Skip to main content

What's in the NPI database download: NPPES file contents

If you're about to download the NPI database, the useful question is what you'll get. (Our NPI Number guide covers the identifier itself.) The CMS NPPES monthly full dissemination file (September 2026 release, covering NPIs issued from 23 May 2005 to 12 September 2026) contains 9,798,758 records, and they're not all active individual providers. About three quarters are individuals, one in five are organizations, and a small slice are deactivated NPIs that still sit in the file. A second CMS file, the Deactivated NPI Report, lists the NPIs whose deactivation still stands, each with its date, so you can see how many are switched off each year. That matters if you're building a provider dataset that stays clean over time; that is a data hygiene problem. Every figure here is a Provyx count of these two CMS files: the monthly file as of 13 September 2026 and the deactivation report of 14 September 2026.

Our data

Contribution: first-party data, derived analysis, downloadable dataset

  • Collected: counts from the CMS NPPES September 2026 file, one row per NPI record (aggregated 13 September 2026), counts from the CMS NPPES Deactivated NPI Report of 14 September 2026, one row per NPI whose deactivation still stands (aggregated 14 September 2026), and the header row of the CMS NPPES September 2026 file (aggregated 13 September 2026).
  • Published: 29 record counts, each covering at least 10 NPI records (smaller groups are withheld), 11 deactivation-report counts, each covering at least 10 report rows (smaller groups are withheld), 3 deactivation-report figures, each covering at least 10 report rows (smaller groups are withheld), 1 deactivation-report figure, and 3 file-structure figures (counted from the file's header row).
  • Method: the exact queries, row counts and thresholds are in the methodology.
  • Computed: 5 derived figures (top-3 and top-5 totals) from the published figures. Each is a dataset row with its formula, inputs and expression, so it can be recomputed.
  • Tracked: in our source-fact ledger since 4 October 2026; no changes recorded yet.
  • Sources: the source manifest (JSON) lists the 2 sources behind the figures (registry files, deactivation report), each with its public URL, retrieval date, status and the reason any is left out. A card's or table's Source link goes to that source's own page.
  • Reproduce it: download the data as CSV or JSON (52 rows; variables: Records by entity type; Active NPI records; Deactivations per year; NPIs in the deactivation report; Report NPIs deactivated before 2016; Report NPIs deactivated from 2016; Report NPIs matched to deactivated file rows; Active records by enumeration year; Active records per practice state; Taxonomy slots filled per record; Monthly file structure; Derived figure; license CC BY 4.0).

How the 9.8 million NPPES records break down by entity type

The monthly file mixes three kinds of rows, and the split is lopsided toward individuals. Individual providers (NPI-1, the Type 1 vs Type 2 NPI distinction) account for 7,471,371 of 9,798,758 records, or 76.2%. Organizations (NPI-2) hold 1,972,058 records, 20.1% of the file. The remaining 355,329 records (3.6%) are deactivated NPIs with a blank entity type (CMS NPPES monthly file, September 2026 release, observed 13 September 2026).

Two practical points follow from that split. First, if your pipeline assumes every row is a practicing person, the 20.1% organization rows and the 3.6% deactivated rows will be something else. Second, the deactivated rows don't disappear from the download. They stay in the file with the entity type blank, so a naive import that treats every row as active will carry dead NPIs into your CRM or territory model.

For this breakdown, Provyx counted every record in the September 2026 file by its entity type code: 1 for individuals, 2 for organizations and blank for deactivated NPIs.

What each record holds

Each row of the monthly file has 330 columns. They cover the NPI and its entity type, the provider's or organization's name, mailing and practice-location addresses and phone numbers, the enumeration, last-update, deactivation and reactivation dates, an authorized official for organizations, up to 15 taxonomy codes (each with a license number, license state and primary flag) and up to 50 other provider identifiers.

Most active records use few of the 15 taxonomy slots. Of the 9,443,429 active records, 7,654,220 (81.1%) fill one taxonomy slot, 1,243,106 (13.2%) fill two, 343,088 (3.6%) fill three and 203,015 (2.1%) fill four or more. A slot count is not a count of different codes: the same code can fill more than one slot, usually with a license from another state. Taxonomy codes are self-reported by providers.

New NPIs keep arriving. The active records include 318,014 NPIs enumerated in 2016 and 631,058 enumerated in 2025, and 490,534 were enumerated in 2026 up to 12 September. Deactivated rows carry no enumeration date, so they are not in these counts.

By practice-location state, California has the most active records with a US practice address, 1,188,147 (12.6% of active records), followed by New York (658,554), Florida (654,298) and Texas (608,782). NPI database download by state: record counts gives the count for every state, with its individual and organization split.

Taxonomy slots filled per record

Top 4: Taxonomy slots filled per recordBar chart. One slot: 7,654,220 records; Two slots: 1,243,106 records; Three slots: 343,088 records; Four or more slots: 203,015 recordsOne slot7,654,220Two slots1,243,106Three slots343,088Four or more slots203,015records (0 to 7,654,220)
Top 4: Taxonomy slots filled per record. Source: Provyx aggregation of the CMS NPPES monthly file (September 2026 release, records to 13 September 2026). As of 13 September 2026.

One slot

7,654,220 records

As of 13 September 2026

Share (%)
81.1%

Source: Provyx aggregation of the CMS NPPES monthly file (September 2026 release, records to 13 September 2026), checked 13 September 2026

Two slots

1,243,106 records

As of 13 September 2026

Share (%)
13.2%

Source: Provyx aggregation of the CMS NPPES monthly file (September 2026 release, records to 13 September 2026), checked 13 September 2026

Three slots

343,088 records

As of 13 September 2026

Share (%)
3.6%

Source: Provyx aggregation of the CMS NPPES monthly file (September 2026 release, records to 13 September 2026), checked 13 September 2026

Four or more slots

203,015 records

As of 13 September 2026

Share (%)
2.1%

Source: Provyx aggregation of the CMS NPPES monthly file (September 2026 release, records to 13 September 2026), checked 13 September 2026

Taxonomy slots filled per record
AttributeOne slotTwo slotsThree slotsFour or more slots
Records7,654,220 records, as of 13 September 20261,243,106 records, as of 13 September 2026343,088 records, as of 13 September 2026203,015 records, as of 13 September 2026
Share (%)81.1%, as of 13 September 202613.2%, as of 13 September 20263.6%, as of 13 September 20262.1%, as of 13 September 2026

Active records by enumeration year

Active records by enumeration yearBar chart. 2016: 318,014 records; 2017: 329,676 records; 2018: 390,724 records; 2019: 399,632 records; 2020: 427,166 records; 2021: 463,573 records; 2022: 468,076 records; 2023: 509,596 records; 2024: 547,706 records; 2025: 631,058 records; 2026: 490,534 records2016318,0142017329,6762018390,7242019399,6322020427,1662021463,5732022468,0762023509,5962024547,7062025631,0582026490,534records (0 to 631,058)
Active records by enumeration year. Source: Provyx aggregation of the CMS NPPES monthly file (September 2026 release, records to 13 September 2026). As of 13 September 2026.
Active records by enumeration year
NameRecords
2016318,014 records
2017329,676 records
2018390,724 records
2019399,632 records
2020427,166 records
2021463,573 records
2022468,076 records
2023509,596 records
2024547,706 records
2025631,058 records
2026490,534 records

Active records per practice state

Top 5 of 10: Active records per practice stateBar chart. CA: 1,188,147 records; NY: 658,554 records; FL: 654,298 records; TX: 608,782 records; OH: 408,060 recordsCA1,188,147NY658,554FL654,298TX608,782OH408,060records (0 to 1,188,147)
Top 5 of 10: Active records per practice state. Source: Provyx aggregation of the CMS NPPES monthly file (September 2026 release, records to 13 September 2026). As of 13 September 2026.
Active records per practice state
RankNameRecordsShare (%)
1CA1,188,147 records12.6%
2NY658,554 records7.0%
3FL654,298 records6.9%
4TX608,782 records6.4%
5OH408,060 records4.3%
6MI358,035 records3.8%
7PA336,445 records3.6%
8IL319,057 records3.4%
9NC277,639 records2.9%
10MA259,106 records2.7%

How many NPIs are deactivated each year

The CMS NPPES Deactivated NPI Report of 14 September 2026 lists 355,330 NPIs, each with the date CMS deactivated it, going back to 23 May 2005. They are the same NPIs as the 355,329 deactivated rows in the monthly file, apart from one that the file shows as reactivated. The chart and table count the 269,113 deactivated from 2016 to 2026; the other 86,217 were deactivated from 2005 to 2015. NPIs that CMS deactivated and later reactivated are generally not in the report, so each year's count is the NPIs deactivated that year that are still deactivated. Recent years sit well above the earlier ones.

NPIs deactivated per year

Top 5 of 11: NPIs deactivated per yearBar chart. 2025: 34,080 NPIs; 2024: 30,749 NPIs; 2022: 29,462 NPIs; 2023: 28,198 NPIs; 2026: 26,157 NPIs202534,080202430,749202229,462202328,198202626,157NPIs (0 to 34,080)
Top 5 of 11: NPIs deactivated per year. Source: Provyx aggregation of the CMS NPPES Deactivated NPI Report (14 September 2026). As of 14 September 2026.
NPIs deactivated per year
RankNameNPIs
1202534,080 NPIs
2202430,749 NPIs
3202229,462 NPIs
4202328,198 NPIs
5202626,157 NPIs
6202025,631 NPIs
7202123,249 NPIs
8201921,994 NPIs
9201819,666 NPIs
10201715,938 NPIs
11201613,989 NPIs

The spread across years is wide. The three heaviest years, 2025, 2024 and 2022, account for 94,291 NPIs combined, and the top five years together reach 148,646. The six lighter years, 2016 through 2021, add 120,467 between them, so the recent years carry noticeably more weight than the early ones (CMS NPPES Deactivated NPI Report, 14 September 2026).

One caution on reading this as a trend: each figure is a count from a single report snapshot, and the 2026 figure covers a partial year. The report doesn't state why counts rose, so read the year-over-year movement as an observation only.

What this means for teams buying or building provider data

If you work in healthcare sales, marketing or RevOps, the file contents point to three concrete decisions.

Plan for entity-type handling from day one. With 76.2% of records being individuals and 20.1% organizations, any dedupe or territory assignment logic needs to treat NPI-1 and NPI-2 rows differently. A rep assigned "Dr. X's organization" and "Dr. X" as two targets is a direct consequence of ignoring the entity type field.

Build deactivation screening into your refresh cycle. The deactivation report shows five-figure annual counts in every year since 2016, peaking at 34,080 NPIs in 2025. If you download the NPI database once and never re-check it, some of your records will be for NPIs CMS has since deactivated. The Deactivated NPI Report is the public source to check against, and it's free. Records that stay active can still be out of date: our NPI staleness report measures how long active records go without an update.

Verify against the public source before you buy. Because both files are published by CMS, you can check any vendor's claims about record counts or deactivation handling yourself.

Limits of this data

These figures describe the files in their own terms, and the boundaries matter.

Coverage: the record counts cover the NPPES monthly full dissemination file as released in September 2026, with NPIs issued from 23 May 2005 to 12 September 2026. NPPES is a registry of NPIs, so the counts measure registered NPIs, not practicing clinicians, market size or employment. An NPI can exist for an entity that no longer sees patients, and the 355,329 deactivated records make that explicit.

Definitions: "individual" and "organization" follow CMS's NPI-1/NPI-2 entity types. Deactivated means CMS has flagged the NPI as deactivated in its report; neither file says why (the deactivation reason column in the monthly file is blank on every deactivated record), and an NPI that CMS later reactivates normally drops out of the report.

Snapshot date: the record counts come from the monthly file as of 13 September 2026 and the deactivation counts from the report of 14 September 2026. CMS updates the monthly file regularly, so counts will move; treat these as a dated snapshot rather than a permanent total. The 2026 deactivation count runs only to 14 September and will grow in later reports.

Unit: each record counts one NPI. Each NPI-1 record belongs to one individual, but an active record does not show that the person is still practicing; organization records count organizations, not locations or clinicians within them. Don't read 9.8 million records as 9.8 million practices or 9.8 million providers.

How to use the dataset

The CSV and JSON files hold 52 rows: one row per aggregate group (each published figure), plus one row per derived figure with its formula and inputs (47 observed, 5 derived). The rows behind each group (the collected source rows) are not published.

Columns: id (row id (observed rows r01..., derived figures d01...)), kind (observed or derived), entity, attribute, scenario, value, number (the figure as a plain number), unit, scope (the scope, or a derived figure's input row ids), input_price_scope, scenario_scope, formula (a derived figure's formula in words), expression (its arithmetic over the input figures), source_owner, source_url, observed_at, fixture.

Variables: Records by entity type; Active NPI records; Deactivations per year; NPIs in the deactivation report; Report NPIs deactivated before 2016; Report NPIs deactivated from 2016; Report NPIs matched to deactivated file rows; Active records by enumeration year; Active records per practice state; Taxonomy slots filled per record; Monthly file structure; Derived figure. NPPES figures as of 13 September 2026; Deactivation report figures as of 14 September 2026. License: CC BY 4.0.

Each observed row's source_url is its public source (the page the data came from) or, for a figure built from several sources, the section of the source manifest that lists them; derived rows list their input rows in scope.

To use it, filter attribute to the figure you need, sort by number, and cite the row's source_owner and observed_at. Recompute any derived row from its expression and the input rows its scope lists.