One NPI, up to six Organization records: how orgs multiply in the federal directory

Practitioner records in the CMS National Provider Directory deduplicate cleanly: 7,441,211 NPIs, zero excess rows. One person, one record.

Organizations are a different story. On the 2026-05-08 release:

  • 1,999,118 unique organizational NPIs
  • 70.6% of them map to more than one Organization resource
  • 1,413,126 excess Organization rows beyond one-per-NPI
  • worst case: six separate Organization records for a single NPI

What multiplication looks like

The same organizational NPI appears as several Organization resources, typically with the same name but different addresses, or slight name variants at the same address. These are not different legal entities. They are the same organization represented once per practice location, or once per source system that fed the upstream record.

Normalizing by (name, state, city) instead of NPI shows the same pattern from the other side: the duplication is structural, not a handful of typos.

Why it matters

  1. "How many organizations are in the directory?" has two answers that differ by 1.4 million. Resource count says 3.4 million. Unique-NPI count says 2.0 million. Any statistic built on the Organization table needs to declare which denominator it used.

  2. Joins fan out silently. A practitioner-to-organization join through PractitionerRole multiplies wherever the org NPI has several resources. Aggregates built without deduplication overcount affiliations, locations, and network sizes.

  3. The asymmetry is the finding. The same pipeline that keeps 7.4 million practitioner records perfectly deduplicated produces 70.6% multiplication on organizations. That points at how the upstream sources model facilities versus people, not at random noise.

Verify it yourself

This is finding H14/H15 (duplicate detection). The compute script and the per-release numbers are public, and the raw NDH files are downloadable from CMS.