SHOULD be populated to assist in identifying a particular pharmacy location". Marking pharmacies in CMS NPD helps the NPI path. It does not replace the NCPDP ID, which the CMS NPD record of 1 of 108,637 pharmacies mentions.
  • Implementers, now. Look up the pharmacy in CMS NPD by NPI, take its type from NPPES, and read the address marked type: physical or use: work, not the first one listed. Pharmacies registered recently may be missing; we have not measured how far CMS NPD lags NPPES.
  • AINPI. Watch HealthcareService and the pharmacy counts each release, so everyone can see when the fix lands.

How we measured

The CMS NPD side is the 2026-10-07 release loaded into BigQuery, read from the raw resource JSON of cms_npd.organization rather than from flattened columns. The NPPES side is bigquery-public-data.nppes.npi_raw, whose newest enumeration is 2026-02-07. A pharmacy is an active type-2 NPI (no deactivation date) with any taxonomy code beginning 3336 in slots 1 to 3. The counts were measured on 2026-10-22 and re-run on 2026-10-26 with the same results.

Each query ran in BigQuery with a 100 GB cap:

bq query --use_legacy_sql=false --maximum_bytes_billed=100000000000 < query.sql

Pass the SQL on stdin: a query that begins with a -- comment is otherwise read as a command-line flag.

Pharmacies in NPPES, and how many have an Organization record (109,203 and 108,637; about 0.35 GB scanned):

WITH pharm AS (
  SELECT npi
  FROM `bigquery-public-data.nppes.npi_raw`
  WHERE entity_type_code = 2
    AND npi_deactivation_date IS NULL
    AND (STARTS_WITH(healthcare_provider_taxonomy_code_1, '3336')
      OR STARTS_WITH(healthcare_provider_taxonomy_code_2, '3336')
      OR STARTS_WITH(healthcare_provider_taxonomy_code_3, '3336'))
),
ndh AS (
  SELECT DISTINCT _npi
  FROM `thematic-fort-453901-t7.cms_npd.organization`
  WHERE _npi IS NOT NULL
)
SELECT
  COUNT(*) AS pharmacies,
  COUNTIF(ndh._npi IS NOT NULL) AS with_organization_record
FROM pharm
LEFT JOIN ndh ON ndh._npi = pharm.npi

Taxonomy anywhere in any Organization record (0 pharmacy codes, 0 records mentioning NUCC or the word "taxonomy", which also covers the NDH value set URL; about 2 GB scanned):

SELECT
  COUNT(*) AS org_records,
  COUNTIF(REGEXP_CONTAINS(TO_JSON_STRING(resource), r'3336[0-9A-Z]{5}X')) AS pharmacy_code,
  COUNTIF(STRPOS(TO_JSON_STRING(resource), 'nucc.org') > 0) AS nucc_system,
  COUNTIF(REGEXP_CONTAINS(TO_JSON_STRING(resource), r'(?i)taxonomy')) AS any_taxonomy
FROM `thematic-fort-453901-t7.cms_npd.organization`

A broader search for any NUCC-shaped string ("[0-9]{3}[A-Z0-9]{6}X") finds 5 records, every one in the name field: four organizations whose name is a taxonomy code, and one called "2020EYEMAX". None is a coded taxonomy.

NCPDP and endpoints on the pharmacy records (1 and 464 pharmacies; about 2.5 GB scanned):

WITH pharm AS (
  SELECT npi
  FROM `bigquery-public-data.nppes.npi_raw`
  WHERE entity_type_code = 2
    AND npi_deactivation_date IS NULL
    AND (STARTS_WITH(healthcare_provider_taxonomy_code_1, '3336')
      OR STARTS_WITH(healthcare_provider_taxonomy_code_2, '3336')
      OR STARTS_WITH(healthcare_provider_taxonomy_code_3, '3336'))
)
SELECT
  COUNT(*) AS org_records,
  COUNT(DISTINCT o._npi) AS pharmacies,
  COUNT(DISTINCT IF(STRPOS(LOWER(TO_JSON_STRING(o.resource)), 'ncpdp') > 0,
                    o._npi, NULL)) AS pharmacies_mentioning_ncpdp,
  COUNT(DISTINCT IF(ARRAY_LENGTH(JSON_QUERY_ARRAY(o.resource, '$.endpoint')) > 0,
                    o._npi, NULL)) AS pharmacies_with_endpoint
FROM `thematic-fort-453901-t7.cms_npd.organization` o
JOIN pharm ON pharm.npi = o._npi

The 108,637 pharmacies sit on 113,077 Organization records, because 4,437 text-only ein tax records share a pharmacy NPI with a prov record. Counted by record rather than by pharmacy, 507 carry an endpoint and 1 mentions NCPDP. As a positive control, every one of the 113,077 records carries the NPI identifier system and 113,073 carry a type key, so the zeros above come from records the query did read.

The Pennsylvania pharmacist figures are recomputed from the committed role-gap-composition-pa.csv as described in step 3 above: 11,865 rows with taxonomy_code beginning 1835, 812 of them with ndh_role = Y, and 1 with a non-empty cms_enrollment_org.

Limits

  • The NPPES snapshot is from February 2026. Newer pharmacies are in neither count.
  • The pharmacist figures cover Pennsylvania only.
  • "Pharmacy" here means any 3336 taxonomy in the first three NPPES slots. A narrower definition, such as 333600000X only, gives a smaller count.
  • Endpoint and identifier counts read the raw FHIR JSON of the 2026-10-07 release. They do not test whether an endpoint answers.

Questions or corrections: Eugene Vestel, gene@fhiriq.com. Method and code are open at github.com/FHIR-IQ/AINPI. Every number here should be checked against the primary source before you rely on it.