Predictive analytics helps fraud fighters detect sophisticated schemes [Special Report]

Predictive analytics might be the hottest buzzword in healthcare. Over the last several months alone, data analytics have helped researchers predict where and when the flu is at its peak, allowed providers to determine what kind of care patients will need six months from now, and helped payers identify gaps in care.

Now, predictive analytics is changing the way payers identify instances of fraud, waste and abuse in healthcare. Last month, the assistant inspector general and chief data officer at the Office of Inspector General, Caryl Brzymialkiewicz, highlighted analytics as a key tool in the largest healthcare fraud bust in U.S. history.

Increasingly, both public and private payers are turning to data analytics to identify high risk fraud trends, Andrew Asher, senior fellow and director of data analytics at Mathematica, said in an exclusive interview with FierceHealthPayer: AntiFraud. However, payers are in the early stages of using healthcare claims data to accurately predict fraud schemes. In many ways, fraud fighters are still transitioning from the old "pay-and-chase" models to a more proactive approach, in part because predictive analytics systems are still finding a foothold in the fraud-prevention world.

"Payers remain concerned about the risks of false positives and about the accuracy of models to identify actionable findings," Asher said.

Although predictive analytics is still in its infancy, health insurers are finding notable success in an approach that blends the human element of fraud expertise with customizable models that can rapidly identify sophisticated, high-risk fraud schemes.

A multi-dimensional approach

In an effort to fine-tune fraud prevention efforts, healthcare payers have found that data analytics can be used in a variety of ways, with each method offering distinct benefits.

A rules-based approach identifies claims that are almost certainly erroneous. This might include billing for more than 24 hours in a day, providing services after the date of a beneficiary's death, or transporting a patient in the middle of an inpatient stay. This method highlights a very specific and straightforward error within the billing claim, and allows an insurer to take prompt action against the provider.

Using this approach, payers are able to construct predictive rules that prevent payment of similar claims, Richard Statchen (left), manager of informatics at Aetna, said in an exclusive interview with FierceHealthPayer: AntiFraud. Certain fraud patterns emerge, such as a beneficiary traveling 75 miles for a prescription, or visiting 20 pharmacies in a week. Those types of outliers are sent directly to investigators at Aetna, who can stop payment and request additional records.

A model-based approach is not nearly as cut and dry, but it allows payers to identify much more sophisticated, high-risk fraud schemes that often require additional investigative measures. Using this approach, analytics can help identify utilization anomalies within a geographical area or a certain service line that would have taken significantly longer to recognize, Statchen said.  

"With the new analytical solutions, you're really able to, within a handful of claims, make a quicker determination so that you don't give the providers that much lead time to actually accumulate payments over time," he said. "It's really a much more effective and quicker approach."

Statchen adds that Aetna focuses its detection efforts on broadly identifying schemes rather than targeting certain specialties. However, Aetna currently has a team looking specifically at phantom providers, using the characteristics of previous schemes to prevent new fraudulent providers from entering the system.

These types of provider-centered and claims-centered models emphasize risk, which is the ideal way to detect the most sophisticated schemes, Asher said.

"The best analytics are concentrating on finding fraud networks," he said. "They are looking to find fraud networks which can have a very expensive impact and they can rapidly respond to detection by trying new billing strategies."

Key factors: Customized technology and the human touch

Although data analytics has firmly implanted itself in the fraud prevention world, a number of obstacles still remain as payers look to use new data systems more efficiently.

First, payers need a system that can be "deployed flexibly," Asher said. Typically the commercial, off-the-shelf systems are not dynamic enough to keep up with a modern claims processing system or keep pace with sophisticated fraud schemes.

"[Payers] need a customized approach to their own system and their own payment rules, data and policies, all which can be highly varied," he said.

Although larger payers may have the capital to invest in that kind of sophisticated technology, mid-size and smaller insurance companies can find that challenging, Asher adds.

According to Statchen, Aetna uses a combination of vendor-based solutions and internal systems. In recent years, however, that balance has shifted toward predominantly internal systems, in part because it is allows for more flexibility and customization.  

"In the past, the resources to develop this type of data science and analytics were not present in Aetna, but over the past few years we've really invested in that skill set, so we are shifting more toward internally based solutions," he said.

Secondly, a purely technology-based approach to fraud prevention fails to account for the nuances of healthcare reimbursement. Although data systems are great at identifying potential problems, the complex nature of healthcare and the structure of various plans and payment systems require a decidedly human element. Unlike the credit card industry, which can send automated alerts for potentially fraudulent charges, healthcare claims require the expertise of someone familiar with the various contractual agreements with providers, Statchen said.

In September, the Medical Identity Fraud Alliance published a survey indicating that although software and hardware used to identify fraud still dominates budgets, the healthcare industry is beginning to place greater emphasis on personnel to supplement data analytics.

"The best solution is one where sophisticated technology with a deep understanding of data, and that human element, is all brought together," Asher said. "It's not 'either-or;' it's bringing all of those together in a very deep and dedicated way."

Keeping pace with evolving fraud schemes

The shift toward predictive analytics is promising, particularly as many organizations look to move away from the traditional "pay-and-chase" methodology of combating fraud. But in many ways, this method is still evolving.

Asher says that there is still room for payers to bolster fraud detection systems by using internal and external data, incorporating electronic health record information, and employing broader data sets that enable models to identify fraud trends more rapidly.

Additionally, payers continue to fine-tune the ways they use analytics to take appropriate action against providers or conduct follow-up reviews. In the future, fraud prevention will rely on a closer partnership between payers and providers that uses analytics to improve billing compliance.

"[Future efforts will] couple analytics with rapid-cycle testing strategies to determine which provider feedback mechanisms are most effective to prevent future erroneous billing," Asher said. "This approach would not be used for the providers suspected of fraud, but for ones with billing errors."

In an industry in which fraud schemes are constantly evolving, payers need to ensure that their data is in "good condition," Statchen said. Data integrity issues will sink even the best models and rules, so payers need to make constant adjustments to their models and conduct regular testing to ensure analytics are accurate and effective.

He adds that once an analytics system is firing on all cylinders, the problem is managing the increased workload. Over the last several years, he estimates Aetna has generated 60-70 percent more cases because better data analytics have generated more leads.

"There is always going to be fraud out there, and there are always things you're not uncovering," he said. "The more you get into it, the more cases you have."