🤖 ML Fraud Detection

Machine learning model identifies Medicare prescribers with patterns consistent with confirmed fraud cases

Share:

⚠️ Important Disclaimer

ML predictions are statistical indicators only, not accusations of fraud. These providers exhibit prescribing patterns similar to confirmed fraud cases, but there may be legitimate medical explanations. High-volume pain specialists, for example, may appear flagged due to medically appropriate opioid prescribing. Always consider clinical context.

1,077,354

Providers Scored

4,183

ML Flagged (≥80%)

76.5%

Known Fraud Recall

83.0%

Model Precision

How the ML Model Works

Training Data

  • 281 confirmed fraud cases from the OIG LEIE exclusion list cross-matched to active Medicare prescribers
  • 1,077,354 total providers with ≥50 claims in 2023
  • 20 prescribing features per provider including opioid rates, costs, brand preferences, specialty-adjusted z-scores, and drug combination patterns

Model Performance

  • Precision: 83.0%83% of flagged providers match fraud patterns
  • Recall: 66.6% — catches 67% of known fraud cases
  • F1 Score: 0.738 — harmonic mean of precision and recall
  • • 5-fold cross-validation on held-out data

Model: BaggedDecisionTrees ensemble with 20 trees. Trained with oversampled fraud cases and reservoir-sampled negatives. See full methodology.

Risk Tiers

785 providers

Very High (≥95% ML confidence)

Strongest match to fraud patterns

1,215 providers

High (85-94% ML confidence)

Strong match to fraud patterns

0 providers

Elevated (80-84% ML confidence)

Notable pattern similarities

Highest ML Fraud Scores

Top 100 providers ranked by ML fraud probability. These providers were not in the LEIE exclusion list — they are new predictions.

ProviderML Score
Jerry FlynnVery High
Milla KarevVery High
Marcel HaulardVery High
Herman ChavisVery High
Jason HamVery High
Joseph ScottVery High
Vadim BaramVery High
Dumitru SandulescuVery High
Jackie MaxeyVery High
Jon BowenVery High
Freddy GatonVery High
Mario Vasquez AguilarVery High
George SmithVery High
Richard TrenbathVery High
Eugene O'heaVery High
Dennis DobardVery High
Marcel FilartVery High
Kurt KraftVery High
Michael BusmanVery High
Cecil HolbertVery High
Japheth HardingVery High
John WilliamsVery High
Warren StumboVery High
Sharon ColtonVery High
Caterina Goldberg-DunnettVery High
Evgueni RoudachevskiVery High
James ChudleighVery High
David DavisVery High
Terence FrinksVery High
Constance SweetVery High
Benjamin SmithVery High
Preston McdonnellVery High
Alphonsus FlanneryVery High
Brendan BagleyVery High
Naser SoghratiVery High
David GuillotVery High
Heywood GayVery High
Sonya GoodrichVery High
Dan SpringerVery High
Frank BixlerVery High
Andrew BoydVery High
William GammillVery High
Ralphie Rosario FelicianoVery High
Murugesan SiddhappanVery High
Nicholas MorrowVery High
Ilya FishmanVery High
Robert KentVery High
Elizabeth CrabtreeVery High
Travis NovingerVery High
Andrew PerryVery High
Ahmad WardehVery High
Harry McdonaldVery High
Tracy ChapmanVery High
Adolfo DulayVery High
Joshua CarpenterVery High
Shyam AkkulugariVery High
Mark CrickVery High
Christine LerbackVery High
Clyde GreenVery High
Charles MooreVery High
Robert ClampittVery High
Heather SergentVery High
Enrique RodriguezVery High
Steve WamplerVery High
Farooq AhmedVery High
Brent ArdoinVery High
James SinclairVery High
Robert SweetenVery High
Earl EdwardsVery High
Christian WeigelVery High
Harshul PatelVery High
Min GaoVery High
Angela JonesVery High
Jamie PepplerVery High
Kristen MasseyVery High
Ladonna GeorgeVery High
Timothy BaileyVery High
Christopher OldfieldVery High
Edward ReasonVery High
Jason GlowVery High
Magdaleno CardenasVery High
Daniel RenoisVery High
L C TenninVery High
Collins FomunungVery High
Harold SmartVery High
Michael SatchellVery High
Debra WilderVery High
Joseph PrattVery High
Timothy SimonVery High
Sunil SharmaVery High
Rosaria TaimouriVery High
Randy WoodsVery High
Joanna MulderVery High
Joseph RobertsVery High
William BoulwareVery High
Lawrence HanauVery High
Ashok KancharlaVery High
Suresh DidwaniaVery High
Armand MasongsongVery High
Harold RothVery High

Top Flagged Specialties

Family Practice823
Internal Medicine769
Nurse Practitioner188
General Practice70
Physician Assistant32
Geriatric Medicine21
Psychiatry15
Physical Medicine and Rehabilitation10

Top Flagged States

Pennsylvania130
Texas119
Georgia109
California108
Alabama98
Ohio96
Florida93
Tennessee84
North Carolina82
Illinois76

ML vs. Rule-Based Scoring

OpenPrescriber uses two complementary fraud detection approaches:

📋 Rule-Based (10-Component Score)

  • • Hand-tuned thresholds and z-scores
  • • Transparent and explainable
  • • Good at catching extreme outliers
  • View flagged providers →

🤖 Machine Learning (This Page)

  • • Trained on confirmed fraud cases
  • • Finds non-obvious pattern combinations
  • • Better at detecting subtle fraud
  • • Identifies providers rules miss

Data from CMS Medicare Part D Prescriber Public Use File, 2023. Fraud labels from OIG LEIE. ML predictions are statistical indicators, not accusations.