Federally excluded providers paid by Medicaid (HHS spending dataset)
Joins the AINPI federally-excluded cohort (active OIG LEIE or SAM.gov listings) against the HHS Medicaid Provider Spending dataset (2018–2024, NPI-keyed, public). Every match is a § 455.436 audit-referral candidate for the state where the spending occurred — and catches MCO-side exposures that AINPI's H26 4-payer sweep currently misses behind authentication walls. Virginia is the Phase 1 pilot state.
Headline
Of 6,840 currently-active federally-excluded NPIs across 52 state cohorts (per NPPES practice state), 1180 received Medicaid payments somewhere in T-MSIS 2018–2024 totalling $481,682,270 across 12,539,039 claim lines — but only 12 of those 1180 matches received payment STRICTLY AFTER the earliest of their LEIE / SAM exclusion-effective dates, totalling $3,249,457 in strict post-exclusion payments. The strict-filter result is the regulatorily significant signal — pre-exclusion billing reflects work the provider was legitimately authorized to do at the time, while post-exclusion billing is a direct 42 CFR § 455.436 audit-referral candidate. Source: HHS Medicaid Provider Spending dataset (2026-02-09 release). Two source-data limits: (1) the HHS file has no state-of-payment column, so paid amounts aggregate across every state Medicaid program that paid the NPI — state attribution is per NPPES practice state, not per state of payment; (2) NPPES deactivation is NOT used as an exclusion-effective date because it can be triggered by retirement / death / voluntary closure of practice.
12 / 6.8K = 0.18%
unit: count
What this means
Regulators
Direct § 455.436 audit-referral signal. State PI units can pull the per-state CSV and feed it into the MMIS reconciliation queue. Every row carries the directory-side context (entity type, NPPES status, exclusion source + date, top HCPCS codes) so MMIS triage doesn't over-index on entity-NPIs that mix wide procedure ranges into one identifier.
Payer ops teams
The HHS file aggregates fee-for-service and managed care. Matches in your state's slice are operationally yours to investigate even when the payment flowed through an MCO. The row-level context columns are the de-noise layer — read entity_type=2 and high top_hcpcs_codes diversity together before treating a paid_amount as comparable.
Researchers
Publication policy resolved 2026-05-14: paid amount with context, anchored in AINPI's directory-side priors (H10 NPPES match, H14/H15 duplicates, H22 endpoint liveness for the entity's parent organization where applicable). No state-comparative ranking; per-state slices only. KFF 2026-02-20 caveat about non-comparable procedure codes is addressed by the row-level top_hcpcs_codes column rather than by aggregating amounts upward.
Null hypothesis
Zero NPIs currently active on OIG LEIE or SAM.gov appear as Billing NPI or Servicing NPI in the HHS Medicaid Provider Spending dataset (2018–2024).
Denominator
Active LEIE rows with a populated NPI (~8,551) ∪ active SAM rows with a real NPI (~4,517), joined against every (Billing NPI, Servicing NPI) appearing in the HHS Medicaid Provider Spending dataset. State attribution via the source file's T-MSIS state code. Virginia subset comes from the existing 131-NPI cohort at `/api/v1/states/va-cohort-critical.csv`.
Data source
OIG LEIE + SAM.gov Public Extract V2 (already ingested as `cms_npd.oig_leie` and `cms_npd.sam_exclusions`) × HHS Medicaid Provider Spending dataset (2026-02-09 release, opendata.hhs.gov/datasets/medicaid-provider-spending — 238M rows, 7 columns: billing/servicing NPI, HCPCS, claim month, patients, claim lines, paid amount; no state-of-payment column). State-scoped CSV at `/api/v1/states/<state>/h29-excluded-paid.csv` carries one row per matched NPI with the AINPI directory-side priors (exclusion source, top HCPCS codes, billing/servicing axis, first/last paid month) anchoring the paid-amount headline. See `/smd-revalidation/cross-audit-roadmap` §10b for the per-row schema and `analysis/claims_sources/medicaid_provider_spending.py` for the pyarrow filter implementation.
Notes
All-states refresh. Per-state CSVs at /api/v1/states/<state>/h29-excluded-paid.csv carry one row per matched NPI with the directory-side context columns (top_hcpcs_codes, exclusion_source, billing-or-servicing axis, first/last paid month) needed to interpret the paid-amount headline. Source-file schema: 7 columns (billing/servicing NPI, HCPCS, claim month, patients, claim lines, paid amount), 238M rows, no state-of-payment column. State attribution comes from NPPES practice state, not the spending file. The HHS source aggregates fee-for-service, managed care, and CHIP; matches catch MCO-side exposures that AINPI's H26 4-payer sweep currently misses behind authentication walls. Disclosure timing: publish when available and high confidence (locked-in 2026-05-14 per roadmap §10). DMAS receives 5-business-day review courtesy on VA-attributed rows before each refresh — operational courtesy, not a publication gate.