Published on October 7th, 2014 | by Travis Korte0
How CMS Can Improve its Fraud Prevention System
In 2011, the Centers for Medicare and Medicaid Services (CMS) implemented the Fraud Prevention System (FPS), a system designed to analyze Medicare claims data to predict potentially fraudulent behaviors and report the results to analysts, who can use the information to prioritize investigations. This is a critically important system in the fight against Medicare fraud, which costs taxpayers $60-$90 billion annually. While the system is useful, there are opportunities for improvement. Specifically, CMS should better integrate FPS into its payments systems so that it can suspend payments for suspicious claims, develop better performance targets to ensure the system is performing as desired, and improve its predictive models to detect more fraudulent claims.
First, CMS should ensure it is capable of suspending payments when FPS detects suspicious claims. While the agency’s fraud inspectors can deny payment for claims found to be fraudulent, they cannot for claims that are under investigation. As a result, CMS routinely pays suspicious claims that it is investigating, meaning that the agency must try to recover the funds if the claims turn out to be fraudulent after all. In many cases, it is unable to do so. Part of the motivation behind using predictive analytics to detect fraud is to move away from this “pay-and-chase” model and toward preventing fraud before it happens, but as long as CMS cannot suspend claims during ongoing investigations, the problem will persist. The Government Accountability Office (GAO) recommended in 2012 that CMS better integrate FPS with its Medicare payment processing system to allow this kind of payment suspension, and although CMS has made progress by implementing the ability to deny certain suspicious claims outright through FPS, it still has not resolved the problem of payment suspension. If it is to truly abandon the “pay-and-chase” approach to fraud, CMS should integrate the FPS and payment processing systems per GAO’s recommendations.
Second, CMS should develop stronger performance targets and milestones to evaluate FPS. Such performance measures are critical to ensuring the program is succeeding in its mission to reduce Medicare and Medicaid fraud, as well as helping inform Congress’s future funding decisions around the initiative. Although GAO raised this issue in 2012, CMS has argued that such targets and milestones are difficult to create, both because investigators’ performance according to these metrics may not accurately capture their merit and because the very existence of FPS should have a deterrent effect on fraud that is difficult to capture numerically. Even acknowledging both of these difficulties, CMS can do more to rigorously evaluate the extent to which FPS is working. For example, as of 2012, only five percent of CMS’s fraud investigations originated from FPS (the remainder coming from traditional means of fraud alerting, such as consumer complaints). One reasonable performance target could involve the system’s adoption rate among investigators. In addition, CMS could strive to reduce FPS’s false positive rate, i.e., the proportion of claims FPS flags as suspicious that do not ultimately lead investigators to fraud; or its false negative rate, i.e., the proportion of all fraud findings where FPS processed the claim but did not detect any issues. In general, the agency should continue to seek quantifiable metrics for FPS’s success.
Finally, CMS should improve FPS’s algorithms to ensure it is detecting as much fraud as possible. Because so much fraud goes unrecovered, even a modest improvement in algorithmic performance could mean major savings. CMS’s current process for refining FPS models involves collecting feedback from a variety of sources, including investigators, data analysts, and others. However, more effective means of improving algorithms exist. Private and public sector entities alike, such as streaming video company Netflix and the U.S. Defense Threat Reduction Agency, have hosted challenges to improve algorithms. CMS should host a similar competition, offering a public claims data set and challenging developers to exceed FPS’s accuracy in detecting fraud. The agency could offer a reward for the competition based on the improvement in accuracy, like the ones Dun & Bradstreet is offering for an upcoming fraud detection challenge of its own. Ultimately, agencies like CMS should be working to not only use analytics, but to build a platform that allows it to take advantage of the best analytics capabilities available.
Medicare and Medicaid fraud represents a massive annual cost to taxpayers, and CMS has a major opportunity to generate substantial tax savings with FPS. While CMS has made some progress at reducing fraud, it can still improve its efforts with greater payment systems integration, better performance targets, and more accurate algorithms.