Federally excluded NPIs billing Medicare Part B by HCPCS code post-exclusion
H30a established that federally-excluded NPIs continue to bill Medicare Part B post-exclusion. H40 sharpens the signal from "billed Part B for $30K" to "billed HCPCS J0897 47 times from POS 02 (telehealth) in [year], strictly after the LEIE exclusion effective date" — the unit of work State Medicaid PI offices write recoupment letters against. Joins the federally-excluded cohort (active LEIE or SAM, score ≥ 1.5) against the CMS Medicare Physician & Other Practitioners by Provider AND Service file (one row per NPI × HCPCS × place-of-service × year). Strict-post-exclusion attribution uses the per-NPI leie_excldate / sam_active_date already carried by the cohort exporter (shipped 2026-05-14 with the H29/H30a/H30b/H32 retrofit). Per-state CSVs at `/api/v1/states/<state>/h40-excluded-partb-by-hcpcs.csv`.
Headline
194 of 6,840 currently-active federally-excluded NPIs billed Medicare Part B in CY 2023 across 1,686 (NPI, HCPCS, place-of-service) row combinations (est. $27,384,846 paid on 551,657 services). Of those, **4 NPIs were billing STRICTLY POST-EXCLUSION** across 57 (NPI, HCPCS, POS) rows (est. $1,045,336 paid on 66,445 services). Per-row detail is the unit-of-work State Medicaid PI offices write recoupment letters against — per-claim, not per-provider.
4 / 6.8K = 0.06%
unit: count
What this means
State Medicaid PI offices
Recoupment is per-claim, not per-provider. H40 is the per-(NPI, HCPCS, POS, year) working list your team builds the recoupment letter from. Pairs with H29 (Medicaid spending) so the same NPI shows up in both the state-side and federal-side audit trail with the procedure mix attached.
Regulators
High-risk procedure patterns surface that the aggregate file masks: POS 02 (telehealth) E/M codes by an excluded NPI, controlled-substance J-codes, expensive injectables. Each pattern has a defined fraud-recoupment posture and a distinct prosecutorial weight.
Researchers
Strict-post-exclusion filter applied at the (NPI, year) grain because CMS publishes year-level service totals — same temporal precision as H30a, with per-HCPCS, per-POS specificity added. Full-window sidecar available for sensitivity analysis.
Null hypothesis
Zero currently-LEIE/SAM-excluded NPIs appear as billing providers for any HCPCS code in any place of service in the CMS Medicare Physician & Other Practitioners by Provider AND Service file for any year strictly after their exclusion effective date.
Denominator
Federally-excluded cohort (active LEIE or SAM, score ≥ 1.5; ~8,619 NPIs nationally per the latest `high-risk-cohort-export.csv`) joined against every (Rndrng_NPI, HCPCS_Cd, Place_Of_Srvc, year) row in the by-Provider-AND-Service file (latest available release; ~10M rows / ~3 GB per service year). State attribution via NPPES practice state, same partitioning the H29/H30a/H30b/H32 streams use.
Data source
CMS Medicare Physician & Other Practitioners by Provider AND Service (data.cms.gov/provider-summary-by-type-of-service/medicare-physician-other-practitioners/medicare-physician-other-practitioners-by-provider-and-service). One row per (Rndrng_NPI, HCPCS_Cd, Place_Of_Srvc, year) with services, beneficiaries, charges, and Medicare-allowed/paid amounts. Strict-post-exclusion filter at the (NPI, year) grain since CMS publishes year-level totals (same caveat as H30a). Streaming-once partition pattern at `analysis/claims_sources/medicare_partb_by_hcpcs.py` (to be added) mirrors `medicare_partb.py`. Public domain (CC0-equivalent), 2-year publication lag.
Notes
Per-state CSV at /api/v1/states/<state>/h40-excluded-partb-by-hcpcs.csv carries one row per (NPI, HCPCS, place_of_service) combination. estimated_paid_total = Tot_Srvcs × Avg_Mdcr_Pymt_Amt (CMS publishes averages, not totals, in this file). Place_Of_Srvc is aggregated to 'F' (Facility) or 'O' (Office/non-facility) at file build time — claim-level POS codes (02 telehealth, etc.) are not directly recoverable. Source CSV streamed once and partitioned across all state cohorts in memory — same I/O pattern as H30a.