AINPI 2026-05-14 update — claims-side cross-audit shipped
For: AINPI subscribers Date: 2026-05-14 Methodology version: 0.6.1-draft Source data: 2026-05-08 CMS NDH + claims-side public datasets (CY 2023 / CY 2024 / Q1 2026)
What's new since last week
The May 8 update was about the directory — what changed between the April and May NDH releases. This update is about the money that flows past the directory.
In one week we shipped the full claims-side cross-audit roadmap: 8 new findings that join AINPI's directory-side cohort to Medicaid spending, Medicare Part B billing, Medicare Part D prescribing, NPPES-deactivated-still-billing, Open Payments industry payments, DMEPOS supplier records, nursing-home / hospice / HHA / hospital ownership disclosures, and NDH completeness against the Medicare Part B universe.
Two methodology corrections also shipped — both of them lifted earlier numbers off the floor or fixed a structural null. We document both below.
The 8 new findings
| Hypothesis | Slug | Headline (regulatory-significance framing) |
|---|---|---|
| H29 | /findings/excluded-paid-by-medicaid | 0 of 28 VA-cohort NPIs paid by Medicaid strictly post-exclusion in 2018–2024 (full-window: $8.5M to 28 of 125) |
| H30a | /findings/excluded-billing-medicare-partb | 0 of 8 VA-cohort NPIs billing Part B strictly post-exclusion in CY 2023 (full-window: 8 of 125) |
| H30b | /findings/excluded-prescribing-medicare-partd | 0 of 10 VA-cohort NPIs prescribing Part D strictly post-exclusion in CY 2023 (full-window: 10 of 125, 6 opioid prescribers) |
| H31 | /findings/deactivated-still-billing | 3 of 1,495 VA-state NPPES-deactivated NPIs billed Medicaid / Medicare strictly after their NPPES deactivation date |
| H32 | /findings/excluded-receiving-industry-payments | 198 of 8,619 LEIE/SAM-active NPIs received industry payments strictly post-exclusion in PY 2024 ($167K) — full-window: 350 ($3.8M) |
| H33 | /findings/dmepos-excluded | 0 of 63,988 DMEPOS suppliers (DY 2023) are on the LEIE∪SAM active set |
| H35 | /findings/nh-hospice-hh-ownership-flags | 0 confirmed-NPI / 1,779 candidate-demographic matches between SNF + hospice + HHA + hospital owners and federal exclusion lists (17 candidate-demographic for VA facilities) |
| H36 | /findings/ndh-completeness-gap | 99.99984% NDH completeness against Medicare Part B billers with material CY 2023 paid amounts (only 2 of 1,259,343 individual NPIs absent) |
The headline pattern is consistent: when federal exclusion takes effect, federal-program payment stops in the data. The strict-post-exclusion column reads "0" across H29, H30a, H30b, and the strict subset of H32 dominates the dollar volume. The system at the payment gate is mostly holding for the active cohort.
What remains visible — and is what state Medicaid PI offices should act on — is the directory-side problem:
- H10 (NPPES-deactivated still listed in NDH).
- H24 / H25 (federally-excluded still appearing in the NDH bulk export).
- H31 (3 of 1,495 VA-state NPPES-deactivated NPIs are still billing somewhere).
- H35 candidate-demographic tier (1,779 candidate matches across the four facility types nationally).
Methodology correction #1 — strict post-exclusion attribution
Earlier drafts framed H29's $8.5M Medicaid headline as "VA-excluded NPIs paid by Medicaid 2018–2024." Technically true, regulatorily incorrect — because most of that $8.5M was pre-exclusion legitimate billing, not § 455.436 violations.
The fix: the H23 cohort exporter now carries per-NPI leie_excldate, sam_active_date, and nppes_deactivation_date. Each downstream claims-side finding (H29 / H30a / H30b / H32) filters claim-month against the earliest exclusion date, and the strict-post-exclusion column became the regulatory headline. The full-window column stays as a sidecar for completeness.
For H32 specifically: full-window $3.8M includes the $3.08M FRANK ALEXANDER payment received before his exclusion took effect. Strict-post-exclusion gives the much smaller $167K — exactly the number a Medicaid PI office should treat as a payment-gate failure signal.
Methodology correction #2 — H35 Stage B
The first H35 release reported "0 demographic matches between owners of SNFs / hospices / HHAs / hospitals and the OIG LEIE active exclusion list." That zero was a structural null, not a true finding. Investigation revealed two source-data facts:
- The CMS All Owners files do not populate STATE - OWNER for individuals — 100% empty for TYPE='I' rows in the 2026-04-01 release. The v1 demographic match joined on this empty column, so the key never collided with LEIE's populated STATE.
- The All Owners files do not carry an owner-NPI column either. Owner identity is keyed on PECOS_ASCT_CNTL_ID (an internal CMS associate ID).
Stage B (shipped today) introduces the CMS Medicare Fee-For-Service Public Provider Enrollment File (PPEF, 2026-04-01, 2.98M rows, 2.47M individual NPIs) as the shared cross-walk:
- Tier 1 (confirmed_npi): PECOS_ASCT_CNTL_ID → NPI via PPEF, then check NPI against LEIE.NPI ∪ SAM.npi. Authoritative — same NPI, same person. Result: 0 nationally. This null is itself evidence: exclusion forces revocation of Medicare enrollment, so only 25 of 8,619 LEIE/SAM-active NPIs are still in PPEF, and 0 of those 25 are listed as owners.
- Tier 2 (candidate_demographic): (LAST, FIRST, FACILITY_STATE) demographic match against LEIE, with facility state resolved via PPEF ENRLMT_ID (100% lookup hit, replacing the structurally empty owner-state column). Result: 1,779 candidate-demographic matches nationally, 17 for VA-state facilities.
Tier 2 is the verification surface state survey agencies / DMAS work from. Every row carries LEIE-portal verification URLs.
The compounding-signal pattern
One name surfaces in five independent public-data joins now: BREWER, STEVEN (NPI 1801070313) — H23 (federally excluded), H24 (active on OIG LEIE), H26 (listed in Cigna's payer directory until recently), H29 (Medicaid paid full-window), H30a + H30b (Medicare full-window). Each individual signal is a low-priority flag. The cross-product is the high-priority triage target.
This is the design pattern of AINPI's claims-side audit. Single-source flags are noise; multi-source flags converge on real cases. The cross-audit roadmap and the per-NPI rollup are how that converging happens at scale.
What this means for state Medicaid offices
State Medicaid Director-letter responses (deadline 2026-05-23) need to demonstrate the five elements of CMS's strategy submission. AINPI now covers:
| § 455.436 element | AINPI coverage |
|---|---|
| NPPES database check | H10–H13 (NPPES match rate) |
| OIG LEIE database check | H24 + H29 / H30a / H30b / H32 / H35 cross-audit |
| SAM.gov database check | H25 + H29 / H30a / H30b / H32 / H35 cross-audit |
| SSA-DMF database check | Out of scope (restricted access) |
| Public-facing metrics for Element 2 | All 30+ findings, manifest at /api/v1/manifest.json |
VA-attributed deliverables for DMAS:
/api/v1/states/va-cohort-critical.csv— 131 federally-excluded VA NPIs (LEIE or SAM, score ≥ 1.5)/api/v1/states/va/h29-excluded-paid.csv— Medicaid paid amounts per cohort NPI/api/v1/states/va/h30a-excluded-billing-partb.csv— Medicare Part B/api/v1/states/va/h30b-excluded-prescribing-partd.csv— Medicare Part D, with opioid metrics/api/v1/states/va/h31-deactivated-paid.csv— NPPES-deactivated VA-state billers/api/v1/states/va/h32-excluded-industry-payments-va.csv— Open Payments/api/v1/states/va/h35-nh-ownership-flags.csv— Stage B two-tier nursing-home / hospice / HHA / hospital owner matches
All MMIS-ready. All carry LEIE / SAM / NPPES portal verification URLs. Every row's directory-side context columns travel with the claims-side number, so the headline is never read in isolation.
What's next
- H34 (POS-deactivated × NPPES-active) remains pre-registered with the blocker documented: the CMS POS files (iQIES, Hospital_and_other, CLIA) are CCN-keyed and do not carry NPI. PPEF doesn't carry CCN either — so the CCN ↔ NPI cross-walk needs an authorized CMS source. If anyone on this list has a line on that, please reply.
- Second-state pilot (per the roadmap §10b): South Carolina remains a candidate if the SC referral lands.
- Subscriber digest cadence stays release-driven. Next CMS NDH bulk-export drop will trigger the next release-update report.
Read the data
- Findings index: https://ainpi.dev/findings
- Per-finding pages: each links to its JSON, per-state CSV, and a reproducibility note.
- Manifest: https://ainpi.dev/api/v1/manifest.json
- Methodology: https://ainpi.dev/methodology
- Virginia briefing: https://ainpi.dev/briefings/va
- SMD-revalidation alignment: https://ainpi.dev/smd-revalidation
- Cross-audit roadmap with all the decisions in §10: https://ainpi.dev/smd-revalidation/cross-audit-roadmap
— Eugene Vestel, FHIR IQ