The fight against health care fraud
- By Paul McCloskey
- Aug 18, 2015
Health care fraud is one of the government’s costliest problems. It’s a hall of mirrors where billions of dollars are lost to swindlers looking to cash in on the millions of transactions generated by insurance-paying agencies every day.
Last year, the federal government lost $124.7 billion in fraudulent or improper payments through 124 programs, according to the House Ways and Means Committee’s Oversight Subcommittee. Medicare fraud accounted for about half, or $60 billion, of the losses.
Behind that backdrop, for the past three years officials at the Centers for Medicare and Medicaid Services (CMS) have been working on a system designed to scan for clues to fraud in aggregated claims data. The Fraud Prevention System (FPS) flags anomalies before a payment is made, much like credit industry systems can spot a potentially fraudulent charge and withhold payment while the transaction is investigated.
FPS’ developers say the system’s biggest challenge could be the sheer complexity of the government’s health care payment system. According to the National Health Care Anti-Fraud Association, Medicare Parts A and B process 4.5 million claims every day from 1.5 million health care providers, which means fraudulent patient records and treatments could slip through.
“Preventing fraud is so difficult because the schemes and participants are constantly...evolving to elude enforcement actions,” Charlene Frizzera, president of CF Health Advisors, told the subcommittee.
As FPS continues to grow, companies are also building new analytics tools that identify fraud more quickly, accurately and earlier in the payment cycle.
Those companies cite advances in predictive analytics to flag probable fraud leads, case-management platforms to help assess risk throughout a fraud case and a greater emphasis on sharing information to help catch fraudsters.
“There is no silver bullet in protecting against fraud,” said Mark Nelsen, senior vice president for risk products and business intelligence at Visa. Instead, payers should field a combination of technology, processes and people, and use data analytics as a “critical component in making all three of these effective.”
The company’s analytics suite evaluates as many as 500 data elements to identify suspicious transactions as they occur. The evaluations provide an instantaneous rating of a transaction’s fraud potential by checking the history, geolocation and transaction speed of a potentially fraudulent event.
The investigation continuum
Vendors say anti-fraud campaigns depend on the sharing of pertinent datasets so that teams of data scientists, program integrity experts and software developers can mount successful fraud cases.
“Integration within the data warehouse is absolutely vital because it lets you visualize things you would not see with smaller datasets,” said Elizabeth Snavely, director of fraud, waste and abuse products at General Dynamics IT.
Data-visualization techniques also allow risk managers to monitor a fraud scheme as it unfolds, which Snavely compared to seeing maps of an epidemic spreading from one side of the country to another.
General Dynamics offers a number of applications designed to support the individual stages or the continuum of a fraud case, she said. The first application generates investigative leads by, for example, flagging the number of patients seen by a clinical provider in a single day.
“Let’s say you are looking at a series of providers and they are indicating that they saw 52 patients in a day,” Snavely said. “That’s a red flag. Either that provider is extremely dedicated or the dates in their computer program are wrong.”
Could they be telling the truth? “Absolutely,” she said, and that’s why lead generation is only the first part of the continuum.
Snavely added that although it seems straightforward, “it can take two years to train someone to be a good investigator and pick out those data anomalies.” In the early stages of an investigation, “it’s not necessarily fraud you’re seeing, but it’s finding that string of thread that you’re going to start pulling on.”
The next stage of an anti-fraud sequence involves applying analytics to promising leads. In the scenario in which a doctor claimed to see 52 patients a day, risk managers would look beyond report summaries and ask for all the data on the claims received.
At this stage, investigators might seek information on whether all 52 claims had been paid, whether they were all for the same procedure and whether they all had the same diagnoses. “It’s where you start to follow a process and use analytic tools to see what you really have,” Snavely said.
In addition to lead generation and analytics, General Dynamics offers a fraud case-management tool that helps maintain program integrity by moving an investigation forward based on the best anticipated financial recovery.
One of the last pieces in the company’s product line is a prepayment review that provides a final layer of analytics. “You take all that stuff that you learned during your investigation and you feed it into a prepayment system that begins to flag [problems] before the payment is made,” Snavely said.
Setting the risk dial
Given the complexity of a typical fraud case, industry executives say prepayment analytics are useful but not perfect tools. For one thing, most systems flag too many false positives, creating conflict between payers and providers.
“I’ve seen these put in place in a few different states,” said Monty Faidley, director of market planning, health and human services at LexisNexis. “The provider community gets upset because too many claims are being stopped or being flagged; they’ve got delays in payments. Those provider networks are often strong lobbying groups, and their complaints get heard very quickly.”
Therefore, using prepayment analytics requires the ability to fine-tune the risk equation by balancing the requirements of payers and providers. “You need a soft touch,” which might involve putting a test in place and tracking its impact over several months, Faidley said.
LexisNexis recently helped New York City’s Human Resources Administration use predictive analytics to study costs related to providing benefits to its 2.9 million Medicaid recipients, 1.8 million Supplemental Nutrition Assistance Program recipients and 350,000 Cash Assistance recipients.
The company combined the agency’s data with LexisNexis’ public records data, “which gives us broad context of information about each beneficiary,” Faidley said. By applying its analytics scoring model to differentiate low-risk beneficiaries from those who “definitely need to be investigated,” the city was able to flag costs associated with 9,700 cases and save more than $52 million.
“Fraud is always evolving, but these solutions are beginning to make a difference,” Faidley said. “We see more procurement and interest, especially at the legislative levels. They’re starting to ask some hard questions and are saying, ‘It’s time to emphasize this and really make a difference.’”
Paul McCloskey is senior editor of GCN. A former editor-in-chief of both GCN and FCW, McCloskey was part of Federal Computer Week's founding editorial staff.