Eliminating fraud (in healthcare) with EDM
Bill Briggs over at Health Data Management wrote a nice little piece on how I.T. Helps Payer Smoke Out Fraud. This article brings up what is a huge issue - healthcare payers are paying tens of billions in erroneous, abusive and fraudulent claims that look just fine in editing and adjudication systems. As the article points out, even rules-driven fraud detection systems won't catch all this kind of fraud as they rely on the payer knowing, at some level, what rules to apply. In reality, while the majority of fraudulent claims appear legitimate when viewed in isolation, they actually shouldn’t be paid in full, and many of them should not be paid at all. Once they are, only a tiny fraction of the dispersed funds will ever be recovered.
The problem is that claims data, in and of itself, only goes so far in revealing billing problems. Claims edit and adjudication systems, as well as fraud detection systems that rely on rules alone, do not go much beyond examining the data on the claim—something like looking at just one leaf on a tree. These systems do not recognize that claims which are correct in isolation may actually be part of a larger pattern of fraudulent activity, repeated error or systemic weakness. They do not notice providers manipulating the system by upcoding or those billing for services that are unrealistic and for amounts that are out of alignment with their peers. Focusing on the leaves, they cannot see the trees—much less the vast forest of costly billing problems.
Nevertheless, claims data is the key to detecting and stopping these problems. It contains rich data on providers, patients and other healthcare participants. A detection system with powerful predictive analytics extracts data from all payer claims on an ongoing basis. It should capture meaningful information from this vast quantity of data and mathematically distill it down into a highly compressed and efficient form, analyzing each incoming claim against this rich context. The results of this complex multidimensional analysis—which enables predictive analytics-based solutions to minimize losses and even prevent them before payment—should be output in simple, actionable form:
- Scores and rankings to focus analysts on the most problematic claims and providers
- Explanations, with links to evidence, to enable analysts to rapidly understand the source of a problem
- Correcting errors on the spot through integration of the fraud decision with the claims process
- Opening and referring investigative cases and recommending policy changes to address systemic issues where a simple action is not sufficient
Healthcare payers stepping up to this level of decision management can achieve real savings. They can detect fraud in prepayment to prevent funds from being dispersed unnecessarily. They can get weekly rankings of problematic providers that enable early investigation. They can conduct comprehensive post-payment analyses that reveal large-scale fraud. And so on. Fair Isaac's experience is that customers using such predictive analytics-based detection systems consistently produce savings that add up to as much as 10:1 ROI or more.
Thanks to Teri Kim for some extra information here and for reminding me I blogged about this before - Predictive analytics can detect growing fraud in Healthcare Claims