Filter Out the Frauds

Fair Isaac's version 2 software, released earlier this year, uses pattern-matching technology that the company picked up when it bought HNC Software in 2002. The software can sift through claims and segment them by groups, such as the claims coming from out-of-network healthcare providers performing surgery. Traditionally, insurers hunt for fraud with rules-based data mining systems, searching for claims that meet well-defined conditions, such as doctors who bill for 25 hours of office visits in a single day, and applying rules that can detect fraud. But, if crooks vary their schemes from case to case, that method isn't as effective.

Using pattern matching, insurers won't need to know all the details of how a scam operates before detecting it, says Fair Isaac healthcare vice president Joel Portice. Banks, insurance companies and retailers have used HNC's technology to detect patterns from masses of complex data. "It makes sense to apply the same technology to healthcare," Portice says.

Sounds great, but there are lots of reasons it hasn't been done. Pattern-recognition systems such as Fair Isaac's Payment Optimizer or rules-based systems such as IBM's Fraud and Abuse Management can only see shades of gray in a mass of healthcare claims. They seldom can put their digital fingers precisely down on fraud. That kind of intuition takes human investigators, who can decide whether the claims constitute fraud or abuse of the healthcare system.

But flagging more cases might mean delaying payment to legitimate providers. "We want to pay claims in a timely manner," says Rich Appel, Philadelphia-based Cigna Corp.'s director of special investigations. "It's a fine line to walk when you work with these kinds of tools."

Another risk: gumming up the works that get health providers and patients paid on time. Prepayment analysis of claims threatens to create delays that run afoul of states' "prompt payment" laws, which require insurers to get their payments to doctors and hospitals in 14 to 45 days. That leaves little time to investigate suspicious claims. Forty-nine states have such laws on the books.

Double-Edged Sword

Payment delays rile providers, whose offices depend on the cash flow, adds Steven Skwara, associate general counsel and director of fraud investigation at Blue Cross Blue Shield of Massachusetts (Boston). "Nowadays, the fraudulent claim looks exactly like the legitimate claim next to it," he says.

When inspectors find a suspicious claim, they often ask providers for medical records, bills and other documentation. "Thirty to 45 days can drop down the drain right there," Skwara says. "Strictly electronic processing takes human eyes away from claims. The technology gain cuts both ways."

Then there's the fact that computer programs - no matter how sophisticated - just aren't as good as human eyes at spotting fraud. With 4 billion healthcare transactions to process a year and unyielding pressure to keep costs down, insurers' claims processing systems "operate on a trust factor" that claims are valid, according to the National Health Care Anti-Fraud Association's Hennings.

The 1996 Health Insurance Portability and Accountability Act (HIPAA), a federal law passed to ensure the privacy of patients' medical records, was meant to reduce human involvement in claims processing, he adds. But turning over that work to machines could cause some fraud to go undetected.

Even if pattern-matching systems work as advertised, they're not the whole answer. Vista Healthplan's Rushton says the company will combine Fair Isaac's new technology with software that sifts through claims after they're paid, looking for anomalies and evidence of known schemes, as well as emerging scams sought by Fair Isaac. The post-payment data mining technology can react to rules drawn up to detect known scams, such as bills for tests unrelated to a procedure or a provider that bills for the same patient twice.

At Cigna, investigators this year have flagged 25 cases as "most suspicious" and needing further investigation, culling them from a larger pool of claims the software identified as unusual for some reason, says the insurer's Appel. He adds that selectivity shows that the rules engine software is picking out only the most worthwhile cases to pursue. Cigna is considering augmenting IBM's rules-based system with pattern recognition technologies, such as Fair Isaac's, that could spot patterns hidden in the data that don't make sense. "Technology is just a piece of the puzzle," Appel says. "The investigators do the legwork."

That conservative approach to technology is typical of many insurers. Harvard Pilgrim Health Care, a Boston-area health insurer, uses Fair Isaac's Payment Optimizer, but so far the carrier utilizes it only to examine claims data after payment. The pattern-matching system is good at looking for aberrations in claims by comparing a provider to a peer group. If a pediatrician starts billing for allergy treatments, Fair Isaac may single out the claim as departing from the norm and thus suspect. Out of a group, "Fair Isaac picks out what's different," says Kimberly Grose, Harvard Pilgrim's vice president of network services operations. For now, Harvard Pilgrim is satisfied with using the system in post-payment analysis, she says.

Everyone in law enforcement and the insurance industry knows a new technology tool isn't going to solve the whole problem. Says the FBI's Delaney, "There's no easy fix."

Aetna manager Wright adds that once fraudulent claims are detected, con artists will likely "just find another way to submit the claims. The game changes," he says. "Whether we can reduce the number of perpetrators, I'm not sure. But we think we will reduce the total loss."

This article was originally written for InformationWeek, a sibling publication of Insurance & Technology.

2 of 2