CMS Ramps Up Use of Predictive Analytics to Combat Fraud and Abuse

On June 18, 2015, the United States Department of Justice announced that the Medicare Fraud Strike Force has charged 243 people in 17 cities with $712 million in alleged fraudulent Medicare billing.  In bringing these charges, the strike force relied on advanced predictive modeling technology among its investigatory resources and tools.

As detailed in a 2012 press release, the Centers for Medicare and Medicaid Services (CMS) Fraud Prevention System (FPS) has been applying various forms of statistical analyses, known as “predictive analytics” in an attempt to identify or prevent improper payments.  It is estimated that in its first two years of operation, the FPS prevented over $300 million in improper Medicaid fee-for-service payments, which is a sizable proportion of the estimated $19.2 billion in fraud recoveries over the last five years.  Recently, the CMS Office of Inspector General (OIG) announced its intention to increase its enforcement efforts predicated on similar data-mining techniques.

Predictive analytics begin with gathering historical and transactional data which are then subjected to various forms of statistical analyses, usually using sophisticated modeling, such as machine learning, and other techniques.  According to CMS, its analysts use near-real-time data to examine Medicare claims to identify potentially fraudulent providers and patterns of suspected fraud. This analysis includes studies of disproportionate payment levels for various services and ratios of services as compared with national averages. These and other assessments enable OIG to identify and adapt enforcement efforts to emerging trends as they evolve.

The use of sophisticated data collection and analysis to build a predictive score, or probability, that each individual among a large number of individuals will act in a particular manner is already widely used by retail stores, investment banks, grocery chains, and even the U.S. Postal Service.  Its application to healthcare fraud detection in this context is neither unprecedented nor surprising.

A recent settlement with a California pharmacy illustrates the manner in which “near real time” data analysis is being used by OIG. Using information provided to it, the OIG alleged that the pharmacy and its owner submitted Part D claims for brand name prescription drugs that it knew, or should have known, were not provided as claimed, and were false or fraudulent.  These allegations were based on inventory records that indicated that the pharmacy could not have dispensed the quantity of Part D drugs which were billed to Part D.  The government’s allegations were settled after the pharmacy and its owner agreed to pay over $1.3 million.

As analytics are applied to the vast amounts of healthcare claim and transaction data that have been collected by CMS, investigations like this one are likely to become more common.  Indeed, at the recent American Health Lawyers’ Association Annual Conference, OIG representatives praised data-mining based investigations and indicated that providers and suppliers could expect increased activity in this regard.  Already, in a nationwide series of audits of Medicare outpatient services in which payments exceeded charges, an OIG audit team used of data mining and analysis to issue a total of 26 final reports to the Medicare contractors, resulting in expected recoveries totaling $106 million. In addition, because of CMS’s verification policy edit implemented as a result of these OIG audits, the Federal Government is expected to save about $30.3 million in future Medicare payments each year.

Utilization of these techniques is not limited to the federal government.  Beginning in 2013, the OIG issued a final rule permitting Federal financial participation in the costs of data mining by state Medicaid Fraud Control Units (MFCUs), provided that certain criteria are satisfied.  To date, six state MCFUs participate in this program:  Florida, Michigan, Missouri, Oklahoma, California, and Louisiana.

Data analyses and predictive analytics are becoming increasingly cost-effective, and therefore popular, fraud-detection tools.  Instead of investigators spending hours poring over paper records, outlier situations can be identified through predefined automatic algorithms, or a few clicks of a spreadsheet.  It is likely that the government and other payers will continue to scale-up use of these tools, possibly capturing conduct which would have remained otherwise undetected absent a more obvious and focused forensic analysis.