Definitions for every term, metric, and abbreviation used across OpenMedicare. See also our methodology and data sources.
The maximum amount Medicare has determined is appropriate for a given service. This is typically less than the provider's submitted charge but more than the actual payment after cost-sharing.
A model accuracy metric measuring how well the algorithm distinguishes between fraudulent and legitimate providers. Our model achieves 0.83 AUC, meaning 83% discrimination accuracy. A score of 0.5 would be random chance; 1.0 would be perfect.
A person enrolled in and receiving benefits from Medicare. Also referred to as a Medicare patient. Each beneficiary has a unique identifier (not the same as the provider's NPI).
Intentionally submitting false or misleading claims to Medicare for payment. Includes upcoding, unbundling, phantom billing, and billing for services not rendered. A federal crime under the False Claims Act.
The federal agency within the Department of Health and Human Services that administers Medicare, Medicaid, and the Children's Health Insurance Program. CMS publishes the provider utilization data that powers this site.
A statistical range indicating the uncertainty around a fraud probability estimate. A narrower interval means more certainty. We report 95% confidence intervals for our risk scores.
The ratio between what a procedure actually costs to perform and what the provider bills. Extremely high ratios can indicate inflated billing, though they vary significantly by specialty.
An in-depth analysis of a specific flagged provider, examining their billing patterns, peer comparisons, geographic context, and specific anomalies that triggered the fraud detection model.
Medical equipment and supplies prescribed by providers and paid for by Medicare. Includes wheelchairs, oxygen equipment, diabetic supplies, and prosthetic devices. Historically a high-fraud category.
A provider who has been banned from participating in Medicare and other federal healthcare programs, typically due to fraud convictions, license revocation, or other misconduct. Listed on the LEIE.
Federal law (31 U.S.C. §§ 3729–3733) that imposes liability on persons who defraud the government. The primary legal tool for prosecuting Medicare fraud. Allows whistleblowers (qui tam relators) to file suits on behalf of the government.
A machine learning concept measuring which data signals contribute most to the model's predictions. In our fraud model, top features include markup ratio, services per beneficiary, total payment amount, and deviation from specialty peers.
The machine learning model's estimate that a provider's billing patterns match known fraud patterns, expressed as a percentage from 0–100%. This is a statistical risk indicator, not an accusation. High scores warrant further investigation.
Our curated list of 500 providers flagged by the machine learning model as having the highest probability of fraudulent billing patterns. Updated with each data release. Not an accusation — a starting point for investigation.
A standardized coding system used to identify specific medical services, procedures, and equipment for billing purposes. Level I codes are CPT codes for physician services; Level II codes cover non-physician services like ambulance rides and DMEPOS.
The Department of Health and Human Services Office of Inspector General, responsible for fighting fraud, waste, and abuse in Medicare, Medicaid, and other HHS programs. Maintains the LEIE and publishes fraud case outcomes.
Billing volumes that are physically impossible for a single provider — such as billing for more than 24 hours of services in a day, or seeing 400+ patients daily. A strong indicator of fraudulent billing.
A database maintained by the HHS OIG containing individuals and entities excluded from federally funded healthcare programs. Our model cross-references this list to identify providers with prior fraud history still billing Medicare.
The ratio of a provider's submitted charges to Medicare's allowed amount. A markup ratio of 3.0 means the provider charges three times what Medicare considers reasonable. Extremely high ratios (>5x) are a fraud indicator.
The component of Medicare that covers physician services, outpatient care, medical supplies, and preventive services. All data on OpenMedicare comes from Part B claims. (Part A covers hospital/inpatient; Part D covers prescription drugs.)
A unique 10-digit identification number assigned to every healthcare provider in the United States. NPIs are permanent and never reassigned. This is the primary key we use to track providers across years and datasets.
Statistical methods for identifying providers whose billing patterns deviate significantly from their peers. We use z-scores and interquartile ranges within specialty and geographic groupings to flag anomalies.
Evaluating a provider's billing patterns against other providers in the same specialty and geographic region. A cardiologist billing 10x the average for their specialty is more suspicious than one billing 1.5x.
Submitting claims for services that were never actually provided to patients. One of the most common and egregious forms of Medicare fraud.
An aggregated view of a provider's billing history, including total payments received, services rendered, beneficiaries served, specialty, location, and fraud risk indicators across all available years (2014–2023).
An ensemble machine learning algorithm that builds multiple decision trees and merges their predictions. Our fraud detection model uses a Random Forest classifier trained on features extracted from Medicare billing data and validated against known fraud cases.
A provider's position among the 500 highest-risk flagged providers, where Rank 1 has the highest fraud probability. Rankings are based on the model's composite risk score combining multiple fraud indicators.
The average number of services a provider bills per unique patient. Abnormally high values suggest potential upcoding, unbundling, or phantom billing — especially when far exceeding specialty averages.
The medical specialty a provider is classified under in Medicare data (e.g., Internal Medicine, Cardiology, Ophthalmology). There are 100+ specialties in CMS data. Risk analysis is always performed within specialty groupings.
The amount a provider bills Medicare for a service, before any adjustments. Typically higher than the allowed amount. The gap between submitted charges and actual payment is the basis for markup ratio calculations.
The fraudulent practice of billing separately for services that should be billed together under a single code, resulting in higher total reimbursement. The opposite of bundling.
Billing Medicare for a more expensive service than what was actually provided. For example, billing a complex office visit (99215) when only a simple one (99213) was performed. A common and costly form of fraud.
A measure of how frequently a particular service or procedure is used, typically expressed per 1,000 beneficiaries. High utilization rates for expensive procedures in a geographic area may indicate overuse or fraud.
An individual, often a healthcare worker, who reports suspected Medicare fraud. Protected under the False Claims Act and eligible for a percentage of recovered funds. The HHS OIG hotline (1-800-HHS-TIPS) accepts anonymous reports.
A statistical measure indicating how many standard deviations a value is from the mean. In our analysis, providers with billing z-scores above 3.0 in multiple categories are flagged for further review.
Missing a term? Let us know.