OPEN DATA ANALYSIS
Anomalies in $1.09 Trillion
of Medicaid Claims Data
On February 14, 2025, the HHS DOGE team open-sourced the largest Medicaid provider spending dataset in department history — 227 million rows of aggregated, provider-level claims data spanning January 2018 through December 2024. This analysis identifies providers exhibiting statistically anomalous billing patterns across three dimensions: year-over-year spending spikes, autism services billing surges, and claims-per-beneficiary outliers.
Data source: HHS Medicaid Provider Spending (open data release, Feb 2025) · 617,503 billing providers · 1,627,362 servicing providers · 10,881 HCPCS billing codes
Total Rows
227M
Provider-code-month records
Total Spending
$1.09T
Jan 2018 - Dec 2024
Providers Flagged
12
Across all 3 analyses
Billing Codes
10,881
HCPCS procedure codes
Year-over-Year Spending Spikes
Providers whose Medicaid billing tripled or more in a single year, starting from a baseline above $1 million. Growth of this magnitude — without proportional increases in beneficiary count — is a strong anomaly signal.
Community Assistance Resources & Extended Services Inc
NPI 1396049987 · New York, NY
2019
$77K
2020
$1.6M
20.9x YoY
2021
$112.6M
69.9x YoY
2022
$486.1M
4.3x YoY
2023
$297.6M
2024
$144.9M
CARES Inc — 70x Spike in One Year
NPI 1396049987 — Community Assistance Resources & Extended Services Inc, New York, NY.
In 2019, this provider billed Medicaid $77,059. By 2022, that figure was $486 million. The spending trajectory: $77K → $1.6M → $112.6M → $486M. That is a 6,309x increase over three years.
Their claims volume went from 586 to 1.2 million annually. Peak monthly beneficiaries grew from 29 to 3,037 — but the spending-per-beneficiary reached $160,086/year at peak, far exceeding norms for personal care services.
After peaking in 2022, spending dropped to $297M in 2023 and $145M in 2024 — a pattern consistent with either detection/enforcement or deliberate drawdown.
Tradition Choice — 52x Spike
NPI 1568910677 — Tradition Choice, Brooklyn, NY.
From 2019 to 2021, this Brooklyn-based provider billed under $800K per year. In 2022, billing jumped to $39.9 million(a 51.6x increase), then doubled again to $83.2 millionin 2023 and $83.7 million in 2024.
Beneficiary count went from 20 to 1,536 in a single year — a 77x increase. The sudden jump from micro-provider to $80M+ annual billing, sustained across years, is atypical.
Human Care Services — $1M to $90M
NPI 1730369760 — Human Care Services for Families and Children Inc, Brooklyn, NY.
First appeared in the dataset in 2021 billing $977K. By 2022: $89.7 million (a 91.8x increase). Beneficiary count went from 67 to 904. The per-beneficiary spend of $99,218/year is notably high for community habilitation services.
A Better You Wellness — 27x then Collapse
NPI 1023788031 — A Better You Wellness Center, LLC, Phoenix, AZ.
Billed $3.7M in 2021, then $99M in 2022 (a 26.8x increase) with only 309 beneficiaries. That is $320,527 per beneficiary in a single year. Spending then collapsed to $12.6M in 2023 — an 87% drop — suggesting either intervention or billing correction.
Americare — 18.6x Spike in Latest Year
NPI 1770513145 — Americare Certified Special Services, Inc., New York, NY.
Spent years at $3-5M annually. In 2024, billing jumped to $61.4 million — an 18.6x increase. Claims went from 38,950 to 266,458. This spike is happening now in the most recent data.
The Minnesota Autism Pattern
In Minnesota, large-scale autism diagnosis fraud was perpetrated through inflated billing of H2015 (community habilitation) and T1019 (personal care services). This analysis searches the national dataset for the same pattern: providers with explosive year-over-year surges in these specific billing codes.
What are H2015 and T1019?
H2015
Community habilitation services, waiver — per diem. Covers day-to-day support for individuals with developmental disabilities. Median claims-per-beneficiary across all providers: 256.6. 99th percentile: 3,352.
T1019
Personal care services, per 15 minutes. Covers assistance with daily living activities. Median claims-per-beneficiary: 747.7. 99th percentile: 2,287.
H2015 / T1019 Billing Surges by Provider
Year-over-year spending multiplier for autism/personal care codes. A provider going from $220K to $34.6M in one year (157x) is a statistical impossibility under normal operations.
Human Care Services — 157x H2015 Surge
NPI 1730369760 went from $219,957 to $34.6 million in H2015 billing in a single year (2021 → 2022). That is a 157.4x multiplier.
Claims went from 769 to 80,622. With 755 peak beneficiaries, that is 106.8 claims per beneficiary — well below the 99th percentile individually, but the spending growth rate is 157 standard deviations above the typical year-over-year change.
CARES Inc — 106x H2015 Surge
NPI 1396049987 went from $430K to $45.5 million in H2015 billing (2020 → 2021), then to $245.8 million (2021 → 2022).
This is the same provider that appeared in the spending spikes analysis. The H2015 code alone accounts for roughly half their total billing. At peak, they billed 191.8 H2015 claims per beneficiary — the 95th percentile is 1,650. Combined with the overall spending trajectory ($77K to $486M), this is the strongest fraud signal in the dataset.
Tradition Choice — 52x T1019 Surge
NPI 1568910677 went from $773K to $39.9M in T1019 billing (2021 → 2022), a 51.6x increase. At 188 claims per beneficiary, the volume itself isn't unusual — but the growth rate is. A legitimate provider does not go from $773K to $40M in personal care services overnight.
Public Partnerships LLC — 7x T1019 Surge
NPI 1417262056 — Public Partnerships LLC is a fiscal intermediary, not a direct care provider. Their T1019 billing went from $31.6M to $224.2M (7.1x) in 2018 → 2019, reaching 210.4 claims per beneficiary across 18,410 beneficiaries. As a fiscal intermediary, the anomaly may reflect downstream provider fraud being channeled through their billing infrastructure.
Claims-per-Beneficiary Outliers
When a provider bills thousands of claims per patient, it suggests either billing for services not rendered, unbundling services to inflate claims count, or systematically upcoding. This analysis flags provider-code pairs where the ratio exceeds 1,000 claims per beneficiary.
Claims-per-Beneficiary vs. Total Spending
Each dot is a provider-code pair. The further right, the more claims per patient. The higher up, the more total money. Dots far to the right represent physically implausible service volumes.
LA County DMH — 4,807 Claims per Patient
NPI 1699703827 — Los Angeles County Department of Mental Health, Los Angeles, CA.
For H2017 (community habilitation, per 15 min): 4,806.8 claims per beneficiary over 82 months with only 899 peak monthly beneficiaries. Total: $870.7 million. That works out to $968,475 per beneficiary and roughly 13.2 claims per patient per day for the entire dataset period. Each claim is a 15-minute unit, meaning they billed an average of 3.3 hours of community habilitation per patient per day, every day, for nearly 7 years.
The same provider also has 3,294 claims per beneficiary for H2015, totaling $2.73 billion. Combined H2015+H2017 billing: $3.6 billion with fewer than 4,000 peak patients.
Alabama DMH — $4.3M per Beneficiary
NPI 1982757688 — Alabama Department of Mental Health, Montgomery, AL.
For T2016 (residential habilitation, per diem): only 423 peak monthly beneficiaries but $1.83 billion in total spending over 84 months. That is $4,333,133 per beneficiary and $3,837.56 per claim.
At 1,129 claims per beneficiary over 7 years, this implies roughly 0.44 claims per day, which is plausible for daily residential care. The anomaly here is the per-claim cost ($3,838) and per-beneficiary cost ($4.3M), which are extreme outliers. For context, the median Medicaid cost for institutional residential care is approximately $200-400 per day — not $3,838.
Tempus Unlimited — 1,551 Claims/Beneficiary
NPI 1376609297 — Tempus Unlimited, Inc., Stoughton, MA.
The single largest T1019 (personal care) biller in the entire dataset: $5.46 billion over 84 months with 39,834 peak beneficiaries and 1,551.4 claims per beneficiary. At $88.41 per claim, the unit cost is reasonable. The question is volume: 1,551 claims per beneficiary over 7 years is 18.5 claims per beneficiary per month, or about 4.6 per week. Each claim is a 15-minute unit, so roughly 70 minutes per patient per week. This is within plausible range for a large fiscal intermediary.
Full Outlier Table
| Provider | Code | Total Paid | Claims/Bene | $/Beneficiary | $/Claim |
|---|---|---|---|---|---|
| Los Angeles County Department of Mental Health 1699703827 | H2017 | $870.7M | 4,806.8 | $968K | $201 |
| Los Angeles County Department of Mental Health 1699703827 | H2015 | $2.7B | 3,294.2 | $694K | $211 |
| 1356709976 1356709976 | T2033 | $1.0B | 2,332.3 | $928K | $398 |
| 1629436241 1629436241 | T2033 | $1.5B | 2,596 | $649K | $250 |
| County of Santa Clara 1528263910 | H2015 | $890.2M | 1,754.6 | $753K | $429 |
| 1962650622 1962650622 | T1019 | $1.6B | 1,705.3 | $251K | $147 |
| Public Partnerships LLC 1417262056 | S5126 | $2.4B | 1,670.6 | $139K | $83 |
| Tempus Unlimited, Inc. 1376609297 | T1019 | $5.5B | 1,551.4 | $137K | $88 |
| Alabama Department of Mental Health 1982757688 | T2016 | $1.8B | 1,129.1 | $4.3M | $4K |
Methodology & Caveats
How this analysis was conducted and what it does not tell you.
Data Source
The raw dataset is the HHS Medicaid provider-level claims file released on February 14, 2025. It contains 227,083,361 rows, each representing a unique combination of billing provider NPI, servicing provider NPI, HCPCS code, and claim month. Fields include total unique beneficiaries, total claims, and total amount paid.
Analysis Pipeline
The 11GB CSV was converted to a 2.7GB Parquet file using DuckDB with ZSTD compression. All queries were executed locally using DuckDB analytical SQL. Provider identities were resolved via the NPI Registry API (NPPES). No data was modified or filtered beyond the aggregation queries shown below.
Anomaly Detection Criteria
- Spending Spikes: Year-over-year spending increase of 3x or more, with a prior-year baseline of at least $1M and current-year spending of at least $10M. This filters out new providers and small-dollar noise.
- Autism Pattern: Year-over-year increase of 2.5x or more in H2015 or T1019 billing specifically, with a baseline above $100K and current spending above $5M.
- Claims Ratio: Total claims divided by peak monthly beneficiary count exceeding 1,000, with total spending above $5M. This is a lifetime ratio across the 7-year dataset, not a monthly figure.
Important Caveats
- Anomaly is not fraud. Statistical outliers may reflect legitimate operational changes: rapid expansion, new contracts, Medicaid policy changes, or data reporting adjustments. No accusation of fraud is made or implied.
- Fiscal intermediaries like Public Partnerships LLC and Tempus Unlimited aggregate billing for thousands of individual caregivers. Their high volumes may reflect their role, not malfeasance.
- Government agencies like LA County DMH and Alabama DMH may bill differently than private providers due to bundled service agreements and statewide contracts.
- Beneficiary counts in this dataset represent the maximum monthly count, not unique lifetime beneficiaries. Claims-per- beneficiary ratios are therefore upper-bound estimates.
- HCPCS code definitions vary by state Medicaid program. T1019 may cover different service durations (15 min vs. per diem) in different states.