Using Predictive Modeling to Detect Medicare Fraud

Since June 30, 2011, the Centers for Medicare and Medicaid Services (CMS) has been using a predictive analytics model to analyze Medicare fee-for-service claims to detect patterns highly associated with fraudulent activity. The technology allows CMS to conduct real-time, pre-payment claim analyses instead of a “pay and chase” approach that requires the agency to reimburse providers first and then attempt to recover overpayments after the fact.

Under the new model, all claims travel through the predictive modeling system, which uses the data to build profiles of providers, networks, billing patterns, and beneficiary utilization. These profiles enable CMS to create risk scores that (1) estimate the likelihood of fraud and (2) flag potentially fraudulent claims and billing patterns. The process is similar to the pre-payment analysis already done in the financial and credit card industries.

Risk scores enable CMS to quickly identify unusual billing activity and subject suspicious claims to more thorough review before making payment. The system automatically prioritizes claims, providers, beneficiaries, and networks that are generating the most alerts and highest risk scores Analysts review prioritized cases looking at claims histories, conducting interviews, and performing site visits as necessary. If an analyst finds only innocuous billing, the outcome is recorded directly into the predictive modeling system and the payment is released as usual. CMS reports that this feedback loop refines the predictive models and algorithms to better pinpoint fraudulent behavior.