# Physician Employment Intelligence — Ground-Truth Reconciliation

Reconciliation of Verzi-derived physician counts (from CMS PECOS reassignment
joined to the Verzi physician_employer_systems crosswalk) against
publicly-stated counts from the systems themselves.

**Generated:** 2026-06-10
**Method:** Public count via system's own about/fact pages; Verzi count via
`GET /systems/{system_id}` total_distinct_npis (sum of distinct NPIs across
all classified PAC IDs rolling up to the system).

## Headline Comparison

| System | Public Count | Verzi Count | Δ vs Public | Public Label / Notes |
|---|---:|---:|---:|---|
| Mayo Clinic | 7,300 | **11,028** | +51% | "physicians + scientists" — excludes residents/fellows in the public count |
| Cleveland Clinic | 6,600 | **7,833** | +19% | "salaried physicians and researchers" |
| Kaiser Permanente | 25,000 | **30,473** | +22% | Permanente Medical Groups across 8 regions |
| Northwell Health | 7,300 employed / 13,500 credentialed | **8,371** | +15% vs employed / -38% vs credentialed | Post-Nuvance merger May 2025 |
| Mass General Brigham | 7,500 | **11,732** | +56% | System-wide; includes APPs and research staff in our count |
| Intermountain Health | 4,600 | **5,260** | +14% | Public count already bundles APPs — close match |
| Sutter Health | 12,000+ | **6,296** | **-48%** | UNDER-COUNT — crosswalk gap |
| UPMC | 5,000 employed / 6,400 affiliated | **9,776** | +53% vs affiliated | Our count likely captures faculty + community in same rollup |
| NYU Langone Health | 3,000 (FGP only) | **5,973** | +99% | PECOS catches voluntary attendings beyond FGP |
| Northwestern Medicine | 5,400 | **4,558** | -16% | Close match, slight under |

**Mean absolute deviation vs employed counts: 35%**
**Median absolute deviation: 22%**

## Interpretation

PECOS reassignment is the authoritative Medicare-billing-relationship signal,
but it is NOT the same population as a system's "employed physician" headcount
for three reasons:

1. **Inclusion drift.** Public "physician" counts almost always include some
   mix of MDs, DOs, fellows, residents, and APPs without clearly stating which.
   Our `?role=physician` filter strictly limits to NUCC taxonomy 207/208
   (MD/DO), which understates total clinical headcount but is the cleanest
   physician-only comparator.

2. **Voluntary attendings reassign too.** Academic medical centers (NYU
   Langone, Mayo, MGB, UPMC) have large voluntary attending populations who
   bill through the system's PAC but are not "employed" in the W-2 sense.
   This systematically inflates our number relative to the system's stated
   employed-physician count, in proportion to the size of the voluntary
   network.

3. **Captive-group gaps.** Where our count is *lower* than the public number
   (Sutter, Northwestern), the cause is crosswalk gaps — a captive medical
   group's PAC ID is not yet mapped in `physician_employer_systems` to the
   parent system. Sutter is the clearest case: 12,000 stated vs. 6,296 in our
   data means we are missing ~half the Sutter affiliated medical foundations
   in the crosswalk.

## Action items

| Priority | System | Action |
|---|---|---|
| HIGH | Sutter Health | Audit unmapped Sutter-related PAC IDs (PAMF, EBMF, SMG, SHCS, SVH, etc.). Crosswalk likely covers only the named hospital PAC. |
| MED | UPMC | Decompose 9,776 into faculty plan vs community physicians using employer_type tagging. |
| MED | NYU Langone | Add second system_id `NYU_VOLUNTARY` (or sub-tag) to distinguish FGP from voluntary attending population. |
| LOW | MGB / Mayo | Re-run role=physician filter on rollups and document the post-filter delta in the API response. |
| LOW | Northwell | Verify post-Nuvance PAC IDs flagged in crosswalk (migration 027 captured Nuvance as separate; should we re-parent to NORTHWELL? Depends on integration date.) |

## Methodology recommendation for clients

When pitching the Physician Employment Intelligence product, lead with PECOS
as the **authoritative Medicare-billing roster**, not a system HR roster.
Frame the use cases that favor PECOS:

- **Consolidation tracking:** PECOS captures every reassignment relationship,
  so it surfaces PE-platform and payer-vertical employment that a system's
  own about page never mentions.
- **Cross-system overlap:** A physician reassigning to *both* a hospital
  system PAC and a PE-backed group PAC is a relationship neither party
  publishes. PECOS is the only national source that surfaces this.
- **Geographic precision:** PECOS rows have per-org practice locations from
  DAC; system HR rosters are usually national totals.

Push back on apples-to-apples count comparisons. Use the `role=physician`
filter to bound the comparison and document the methodology delta in any
client deliverable. The ±20-50% variance we see vs. public counts is
explainable and consistent across systems.

## Sources

| System | Source |
|---|---|
| Mayo Clinic | https://en.wikipedia.org/wiki/Mayo_Clinic — Mayo fact sheet (2022) |
| Cleveland Clinic | https://my.clevelandclinic.org/about/overview/who-we-are/facts-figures |
| Kaiser Permanente | https://about.kaiserpermanente.org/who-we-are/fast-facts |
| Northwell Health | https://www.northwell.edu/about-northwell |
| Mass General Brigham | https://www.massgeneralbrigham.org/en/about |
| Intermountain Health | https://intermountainhealthcare.org/ |
| Sutter Health | https://www.sutterhealth.org/about-us/sutter-health-overview |
| UPMC | https://www.upmc.com/about/facts |
| NYU Langone | https://nyulangone.org/ |
| Northwestern Medicine | https://www.nm.org/about-us |