Inside Look At Healthcare Fraud Prevention Algorithms

On July 30, 2011, the Centers for Medicare & Medicaid (CMS) implemented its new Fraud Prevention System (FPS), which uses predictive analytics technology, similar to that utilized by credit card companies, to move away from the “pay and chase” model to instead detect aberrant or fraudulent billing patterns prior to payment of claims. According to CMS, by fiscal year 2013, CMS was able to take administrative action against 938 providers and suppliers using FPS, saving or preventing $210.7 million in payments.

In its recent report to Congress, CMS discussed how the FPS algorithms operate to detect patterns of fraud. The FPS, says CMS, screens all Medicare Part A and Part B claims prior to payment, running each claim against multiple models. FPS also analyzes other data to make predictions, including database with compromised physician and beneficiary identification numbers, the Fraud Investigation Database, that contains information on all investigations developed by CMS’s program integrity contractors, and complaints from the Medicare fraud tipster hotline. FPS then creates alerts as each model identifies claims and other data that suggest aberrant billing, further explained CMS in its report. These alerts are then consolidated on a provider and the FPS adds background information to provide context to the alerts. Finally, reports CMS, these leads are then prioritized by potential fraud risk in the system and law enforcement officials as well as Zone Program Integrity Contractors (ZPICs) further investigate the identified providers and suppliers. Medicare Administrative Contractors also have access to FPS data and can use that data to intervene promptly in cases of fraud or abuse and focus their review processes on particularly high risk areas.

What kind of models does CMS use? CMS identified several analytical models to guide the FPS in its function:

  1. Rules-based model that uses rules to categorize potential cases of fraud, such as, for example, “billing for a Medicare identification number that was previously stolen and used improperly”
  2. Anomaly-based model that searches for abnormalities in relation to peer groups such as, for example, providers that “[bill] for more services in a single day than the number of services that 99% of similar providers bill in a single day”
  3. Predictive-based model that uses established fraud cases as a reference point such as, for example, “[providers] that [have] characteristics similar to those of known bad actors “
  4. Network Analysis-based model that uses associative link analysis to, for example, single out “[providers] that [are] linked to known bad actors through address or phone number”

The FPS is not static and, according to government officials, is constantly reviewed by experts from various fields (economics, statistics, programming) to determine its effectiveness and to make improvements where necessary, while identifying potential vulnerabilities in the system and trends of possible Medicare fraud and waste within the results it produces.