Provider Directory Accuracy Index — Q2 2026
Cross-source field-level agreement between CMS NPPES, CMS Care Compare, and PECOS for every NPI present in ≥2 sources — aggregated to national, state, and specialty scope. Quarterly index; Q2 2026 issuance.
Contents · 7 sections
Executive Summary
- The Provider Directory Accuracy Index measures cross-source agreement between three federal databases for every NPI present in at least two sources. A high score indicates consistent information across databases; a low score flags potential ghost-network or stale-data risk.
- Scored fields: practice address (city + state), primary specialty taxonomy, organizational affiliation (CCN), and phone number. Each field is scored independently; the composite is the average of fields with ≥50% NPI coverage.
- National score, state leaderboard, and specialty breakdown are published at /research/provider-directory-accuracy and downloadable as CSV.
- This index publishes quarterly — Q1 in March, Q2 in June, Q3 in September, Q4 in December — generating four earned-media news hooks per year.
- All data sourced from federal public records. Fonteum does not independently inspect, assess, or rate any provider.
At a glance — for journalists, researchers, and AI agents
What this dataset covers
- Cross-source field-level agreement between NPPES, Care Compare, and PECOS
- National accuracy score
- Per-state accuracy scores and leaderboard
- Per-specialty accuracy scores
- Quarterly trend (when multiple periods available)
What this dataset does NOT cover
- Ground-truth provider accuracy (only cross-source agreement)
- Individual provider-level scores (aggregate only)
- Payer-specific directory accuracy
- Historical accuracy before 2026-Q2
- International providers
Sources
- CMS NPPES
- CMS Care Compare
- CMS PECOS
Snapshot date: Q2 2026
Dataset scope · Snapshot June 15, 2026
Includes: the healthcare-provider records this study covers, each tracing to a dated public-record source named in the citation footer. Does not include: providers outside the source named for this study, or records not present in that source at the snapshot date. Counts describe this Fonteum healthcare-provider dataset — not a representative census of the U.S. healthcare workforce.
Key findings
Overview
The Provider Directory Accuracy Index (PDAI) answers a simple question: how consistently do three federal provider databases describe the same providers?
Why it matters: CMS fines health plans for inaccurate provider directories under the No Surprises Act. State AGs pursue insurers over "ghost networks" — listed providers who no longer accept the plan. Journalists report on the problem every quarter. This index quantifies it.
What it is not: A ground-truth accuracy measurement. If a provider's address is wrong in the same way across all three federal sources, the PDAI scores it as "accurate." Cross-source agreement is a necessary, not sufficient, condition for directory accuracy.
Methodology summary
All computation is performed by Methodology · pdai/v1.
Step 1 — NPI intersection: Include only NPIs present in ≥2 of the three sources. NPIs in only one source have no comparison partner.
Step 2 — Pairwise comparison: For each NPI, compare all source pairs (NPPES × Care Compare, NPPES × PECOS, Care Compare × PECOS when applicable).
Step 3 — Field agreement rate: Count exact-value matches / total comparisons per field.
Step 4 — Coverage filter: Fields where <50% of NPIs have a non-null value in ≥2 sources are excluded from the composite.
Step 5 — Composite score: Simple average of coverage-passing fields.
Limitations
See the Limitations section below. Most important: agreement ≠ ground truth.
Limitations
- Cross-source agreement is a proxy for accuracy — fields that are wrong the same way across all sources score as accurate.
- Address granularity is city + state only in pdai/v1; ZIP-level and street-level matching is deferred.
- Specialty codes are matched exactly; taxonomy synonym expansion not performed in v1.
- Snapshot timing differs across sources — NPPES, Care Compare, and PECOS have different update cadences.
- Type-2 (organizational) NPIs excluded from pdai/v1; scored separately in a future version.
Methodology
Read the full methodology
Sources: CMS NPPES (bulk download, Tier-2), CMS Care Compare (bulk download, Tier-2), CMS PECOS PPEF (bulk download, Tier-2).
Unit of analysis: Individual NPI (Type-1 providers only in pdai/v1).
Fields scored: practice_address (city + state normalized lowercase), primary_specialty (NUCC 10-digit taxonomy code exact match), org_affiliation (6-char CCN or organizational NPI), telecom (10-digit phone digits only).
Minimum sample: 100 matched NPIs per scope. Scopes below threshold return no score.
Composite: Simple average of coverage-passing field rates (≥50% NPI coverage required per field).
Cadence: Quarterly cron, 09:00 UTC on the 15th of March, June, September, and December.
Methodology version: pdai/v1. Pinned constant in src/lib/research/accuracy-index/scoring.ts.
Sources: CMS NPPES (bulk download, Tier-2), CMS Care Compare (bulk download, Tier-2), CMS PECOS PPEF (bulk download, Tier-2).
Unit of analysis: Individual NPI (Type-1 providers only in pdai/v1).
Fields scored: practice_address (city + state normalized lowercase), primary_specialty (NUCC 10-digit taxonomy code exact match), org_affiliation (6-char CCN or organizational NPI), telecom (10-digit phone digits only).
Minimum sample: 100 matched NPIs per scope. Scopes below threshold return no score.
Composite: Simple average of coverage-passing field rates (≥50% NPI coverage required per field).
Cadence: Quarterly cron, 09:00 UTC on the 15th of March, June, September, and December.
Methodology version: pdai/v1. Pinned constant in src/lib/research/accuracy-index/scoring.ts.
Technical appendix
Show technical details · script paths · field names
Scoring module: src/lib/research/accuracy-index/scoring.ts
Compute orchestrator: src/lib/research/accuracy-index/compute.ts
Inngest cron: src/inngest/functions/provider-directory-accuracy-quarterly.ts
Migration: migrations/2026_05_11_provider_directory_accuracy.sql
Methodology version constant: PDAI_VERSION = 'pdai/v1'
Open for the script paths, raw dataset filenames, and per-field aggregation rules behind this snapshot. Reader-facing methodology above already covers source, date, and limitations.