Home/Research/Medicaid Fraud Analysis — Part II
Public Data Analysis · Part II

Medicaid Fraud Detection:
$1.09 Trillion, 617K Providers,
34,710 Red Flags

Multi-dimensional analysis of $1.09 trillion in Medicaid provider spending across 617,503 billing providers over 84 months (2018–2024). Combining statistical z-scores, Isolation Forest ML, and geographic clustering to identify 34,710 high-risk providers accounting for 43% of all spending.

227M records617K providers10,881 HCPCS codes84 months (2018-2024)8 fraud dimensionsIsolation Forest ML15 geographic clusters

Total Paid

$1.09T

2018–2024

Total Claims

18.8B

Individual claims

Providers

617K

Unique billing NPIs

Flagged (3+)

34,710

43% of all spending

02

Top Billing Providers

The 10 largest billing providers account for a disproportionate share of total Medicaid spending. Organizations dominate at 96.2% of all spending, while individual providers account for just 3.1%.

Entity Type Breakdown

Organization
96.2%
Individual
3.1%
Unmatched
0.7%

Top 10 Billing Providers by Total Spending

#ProviderStateTotal PaidAvg/Claim
1Public Partnerships LLCNY$7.18B$79.95
2LA County Dept of Mental HealthCA$6.78B$219.57
3Tempus Unlimited, Inc.MA$5.57B$87.71
4ModivCare Solutions, LLCCO$3.09B$28.71
5Freedom Care LLCNY$3.03B$137.68
6GuardianTrac LLCMI$2.68B$75.47
7TN Dept of IDDTN$2.60B$158.79
8AL Dept of Mental HealthAL$2.25B$1,156.32
9Consumer Direct Care Network VAMT$2.11B$95.02
10County of Santa ClaraCA$1.73B$371.23

Notable: Alabama Dept of Mental Health averages $1,156 per claim — significantly higher than peers — warranting further investigation.

03

Where the Money Goes: Top Procedure Codes

Top HCPCS Codes by Total Spending

CodeDescriptionTotal ($B)Avg/ClaimClaims
T1019Personal care services$122.7B$111.521.1B
T1015Clinic visit/encounter$49.2B$152.54322M
T2016Habilitation (residential)$34.9B$505.8769M
99213Office visit (est. patient)$33B$43.18764M
S5125Attendant care services$31.3B$78.74398M
99214Office visit (detailed)$29.9B$59.53503M

Most Overpriced Procedures (Avg Cost per Claim)

J23263 providers · 1,353 claims
$92,159
J14261 providers · 2,062 claims
$31,833
J717013 providers · 10,264 claims
$24,069
J14281 providers · 4,411 claims
$20,923
J235068 providers · 24,420 claims
$16,045

Notable

Personal care services (T1019) alone account for $122.7 billion (11.2% of all spending), making it the single largest category and a high-priority target for fraud auditing. High-cost injection/infusion codes (J-codes) with very few providers are prime candidates for overcharging.

04

The Billing/Servicing NPI Gap

38.6% of all Medicaid spending involves a billing entity different from the servicing provider. While many represent legitimate billing arrangements, systematically high mismatch rates can indicate shell billing companies, kickback arrangements, or billing fraud.

Same NPI

61.4%

$671.1B

Different NPI

38.6%

$422.5B

Spending Split

Same NPI: 61.4% ($671B)Different NPI: 38.6% ($423B)

Extreme Claims-per-Beneficiary Outliers

1952915910

Saint Louis, MO

497

claims/beneficiary

1417409509

Houston, TX

261

claims/beneficiary

1144989351

Detroit, MI

235

claims/beneficiary

The 99th percentile for claims per beneficiary is 24.6. These providers exceed that threshold by 10–20x, indicating potential phantom billing or extreme upcoding.

05

Multi-Layered Fraud Detection

This analysis employs five independent detection methods: statistical z-scores across 6 dimensions, Isolation Forest machine learning (200 estimators, 2% contamination), geographic anomaly detection, composite fraud scoring, and multi-flag intersection across 8 binary indicators.

Multi-Flag Distribution

0 flags
378,006Low
1 flags
161,456Low
2 flags
31,779Moderate
3 flags
14,344High
4 flags
14,947High
5 flags
2,723Very High
6 flags
2,576Very High
7 flags
118Extreme
8 flags
2Extreme

Providers with 3+ Flags

34,710

5.7% of all providers

Their Combined Spending

$470.8B

43% of total Medicaid spending

Isolation Forest ML Results

12,119

providers (2.0%) classified as anomalous

Trained on 19 features including geographic deviations and spending volatility. Correlation with statistical z-scores: r = 0.43–0.53, validating that both methods detect similar patterns.

Top 10 Highest-Risk Providers

#NPILocationTotal PaidScoreKey Anomaly
11073569034Irving, TX$27.7M75.34$83,133 avg/claim (3,004x geographic avg)
21952915910Saint Louis, MO$717K33.25497 claims/beneficiary
31376609297Stoughton, MA$5.57B32.9063.5M claims, 19 claims/benef.
41861568107Berwyn, PA$35.0M24.43$29,034 avg/claim
51831567858Columbia, SC$13.3M24.40$31,680 avg/claim
61124494059Nashville, TN$781.0M23.02$857 avg/claim, 911K claims
71568936557Milton, WA$678K18.19$22,605 avg/claim (30 claims)
81235261652San Leandro, CA$30.6M18.00$19,492 avg/claim
91013030808Merced, CA$47.6M17.70$18,464 avg/claim
101417409509Houston, TX$549K17.04261 claims/beneficiary

Provider #1 (Irving, TX) has a fraud score of 75.34 — more than double #2. Their avg per claim of $83,133 is 3,004x the geographic average. Provider #2 (Saint Louis, MO) billed 497 claims per beneficiary — a classic indicator of phantom billing.

06

State-Level Geographic Analysis

Top 5 States by Total Spending

#StateProvidersSpendingAvg/Claim
1NY59,321$144.8B$86.04
2CA52,782$129.4B$50.83
3TX31,254$56.2B$50.5
4MA19,346$56.0B$70.92
5NJ15,992$47.0B$59.87

Most Anomalous States

#1

Minnesota

Highest claims/beneficiary (4.39), 66% mismatch rate

1.60

anomaly score

#2

Alaska

Highest avg/claim ($158.56), highest $/provider ($3.32M)

1.56

anomaly score

#3

Maine

Very high avg/claim ($185.39), highest $/provider ($3.72M)

1.34

anomaly score

#4

US Virgin Islands

Extremely high avg/claim ($247.91)

1.10

anomaly score

#5

Arizona

High avg/claim ($184.28)

1.05

anomaly score

Key Finding

Minnesota is the most anomalous state overall, driven by the nation’s highest claims-per-beneficiary ratio (4.39) and a 66% billing mismatch rate. This finding is reinforced by Minnesota appearing in 4 separate geographic fraud clusters. Least anomalous: Illinois (-1.50), Puerto Rico (-1.32), Pennsylvania (-0.89).

07

Geographic Fraud Clusters

DBSCAN clustering on provider fraud feature vectors identifies 15 regional clusters of providers with similar anomalous behavior — potential coordinated fraud rings.

Major Clusters by Spending

StateProvidersSpendingKey CitiesMismatch
NC3,168$16.9BCharlotte, Raleigh, Greensboro20.5%
AZ1,873$15.1BPhoenix, Tucson, Mesa24.4%
MN1,754$13.6BMinneapolis, Saint Paul, Rochester40.7%
NM420$6.1BAlbuquerque, Las Cruces27.8%
ME417$5.9BPortland, Bangor, Lewiston29.2%

Highest Fraud Score Clusters (Investigation Priority)

MN-06 providers
8.49

Mahnomen, Farmington, Redlake

Small group, extreme scores

MN-13 providers
5.73

Ponemah, Moorhead, Bemidji

Northern MN rural cluster

MN-23 providers
4.57

Roseville, Saint Cloud, Rochester

100% mismatch rate

ME-13 providers
1.67

Waterville, South Portland, Augusta

Highest spike z-score (8.46)

Pattern

The small Minnesota clusters (MN-0, MN-1, MN-2) with 3–6 providers each and fraud scores of 4.57–8.49 are the highest-priority targets. Their small size and geographic concentration are consistent with organized billing schemes.

08

City-Level Hotspots

Top 10 Cities by % Providers Flagged (min. 20 providers)

Hialeah Gardens, FL
42%21/50
Caruthersville, MO
40.5%15/37
Palmview, TX
38.5%15/39
Brooklyn Center, MN
35.5%22/62
Ferguson, MO
35%7/20
Mendota Heights, MN
34.8%8/23
Laveen, AZ
34.8%16/46
Windsor, NC
34.6%9/26
Church Point, LA
34.5%10/29
Pembroke, NC
33.3%19/57

Geographic Patterns

South Florida (Hialeah Gardens 42%) remains a historically known Medicaid fraud hotspot. Minnesota (Brooklyn Center 35.5%, Mendota Heights 34.8%) has 3 cities in the top 20. North Carolina (Windsor, Pembroke) aligns with the state’s largest fraud cluster. Missouri (Caruthersville 40.5%, Ferguson 35%) shows rural/small-city hotspots.

09

Deactivated NPI Analysis

Deactivated NPIs in Data

1,641

$3.8B in spending

Full Dataset Deactivated

10,269

$7.1B · 126.5M claims

Flagged Deactivated NPI

1720471568

Brooklyn, NY

$293.4M

fraud score: 4.89

Context

While some deactivated NPI spending may represent legitimate pre-deactivation billing, $7.1 billion flowing through 10,269 deactivated provider numbers demands systematic review.

10

Key Findings & Recommendations

Spending concentration

$470.8B

in spending (43% of all Medicaid expenditure) concentrated among just 34,710 providers (5.7%) with 3+ fraud flags. This disproportionate concentration indicates systemic risk.

Minnesota

#1

Most anomalous state nationally. Highest claims/beneficiary (4.39), 66% NPI mismatch rate, 4 separate fraud clusters, and 3 cities in the top-20 flagged list. Small MN clusters (3–6 providers with scores 4.57–8.49) are consistent with organized fraud rings.

Extreme outlier

75.34

fraud score for Provider #1 (Irving, TX) — their average claim of $83,133 is 3,004x the geographic average. Provider #2 (Saint Louis, MO) billed 497 claims per beneficiary — a classic phantom billing indicator.

South Florida

42%

of providers flagged in Hialeah Gardens, FL — continuing the historically documented pattern of elevated Medicaid fraud in the region.

Deactivated NPIs

$7.1B

in spending by 10,269 deactivated provider NPIs. While not all fraudulent, the volume demands systematic review.

Billing mismatch

38.6%

of all spending ($422.5B) flows through billing entities different from servicing providers, creating opacity that facilitates fraud.

Recommended Investigation Priorities

Tier 1 — Immediate

2 providers with 8 flags · 120 providers with 7+ flags · Small MN clusters (avg scores >4.5) · Provider NPI 1073569034 (score 75.34)

Tier 2 — High Priority

Top 500 flagged providers · Deactivated NPIs with >$10M spending · Cities with >30% flagging rate · Providers with >100 claims/beneficiary

Tier 3 — Systematic Review

Minnesota statewide audit · ND/OK billing mismatch patterns · J-code billing (few providers, high costs) · Personal care services T1019 ($122.7B)

Bottom line

This expanded analysis confirms and deepens the findings from Part I. With 84 months of data and $1.09 trillion in spending, the patterns are unmistakable: a small fraction of providers handle disproportionate volumes, geographic clusters suggest coordinated schemes, and billions flow through deactivated or mismatched billing entities. Targeted auditing of fewer than 35,000 providers could address 43% of all Medicaid expenditure.

11

Methodology Note

This analysis uses CMS Medicaid Provider Utilization & Spending data and NPPES NPI Registry. The dataset contains 227,083,361 records spanning 84 months (2018-01 to 2024-12), covering 617,503 unique billing providers and 10,881 HCPCS procedure codes.

01

NPI enrichment

Spending data enriched with NPI registry data (state, city, entity type, org name, deactivation date, taxonomy) via LEFT JOIN. Match rate: 99.5% billing NPIs, 94.2% servicing NPIs.

02

Statistical z-scores

Per-provider z-scores across 6 dimensions: overcharging, over-utilization, NPI mismatch, low code diversity, geographic deviation (price & volume), and spending volatility.

03

Isolation Forest

200 estimators, 2% contamination rate, trained on 19 provider-level features. Captures non-linear fraud patterns that z-scores miss.

04

DBSCAN clustering

Applied within top 10 anomalous states on standardized fraud feature vectors to detect geographic concentrations of similar anomalous behavior.

05

Composite scoring

Weighted ensemble combining z-scores, Isolation Forest anomaly scores, geographic deviations, and NPI mismatch rates. Multi-flag thresholds: top 5% per dimension across 8 independent indicators.

Limitations

1

Statistical flags ≠ fraud

Anomalous patterns may have legitimate explanations (specialty providers, group billing structures, high-acuity patient populations). Each case requires individual clinical review.

2

No clinical context

The dataset lacks diagnosis codes, patient demographics, and clinical justification necessary for definitive fraud determination.

3

Ecological inference

Geographic clusters identify correlated anomalies, not proven coordination. Provider proximity does not prove organized fraud.

This analysis is for informational and research purposes only. Statistical anomalies identified here should not be interpreted as evidence of fraud without further investigation. Data sources: CMS Medicaid Provider Utilization and Payment Data, NPPES NPI Registry.