A data-driven investigation of $1.09 trillion in Medicaid provider spending
Analysis performed by Claude Code under the guidance of Josh — February 14, 2026
Medicaid is the U.S. government health insurance program for low-income Americans, jointly funded by the federal government and individual states. Each state runs its own version of the program. In 2024, Medicaid covered approximately 90 million Americans, including many children with disabilities.
When a healthcare provider sees a patient, they bill using standardized codes:
Applied Behavior Analysis (ABA) is the primary treatment for autism spectrum disorder. ABA therapists work with patients (often children) on behavioral skills, communication, and daily living. Sessions typically run 2–6 hours per day, multiple days per week, billed in 15-minute increments. A 4-hour session generates 16 billing units.
ABA is delivered by BCBAs (Board Certified Behavior Analysts, master's-level) who design treatment plans and RBTs (Registered Behavior Technicians, entry-level) who deliver therapy under supervision.
This analysis examined 227 million rows of Medicaid billing data representing $1.09 trillion in payments from January 2018 through December 2024. We cross-referenced this with the NPPES provider registry, NUCC taxonomy codes, and the OIG List of Excluded Individuals/Entities. Our findings have been validated against publicly available enforcement actions, news reports, and federal audit results.
| Metric | Value |
|---|---|
| Total Medicaid spending analyzed | $1.09 trillion |
| Records | 227 million rows |
| Period | Jan 2018 – Dec 2024 (84 months) |
| Billing providers | 617,000 |
| Servicing providers | 1.6 million |
| Analyses performed | 32 |
| External validation sources | HHS OIG audits, DOJ enforcement, state AG actions, news reports |
ABA therapy has become one of the fastest-growing categories in Medicaid spending — and one of the most fraud-prone. In 2014, CMS effectively mandated that all state Medicaid programs cover comprehensive autism services. By 2022, every state covered ABA. Spending exploded:
The growth trajectory tells a remarkable story: from $2.2 billion in 2018 to $4.6 billion in 2024. But the national average masks extraordinary variation between states:
| State | 2018 Spending | 2024 Spending | Growth | Context |
|---|---|---|---|---|
| Pennsylvania | $1.8M | $436M | +24,100% | New ABA coverage mandate (Act 62); also rapid provider expansion |
| North Carolina | $7.6M | $309M | +3,960% | State projecting $639M by FY2026; cutting rates; judge froze cuts |
| Texas | $18.7M | $310M | +1,560% | Gov. Abbott ordered targeted ABA utilization review (Jan 2026) |
| Arizona | $25.6M | $89M | +248% | Peaked at $217M in 2022; dropped after enforcement actions |
| Minnesota | $60.7M | $130M | +114% | FBI raided 2 ABA providers; 85+ open investigations |
| Massachusetts | $359M | $183M | -49% | Only state with declining ABA spending |
Note on growth rates: Some growth reflects legitimate policy changes — Pennsylvania enacted its Autism Insurance Act (Act 62) and expanded Medicaid waiver coverage. North Carolina added ABA benefits. However, U.S. autism diagnosis rates grew approximately 50% over this period. Growth of 1,000–24,000% vastly exceeds any plausible increase in the patient population, and the OIG audit findings confirm that a significant portion of this growth includes improper payments.
Many Medicaid services are billed in 15-minute units. If an individual provider (not an organization) bills 100 units in a month, they're claiming to have worked 25 hours that month. We calculated the implied hours per day for every individual provider who billed time-based services as both the billing and servicing provider. Anyone exceeding 24 hours/day has crossed a physical impossibility threshold — no matter how the data is interpreted, one person cannot work more hours than exist in a day.
The dashed line at 24 hours marks the absolute physical limit. Every provider to its right is billing for more hours than exist in a day.
| Provider | State | Service | Peak Hrs/Day | Avg Hrs/Day | Months Active | Total Earned |
|---|---|---|---|---|---|---|
| Eric Lund | WI | ABA Therapy | 341 hrs | 91.3 | 72 | $54.8M |
| Kulmoris Joiner | MS | Personal Care | 151 hrs | 50.6 | 84 | $36.7M |
| Prosper Dzameshie | MN | Personal Care | 31.2 | 21.8 | 72 | $15.3M |
| Xue Li | NY | Personal Care | 27.3 | 17.8 | 36 | $10.1M |
| Jodie Lotti | MA | ABA Therapy | 27.0 | 10.5 | 27 | $6.6M |
| Kaying Vang | MN | Personal Care | 24.5 | 16.5 | 24 | $3.7M |
When we tested specifically for ABA therapy (code 97153), we found 24 providers with impossible or extreme hours. Of these, 12 are in Wisconsin — half of all flagged ABA providers nationwide. This concentration, combined with the OIG's finding of $18.5M in improper payments with a 100% error rate, points to a systemic state-level problem in Wisconsin's ABA billing oversight.
Wisconsin's ABA program presents the clearest evidence of systemic failure. The state's spending grew from $20.8M (2018) to $47.1M (2024), with 12 of 24 impossible-hours ABA providers nationwide located in Wisconsin. The HHS OIG audited the state and found:
Eric Lund alone accounts for more than the total OIG-identified improper amount, suggesting the $18.5M figure significantly understates the problem.
Indiana pays the highest per-patient amount for ABA therapy (code 97153) of any state: $7,132 per patient per month, more than double the next highest state and nearly 3x the national median. The HHS OIG audited Indiana and estimated at least $56 million in improper payments in 2019–2020 (with an additional $76.7M "potentially improper"), noting 97 of 100 sampled enrollee-months had deficiencies including services by uncredentialed staff and billing during nap time.
When Indiana Medicaid began covering ABA in 2016, it adopted a system paying providers 40% of whatever they billed (nicknamed "BILL40"). This created a perverse incentive: providers could inflate their charges and still receive 40% of the inflated amount. Indiana ABA spending grew from $21M (2017) to $611M (2023) — a 2,867% increase. Indiana has since reformed to a flat ~$68/hour rate comparable to neighboring states. A governor's working group recommended usage caps in November 2025.
Our analysis found additional red flags in Indiana:
Minnesota has become the national epicenter of Medicaid fraud concern. The Feeding Our Future scandal ($250 million in food program fraud) was the tip of an iceberg. The U.S. Attorney for Minnesota has estimated that fraud likely exceeds $9 billion across 14 high-risk Medicaid services.
Autism/ABA-specific fraud in Minnesota:
Flagged providers in our analysis:
Arizona experienced a massive behavioral health fraud crisis. Although centered on substance abuse treatment rather than ABA specifically, the dynamics are relevant. Arizona's Medicaid agency (AHCCCS) has suspended 300+ providers, and enforcement actions include:
In our data, Arizona shows ABA spending peaking at $217M in 2022 then dropping to $89M by 2024 — consistent with enforcement actions. We identified a cluster of triple-flagged entities:
In January 2026, Governor Abbott directed the Texas HHS Office of Inspector General to launch investigations into potential Medicaid fraud, specifically ordering a targeted utilization review of autism services with a report due June 2026. The Texas OIG opened 2,118 preliminary provider investigations in FY2023 alone, with 866 referred to the Attorney General.
Our analysis flagged several Texas H2014 (skills training) providers billing at 3.2–5.7x the state median:
| Provider | Price/Claim | State Median | Multiple | Total Paid |
|---|---|---|---|---|
| RX Health LLC | $316 | $55 | 5.7x | $23.4M |
| Winners Circle Group of Texas | $249 | $55 | 4.5x | $19.9M |
| Houston Circle of Hope Services | $254 | $55 | 4.6x | $14.7M |
| Safe Place Counseling & Consulting | $253 | $55 | 4.6x | $9.1M |
Note: Safe Place Counseling is CARF-certified (an independent healthcare accreditation), suggesting it is an operational provider. The pricing outlier may reflect legitimate managed care contract rates, but the 4.6x premium warrants review in the context of the Governor's investigation. A DOJ-confirmed case found a Texas H2014 provider billing 21x the peer average from a residential home — not an approved service location.
Both states show extraordinary ABA spending growth with legitimate policy explanations that partially — but not fully — account for the trajectory.
North Carolina: ABA payments grew from $122M (FY2022) to a projected $639M (FY2026). The state is actively cutting rates (3–8% reductions), though a judge temporarily froze the cuts. We found multiple single-person ABA companies billing $25M–$105M each in NC, which warrants scrutiny.
Pennsylvania: From $1.8M to $436M (2018–2024). Pennsylvania enacted Act 62 (autism insurance mandate) and expanded Medicaid waivers, which explains the coverage expansion. But the rate of growth (over 24,000%) vastly exceeds what coverage mandates alone would predict.
Within each state, Medicaid fee schedules are standardized. Providers receiving 3x+ the state median per claim may be upcoding, exploiting billing errors, or operating under special rate arrangements.
| Provider | State | Code | Their Rate | Median | Multiple | Total Paid | Notes |
|---|---|---|---|---|---|---|---|
| MS. G & Associates | CA | H2019 | $4,115 | $137 | 30.1x | $12.1M | Extreme outlier |
| Amego Inc. † Reassessed | MA | H2014 | $1,666 | $92 | 18.0x | $33.9M | 24/7 residential care; legitimate per-diem rate |
| A Better You Wellness | AZ | H2014 | $1,331 | $224 | 5.9x | $115.4M | Multi-flagged |
| RX Health LLC | TX | H2014 | $316 | $55 | 5.7x | $23.4M | TX investigation target |
| Piece by Piece Autism Center | IN | 97153 | $1,311 | $394 | 3.3x | $41.8M | OIG-audited state |
California uses a cost-based reimbursement model for specialty mental health services where counties receive reimbursement based on actual costs. This means LA County ($423/claim vs $137 median), Santa Clara County ($468), and other California entities billing H2019 at 3–7x the national median are reflecting California's high cost of living and administrative overhead, not necessarily fraud. California is transitioning to fee-for-service to address this. However, MS. G & Associates at 30.1x the median ($4,115/claim) is an extreme outlier even within California and warrants investigation.
Amego Inc. provides 24-hour residential care for individuals with autism and severe behavioral challenges. At $1,666 per claim, this reflects per-diem residential rates for high-acuity patients, not 15-minute skills training units. Residential care legitimately costs far more than outpatient sessions. While the 18x multiple appears alarming, it likely reflects a different service model rather than fraud.
Community Assistance Resources & Extended Services (CARES) has been an established behavioral health nonprofit in New York City since 2005, serving individuals with developmental disabilities and mental health conditions. New York's OPWDD community habilitation rates have been independently found "excessive" by the HHS OIG, reflecting a state-level rate-setting issue rather than individual provider fraud. The 3.9x pricing ratio reflects NY's high state-set rates.
Every healthcare provider registers with a specialty in the national registry. When we find providers billing codes completely outside their specialty, it suggests practicing outside their scope, NPI misuse, or fraudulent billing.
| Provider | State | Registered As | ABA Earnings |
|---|---|---|---|
| Mya Ndiaye | MI | Counselor | $12.3M |
| Scott Staszak | MI | Occupational Therapist | $11.1M |
| Jodie Lotti | MA | Developmental Therapist | $8.1M |
| LaShannon Pinkston | MS | Contractor | $1.5M |
| Eunice Karue | MA | Licensed Practical Nurse | $1.1M |
A "Contractor" is not any type of healthcare professional. An LPN (Licensed Practical Nurse) assists with basic patient care — ABA therapy is well outside nursing scope. Some states do allow non-BCBAs to bill ABA under supervision, which could explain counselors and occupational therapists appearing here. But contractors and LPNs have no legitimate pathway to billing ABA codes.
35 providers registered as acupuncturists in the national registry were billing non-acupuncture medical codes (primarily in New York and California), totaling $35.5 million. This may reflect outdated NPPES registrations rather than active fraud, but the combination of wrong-specialty billing and high dollar volumes warrants verification.
| Organization | State | Revenue Through Top Person | % of Total | Staff |
|---|---|---|---|---|
| Gateway Pediatric Therapy | MI | $108.1M | 100% | 1 |
| ABS Utah | UT | $193.2M | 99.8% | 2 |
| Comprehab LLC | NC | $105.6M | 95.8% | 3 |
| Mosaic Group | NC | $82.6M | 99.1% | 2 |
| Brett DiNovi & Associates | PA | $65.1M | 100% | 1 |
Gateway Pediatric Therapy billed $108 million in ABA therapy over 6 years, all attributed to one person. At $200/hour, this implies ~250 hours per day — a physical impossibility.
The OIG maintains a list of individuals and companies permanently banned from billing federal healthcare programs. Billing after exclusion is categorically illegal under 42 U.S.C. §1320a-7. We cross-referenced 82,714 excluded entities (8,473 with NPIs) against the claims data.
50 NPI matches were found, billing $243 million combined. 5 were confirmed billing after their exclusion date:
| Provider | State | Excluded | Post-Exclusion Billing | Amount |
|---|---|---|---|---|
| New Life Wellness Center | AZ | Nov 2024 | 2021–2023 | $66M |
| Cactus Wren Community Services | AZ | Mar 2025 | 2020–2022 | $10.3M |
| We Care Transportation | KS | 2010 | Through 2020 | $4.4M+ |
Ohio accounts for 6 of the top 15 excluded-provider matches, suggesting a state-level oversight gap.
Office visits come in five complexity levels, from 99211 (~$25, minimal) to 99215 (~$150, high complexity). Nationally, only 2.2% of established-patient visits are 99215. A provider billing 90%+ as 99215 is almost certainly "upcoding" — billing for a higher level of service than provided.
| Provider | State | Total Visits | 99215 Rate | National Avg | Revenue |
|---|---|---|---|---|---|
| Scottsdale Physicians Group | AZ | 37,425 | 93.9% | 2.2% | $6.3M |
| Gabriel Aljadeff | IL | 3,200+ | 92.4% | 1.2% (specialty) | $1.2M |
| Debra Balke | CA | 1,800+ | 100.0% | 4.7% (specialty) | $0.8M |
Scottsdale Physicians Group bills 93.9% of all visits as the highest-complexity code. For a general internal medicine practice, fewer than 5% of visits typically justify 99215. Billing every patient as maximally complex is a statistical near-impossibility in legitimate practice.
Benford's Law is a mathematical principle: in naturally occurring financial datasets, the leading digit is "1" about 30.1% of the time, "2" about 17.6%, and so on. Forensic accountants use deviations from this pattern to detect fabricated data, because humans making up numbers tend to distribute digits more evenly than nature does.
The full $1.09 trillion dataset matches Benford's expected distribution closely, confirming the data is largely genuine. The leading digit "1" appears 31.5% of the time (expected: 30.1%).
The chart shows the frequency of leading digit "1" in ABA claim payment amounts by state. The expected frequency (per Benford's Law) is 30.1%. States with significantly lower frequencies may have non-natural payment distributions.
| State | Digit "1" Frequency | Expected | Deviation | Interpretation |
|---|---|---|---|---|
| Arkansas | 12.7% | 30.1% | -17.4 pts | Extreme deviation; possible flat-rate or fabricated billing |
| Ohio | 19.6% | 30.1% | -10.5 pts | Large deviation with large sample (252K records) |
| Minnesota | 21.8% | 30.1% | -8.3 pts | Notable deviation; consistent with MN fraud findings |
| Florida | 25.0% | 30.1% | -5.1 pts | Moderate deviation |
Research on Benford's Law in healthcare fraud detection shows high false positive rates. Healthcare billing amounts often cluster within ranges defined by fee schedules, authorization caps, and unit-based pricing, which can cause legitimate deviations. A Marquette University study found that providers flagged by Benford's analysis turned out to have "abnormal billing practices" that were not fraudulent. Benford's deviations should be used as one of multiple signals, never as standalone evidence of fraud. That said, Arkansas's deviation (17.4 percentage points below expected) is extreme even accounting for these limitations.
An individual provider can only be in one place at a time. We found 11 individuals billing through servicing providers in 3–8 different states simultaneously:
| Provider | Home State | States Used | Total Earned |
|---|---|---|---|
| Stephen Cauley | GA | 8 (FL, VA, NY, PA, AL, CA, TN, TX) | $1.4M |
| Arthur Kaiser | CA | 6 (AZ, NJ, NV, TX, IL, MA) | $3.5M |
| Meetu Soni | NJ | 5 (VA, PA, NY, MA, OH) | $2.0M |
Some of these may reflect legitimate telehealth practices, but billing across 8 states as an individual is unusual and consistent with billing mill patterns.
Providers earning over $1 million from 20 or fewer patients may be exploiting a small number of beneficiaries or fabricating claims:
| Provider | State | Type | Total Earned | Patients | $/Patient |
|---|---|---|---|---|---|
| New Trails LLC | MO | Residential | $25.2M | 18 | $1.4M |
| R&M Welty Inc. | MN | Residential | $22.5M | 16 | $1.4M |
| Grace Reliant Healthcare | MO | Residential | $22.0M | 19 | $1.2M |
While 24/7 residential care for severe disabilities can legitimately cost $100K–$300K+ per year, $1M+ per patient over the dataset period should be verified.
Several initial findings turned out to have legitimate explanations upon further investigation. Removing these reduces false positives and focuses attention on genuine concerns.
| Finding | Initial Concern | Reassessment |
|---|---|---|
| Non-NPI servicing IDs ($7B+) | Mystery alphanumeric IDs routing billions | Not fraud State-assigned IDs for personal care workers without NPIs |
| Fiscal intermediaries (Tempus Unlimited, FreedomCare) | Single-code billing, massive volumes | Not fraud Doing exactly what they're designed to do: managing personal care attendant payroll |
| FQHCs flagged on 4 signals | Multi-signal fraud flags | Not fraud FQHC bundled per-visit rate (~$203) creates apparent pricing outliers; cost-based model is structurally different |
| Deactivated NPI billing (Easter Seals, NYC H+H) | Billing on "dead" credentials | Data issue Organizational restructuring; services delivered legitimately under successor entities |
| Curative Labs ($87.6M, COVID-era) | Suspicious new entrant | Not fraud Well-known COVID testing company; appearance/disappearance tracks pandemic timeline |
| Same-name duplicate providers | Clone provider scheme | Mostly not fraud Different people sharing a name (different specialties). Only identical-specialty, small-town pairs remain suspicious. |
| California county H2019 rates (3–7x) | Pricing outliers | Structural Cost-based county reimbursement model; CA transitioning to fee-for-service |
| CARES Inc. (NY, $231M at 3.9x) | Pricing outlier | Rate policy Established 20-year nonprofit; NY OPWDD rates found "excessive" by OIG at the state level |
| Northway Academy (MN, 12 NPIs) | Suspicious multi-NPI expansion | Corporate Owned by Sevita Health, a major national disability services company; multi-site expansion is standard practice |
| Amego Inc. (MA, 18x pricing) | Extreme pricing outlier | Residential 24-hour residential facility for severe autism; per-diem rates legitimately much higher than outpatient |
| # | Entity | Total Paid | Key Evidence | External Validation |
|---|---|---|---|---|
| 1 | Eric Lund (WI) | $77.3M | 341 hrs/day, 2,509 claims/patient | OIG WI audit: 100% error rate |
| 2 | Kulmoris Joiner (MS) | $41.1M | 151 hrs/day for 84 months | Matches DOJ prosecution patterns |
| 3 | Gateway Pediatric Therapy (MI) | $108.1M | All revenue through 1 person | — |
| 4 | Robin Boykin (MO) | $48.7M | "Homemaker" billing $580K/month | — |
| 5 | 5 OIG-excluded providers | $5.5M+ | Billing after federal ban | Categorically illegal |
| 6 | A Better You Wellness (AZ) | $115.4M | New entrant + spike + 5.9x pricing | AZ $2-3B fraud cluster confirmed |
| # | State | Issue | Scale | External Validation |
|---|---|---|---|---|
| 7 | Indiana | Highest ABA per-patient cost ($7,132/mo), pricing outliers | $735M on 97153 | OIG audit: $56M improper |
| 8 | Minnesota | ABA fraud cluster, Feeding Our Future connections | $9B+ estimated total | FBI raids, 85+ investigations |
| 9 | Wisconsin | 12 of 24 impossible-hours providers; systemic | $47M/yr ABA | OIG audit: $18.5M, 100% error |
| 10 | North Carolina | 3,960% ABA growth; single-person companies billing $25-105M | $309M/yr | State cutting rates |
| 11 | Texas | H2014 pricing outliers, rapid growth | $310M/yr | Gov. Abbott ordered investigation |
| 12 | Arkansas | Extreme Benford's Law deviation (digit "1" at 12.7%) | Statistical | — |
| # | Entity | Total Paid | Key Evidence |
|---|---|---|---|
| 13 | Piece by Piece (IN, 2 entities) | $71.3M | 3–4x state median, 98% single-person |
| 14 | Scottsdale Physicians Group (AZ) | $6.3M | 93.9% highest-complexity billing |
| 15 | MS. G & Associates (CA) | $12.1M | 30.1x median — even within CA's cost-based model |
| 16 | Prosper Dzameshie (MN) | $15.3M | 31.2 hrs/day, registered as "Contractor" |
| 17 | Riyo International Corp (MN) | $7.4M | $264K/patient, non-healthcare name |
| 18 | Stephen Cauley (GA) | $1.4M | Individual billing through 8 states |
| Source | Description | Size |
|---|---|---|
| CMS T-MSIS | Medicaid provider spending (monthly aggregates via HHS Open Data) | 227M rows, 2.8GB |
| NPPES | National provider registry (NPI, entity type, state, specialty) | 132MB |
| NUCC Taxonomy | Healthcare specialty codes | 884 categories |
| OIG LEIE | List of excluded individuals/entities | 82,714 entries |
32 analyses spanning data quality, physical impossibilities, pricing outliers, credential mismatches, temporal anomalies, network analysis, statistical forensics (Benford's Law), billing uniformity, geographic impossibilities, and ABA-specific deep dives across multiple states.
Findings were cross-referenced against:
All analyses were performed using DuckDB for query execution on the 2.8GB Parquet file. Joins between claims data and the NPPES registry used TRY_CAST for non-numeric NPI values. External validation was conducted through web research of flagged providers and industry context.