Healthcare fraud is not a new problem. Twenty years ago, Harvard Kennedy School professor Malcolm K. Sparrow provided...
a very broad description of the different methods of healthcare fraud in his book License to Steal. The book described how thieves were able to exploit automated healthcare claims processing frameworks to submit false health insurance claims without actually providing the services described, thereby collecting improper payments.
The issue of payments made by government agencies to fraudsters has become so acute that it inspired the passing of the False Claims Act, which imposes liability on individuals or organizations that submit false or fraudulent claims. The problem continues to grow, however, exposing the healthcare industry's lacking fraud prevention and detection efforts: Continued cyberattacks on health insurance companies have led to numerous data breaches that have exposed the personal records of tens of millions of individuals, triggering the HIPAA Breach Notification Rule while providing a treasure trove of data that could be exploited for the submission of fraudulent claims.
In the United States, Medicare and Medicaid spending exceeds $1 trillion each year, while total health spending is approximately $3 trillion. A 2012 RAND Corp. study suggested that the sum of the lowest available estimate of fraud and waste exceeds 20% of total healthcare expenditures, implying that there is an estimated $600 billion in improper healthcare payments annually.
Traditionally, the approach that most payers have taken to combat fraud involves "pay and chase," or identifying providers to whom fraudulent payments have been made and then going after those providers to reclaim the improper payments. This approach is marked by inefficiencies, especially because it targets individuals who may be difficult, if not impossible, to find, and attempts to reclaim the money often come long after it has already disappeared. There are also some experts who believe the estimates named above are low, and that there is a significant amount of false claims that go undetected.
Fraud prevention and detection models
There is a movement to transition from the post-pay fraud detection processes to more proactive, pre-pay fraud prevention models. These approaches are intended to review and identify suspicious claims before they are paid, thereby avoiding improper payments in the first place. At the simplest level, it is possible to define sets of rules that describe suspicious activity, mostly centered on anomalous behavior. Rules-based systems can be used to filter claims and trigger an alert when it identifies a false claim's known pattern. A simple example is a provider submitting two claims for procedures performed at the same time but in physically distinct locations.
Payers are increasingly turning to machine learning to help in predictive fraud prevention, with some adapting machine learning algorithms to gain more precision and accuracy. For example, consider these approaches:
Directed analysis. A large number of fraud cases are reported by whistleblowers, who often are insider employees who have observed what they believe to be suspicious activity. Once the fraud has been verified, the historical claims data submitted by the perpetrators can be used as a training data set. Directed analysis scans data that is associated with known fraud cases to identify behavior patterns and characteristics that correlate with the false claims. The analysis can also reveal new patterns of suspicious behavior, allowing companies to define new sets of business rules that alert specialists when potentially nefarious activity is detected.
Unsupervised analysis. This type of analysis consumes large data sets to look at the types of claims submitted across providers, healthcare beneficiaries or plan members. This approach is intended to expose anomalies and suspicious behaviors that had previously flown under the radar.
Model refinement. This blends the use of directed analysis, unsupervised analysis and business rule models to overlay known fraudulent behavior with predictive models. This model is then used to identify compliance gaps and to try to use machine learning algorithms to fill those gaps and improve fraud prevention and detection.
These approaches employ different analytics methods and algorithms. Clearly, clustering is necessary to characterize providers based on a combination of demographic variables such as geographic location or specialty, and the procedures for which they are submitting claims. Companies can create and put in place decision trees to find anomalous claims based on combinations of provider and claim characteristics. Entity analytics and neural networks can be used to expose self-organized networks to inappropriate activities that warrant closer investigation.
The basic framework combines model development and implementation. Fraud analysts apply both the directed and the unsupervised algorithms to data sets containing claims histories, investigate suspicious activities that are exposed through the analysis, and determine whether the developed analytical models are precise and accurate enough to warrant their integration into the payment processing applications.
If so, enterprises can use the models on the front end of payment processing to submit claims to these models and identify suspicious activity before the payment is processed. This provides a capability for flagging potential fraud and helps with further investigations into improper payments.
These techniques are not limited to healthcare fraud prevention and detection; other industries have similar requirements when it comes to identifying suspicious behavior. Professionals in the financial industry are obliged by law to continuously be on the lookout for unauthorized debit card transactions, check fraud, fraudulent electronic wire transfers, as well as money laundering. Government agencies are constantly combatting billing and payroll/expense reimbursement fraud. Retail organizations must monitor the transfer of goods and ongoing transactions to identify fraudulent activity.
Fraud prevention and detection analytics applications can be used in each of these industries to identify suspicious activities, combat the flood of illicit activity, and save millions, if not billions of dollars.
Watch one healthcare CIO discuss why interoperability and health data analytics are increasingly connected, and learn why healthcare organizations are ramping up adoption of data analysis software. Then get five data analytics priorities that your business should consider in 2016.