Software Snares Medical Claims Cheats

The Problem:
A large health-care insurer needed a means to automate the detection of phony claims by unscrupulous providers in order to reduce fraud losses.

The Details:
At Highmark, Pennsylvania’s largest health-care insurer with 4.6 million members and 48,000 health-care providers, director of special investigations Tom Brennan says accessing and analyzing the company’s claims data for fraud patterns was a slow and often cumbersome process. Staff first had to ask analysts in Highmark’s healthcare informatics department to sift through raw claims data and organize it into reports. Investigators then would review the reports to identify unusual claims patterns. “We had several data sources, and sometimes it took days or even weeks to get a response to a [data mining] request,” Brennan explains.

The Solution:
Brennan teamed with the informatics department in 2002. “They had a lot of analytical skills, and we had knowledge of how these health-care fraud schemes work,” he says. Highmark built a Web-based data mining and analysis application using business intelligence software from SAS Institute called Enterprise Miner, and installed the application on every personal computer in the special investigations unit. “Now we can do our analysis in minutes, rather than days,” Brennan says.

Using a Web interface, investigators select the data they want to look at and then create their own reports based on the data they’ve extracted. Highmark’s enterprise data warehouse of claims information, which gets 15,000 queries daily, runs on a Teradata 525x/5380 system. A typical query now takes an investigator about 20 seconds. “They can do a lot of investigative queries on the spot,” says Brennan’s counterpart in anti-crime, Shawn McNelis, vice president of Healthcare Informatics at Highmark.

Investigators use the system to quickly detect potential fraud patterns, such as an inordinately high dollar amount of claims paid to one provider in a specific time period, or unusual dollar payments for a particular procedure code. For instance, if the typical dollar amount for a certain procedure code is $100 and investigators spot a flurry of $500 claims under that code, fraud is a likely reason for the anomaly. Other strong indicators of fraud include a high number of claims, certain procedures performed too closely in time, and an impossible number of procedures performed by one provider or practice in a given time period.

The Result:
Brennan says that since 2002, the anti-fraudulent-claims campaign has resulted in recovering $23 million from improperly paid claims. What’s more, his staff, which now can handle investigative queries in seconds instead of eight hours, is now managing a 30% increase in investigative caseload, saving the company an additional $1 million a year.

In one case, Brennan says a claims analysis prompted an investigation of chiropractor Douglas Henderson of New Kensington, Pa. Henderson, who pleaded guilty to criminal charges of insurance fraud in U.S. District Court in Pittsburgh in April 2006, is due to be sentenced this month. The federal indictment charged that Henderson and others engaged in a conspiracy to submit claims for services not provided, totaling $7 million from 1995 to 2003. “This chiropractor worked with a broker and put together false documentation, and got a 42-member group of patients underwritten,” Brennan says. “He would pay for their health care if they let him bill, and he split fees with the members of the group.”

The system flagged Henderson’s claims, Brennan says. “Some of the individuals in the group couldn’t have been receiving all the health care that was being billed,” he says, concluding: “That case was a result of our getting together with the informatics department and having them build an application so that we could do our own analytics.”