Provider data landscape · One cell per state × specialty
Where the federal provider directory is accurate, and where it is not
Every state and specialty in the CMS National Provider Directory, scored on six dimensions of accuracy. One tile is one state and specialty. Tile size is the number of active practitioners. Tile color is whichever dimension you pick, and the layout holds still when you switch, so you can watch one metric at a time. Click a tile to check its providers against NPPES, the OIG exclusion list, and SAM.gov. The scoring maps to the REAL Health Providers Act. For excluded providers by state, open the map.
Latest finding · measured
The role gap is a Medicare-billing gap
- 77.9%
- Advanced practice with a role
- 526 of 12,465
- Pharmacy with a role
- 230,837
- PA practitioners measured
Directory coverage is not evenly spread across the professions it appears to describe. It tracks who bills Medicare.
Worked example · one state, end to end
Can software actually reach your clinician?
- 230,837
- PA practitioners traced
- 43.7%
- have an organization
- 18.7%
- reach an endpoint
Pennsylvania traced the whole way through: practitioner to organization to location to endpoint to EHR vendor, with a county map you can zoom and a named list of the organizations nothing public reaches.
Each box is one specialty. Cells inside it are states.
Completeness. Share of provider records carrying name, NPI, address and a contact method. These are the § 6220 fields the NDH schema can actually hold. New-patient acceptance, ADA accessibility and telehealth have no home in the published profile; language has one and is empty, so it is measured separately.
How to read this
- Spatial layout does not change when you flip layers: only color animates. The same cell sits in the same place, so you can learn the geography once and watch each metric move across it.
- Area = scale. A large California allopathic-physician cell carries more practitioners than the entire Vermont workforce; the treemap encodes that directly.
- Cells with fewer than 25 practitioners are suppressed to protect against PHI risk on small populations and to keep the visual readable.
- Color is normalized per layer to a constant diverging scale (rust = worse, blue = better; chosen so it survives colour-vision deficiency and greyscale printing, which a red-to-green scale does not). Higher completeness, agreement, reachability, integrity, and specialty validity are better; lower median update days are better.
Methodology & data lineage
Each metric is computed by pre-aggregation in BigQuery (analysis/landscape.py) and emitted as a typed JSON file: /api/v1/landscape.json. External consumers, regulators, and researchers can pull the same file as the visualization. Methodology version: 0.7.3-draft · Release: 2026-08-20 · Generated: 2026-08-22T13:04:05.655684+00:00.
Per-dimension methodology references: completeness · cross-source agreement · currency · reachability · integrity.