Fraud Risk

Where a year of Medicare billing ranks against providers later excluded for fraud.

Connecting to the models…

About the fraud risk model

This calculator scores a provider-year of Medicare billing against a national model of future exclusion for fraud by the Office of Inspector General (OIG). Only nine HBOT providers have ever been excluded, too few to train on, so the model learns from all Medicare providers and is applied to HBOT practices.

The result is a percentile, not a probability: how a profile ranks against 2019 to 2022 provider-years. It works best as a compliance mirror, showing which measures, such as payment growth, services per patient and charge-to-allowed ratio, look like those of providers later excluded.

Predicts: Rank of fraud-type OIG exclusion risk within three years, as a national percentile.
Model: EasyEnsemble: 10 LightGBM models on undersampled data, trained on 2015 to 2018 providers who never billed HBOT.
Validation: Trained on 2015 to 2016 and tested on 2020 to 2022, so no outcome is shared between training and testing: AUROC 0.84. Reviewing the top 1% caught 21% of future fraud exclusions; the top 5% caught 45%.
Limits: Labels reflect detected and prosecuted fraud. A high percentile is a reason to review, not evidence of wrongdoing.

Research demonstration only. Not legal or compliance advice. A risk percentile is not an accusation.