When organization NPI counts overstate address counts

Our data

We computed this report from 27,965 NPI records in 13 existing legal-organization profiles. The downloadable dataset contains 13 rows at the legal organization level. The cohort consists of these selected legal-name profiles; each retains its own source boundaries.

Download the computed CSV · JSON dataset · Input checksums and provenance · Reproduction method

The biggest differences are concentrated

Wal-Mart Stores East LP has 4,751 NPI records and 3,217 primary-address keys, a difference of 1,534. Costco Wholesale Corporation has 1,917 records and 727 keys, a difference of 1,190. Together those two profiles account for 2,724 of the cohort's 3,831 records above its primary-address count. A single conversion factor would therefore hide the concentration of the issue.

CVS Pharmacy Inc. offers a useful counterexample: 1,234 records map to 1,232 primary keys. Its difference is only two. Both profiles can be valid representations of their underlying registrations while behaving very differently in an address-based territory plan. The source supports comparing these specific legal-name extracts using their retained record and address counts.

Separate a registration from a place

An NPI identifies a registration represented by a row in this dataset. A location key groups the row's primary practice address. The two fields answer different questions. Counting NPIs describes the number of registrations present in the extract. Counting address keys describes how many distinct normalized primary address strings those registrations use within one legal name.

For a workflow built around service locations, the second count may be a more useful starting point. For a workflow that must preserve each registered entity's identity, replacing NPIs with address keys would lose information. Keep both dimensions in the data model. An account record can link to several NPIs, and a registration can link to addresses without forcing a one-to-one relationship.

A collision queue is more useful than automatic deletion

The cohort contains 3,141 primary keys carrying multiple NPIs, with 6,972 NPI records attached to those keys. That is a reviewable queue. The CSV provides the number of collision addresses and associated records for each legal name, so an analyst can prioritize the profiles with the largest reconciliation work before importing the complete source rows.

A reviewer should compare the full address, taxonomy, legal name and other-name fields within each candidate group. A pharmacy registration and an eyewear registration at one retail address might serve different operational needs. A local sales team can use those distinctions to decide which records its workflow needs. Preserve the original NPI and explain any account grouping in a separate, reversible crosswalk.

Why a percentage can mislead without its denominator

The difference column divides records above the primary-key count by the number of NPI records. Manual duplicate review would require a separate decision on each candidate group. The percentage describes what changes when this particular grouping rule is applied. A different rule, such as retaining suites or matching on a complete postal address, can produce a different result.

Use the numerator and denominator from the same legal-name row. Averaging the 13 percentages would give the 61-record New York City Health and Hospitals profile the same weight as the 11,140-record Walgreen Co profile. For a cohort-level comparison, sum the record counts and address counts first, then compute the ratio from those totals.

The place of legal names in an account hierarchy

Three source profiles carry names containing Walmart or Wal-Mart. This analysis keeps all three separate because their legal-name strings define the supplied extracts. Establishing a corporate-family relationship requires separate ownership evidence. Combining them would introduce an ownership assumption that is outside the retained dataset.

The same caution applies when a familiar brand has a surprisingly small count. This CVS Pharmacy Inc. extract covers one legal-name slice within the source. Start from the organization profile, inspect its explanation of similar legal names, and obtain separate ownership evidence before creating a parent-account rollup.

How to use the CSV in a territory review

Import the computed file as a diagnostic table rather than a prospect list. Choose an account grain first: registration, primary address within a legal name, or a separately verified corporate parent. Record that choice in the workbook heading so another person can identify whether the count describes addresses or companies.

Then open the source profile for a high-difference organization. Review several collision groups with different taxonomy combinations, keep their NPIs intact, and document which operational use cases require separate records. If an address appears invalid or outdated, investigate it with a current primary source. Retain that operating-status evidence alongside the dated comparison.

Checks that keep the comparison reproducible

The calculation reads every source NPI as text and checks uniqueness within each organization file before counting. The calculation preserves the source identifiers and requires each file to pass the uniqueness check. The address key is reused from the existing profile rather than regenerated with a different normalization function. That keeps the result tied to a specific, inspectable rule.

If the underlying extraction changes, rebuild the comparison and retain the old dataset as a dated version. A change in an address count can result from new registrations, removals, corrections or normalization changes. Compare the underlying rows and current operating evidence before attributing a change to a facility opening or closure. The checksums in the provenance file make that comparison possible.

Method and reproducibility

For each of the 13 organization CSVs, count unique NPI rows and distinct nonempty location_key values. location_key is the existing normalized primary street line plus five-digit ZIP. Subtract distinct primary keys from NPI rows; divide that difference by NPI rows for the percentage. Count an address collision when more than one NPI in the same legal-name profile has that key. Do not merge records across legal names or include secondary addresses in this particular comparison.

Limitations: A shared address is a matching candidate, not proof of a duplicate registration, a shared owner or a single operating site. The key does not independently verify the address, resolve suites or establish whether two departments have separate buying authority. These ratios are not measured duplicate rates or estimates for every pharmacy or retail health business. The cohort is the 13 profiles already published by Provyx, selected rather than randomly sampled. Totals are sums of within-organization counts, not globally unique facilities. Operating status and corporate-family relationships require separate evidence.

This analysis and its prose were prepared with AI assistance. The 13 input CSV files are the downloads already attached to the organization profiles below. We read identifiers and ZIP codes as text, preserve legal names, and calculate each organization independently. The provenance file records every input SHA256 and row count. The analysis code is included with the reviewable source; the public CSV and JSON contain identical rows. The derived downloads contain aggregate organization-level counts and source-field summaries.

Underlying source: CMS NPPES dissemination files, identified in each profile as the September 2026 monthly release. The public CMS index was under maintenance when this analysis was prepared; we used the retained extracts and their recorded provenance, using the original recorded source vintage. CMS explains the NPI standard.

Source profiles and state drill-downs

• COSTCO WHOLESALE CORPORATION profile, source records and method
• CVS PHARMACY INC profile, source records and method
• GENOA HEALTHCARE LLC profile, source records and method
• LUXOTTICA OF AMERICA INC profile, source records and method
• NATIONAL VISION INC profile, source records and method
• NEW YORK CITY HEALTH AND HOSPITALS CORPORATION profile, source records and method
• NOVANT HEALTH MEDICAL GROUP, LLC profile, source records and method
• PUBLIX SUPER MARKETS INC profile, source records and method
• TOTAL RENAL CARE INC profile, source records and method
• WALGREEN CO profile, source records and method
• WALMART INC. profile, source records and method
• WAL-MART STORES EAST LP profile, source records and method
• WAL-MART STORES, INC profile, source records and method

Each profile links to its existing state page and filtered source records. Those pages provide the next level of detail; use them to inspect the records behind a legal-name distribution. Start with CVS Pharmacy Inc. in Texas, Novant Health Medical Group in North Carolina, or the organization research directory.