Texas Drops a Smart Bomb on Medicaid Fraud
OPERATION JUST DESSERTS
A TECHNOLOGY that helped guide smart weapons during the Gulf war and saves credit-card companies millions of dollars is now being used to take on Medicaid fraud in Texas.
The Lone Star State will be the first in the nation to use neural-network technology to catch those who steal from the federal program that is designed to help the poor with medical bills.
Fraudulent Medicaid claims cost Texas taxpayers an estimated $1 billion per year. Yet only a handful of cases are investigated. Texas Comptroller John Sharp believes the solution may lie in neural networks, complex computer systems that can learn how to detect waste and fraud.
Neural networks have already proved their value in the private sector.
''It's a great technology to help us fight credit- and debit-card fraud,'' says Gail Murayama, a spokeswoman for Visa International, which began using a neural network on its computer system in 1993. ''Our counterfeit-fraud losses are down 18 percent. Since January of 1995, Visa's member financial institutions have saved $100 million by using the neural networks.'' Visa currently uses neural networks in the US and Europe and is testing the procedure in Asia and Latin America.
If neural networks are as effective in combatting Medicaid fraud as they have been in the credit-card industry, they could save taxpayers billions of dollars per year. In 1994, the General Accounting Office estimated that 10 percent of all Medicaid claims nationwide are fraudulent, costing $31 billion per year. In addition, the Federal Bureau of Investigation estimates 15 percent of all health-care costs are attributable to fraud.
Risto Miikkulainen of the computer-science department at the University of Texas here, has been studying neural networks for nine years. He says their application in the area of Medicaid fraud could be fruitful, because the data generated by Medicaid is similar to that created by the credit-card business. ''The tool, neural networks, matches well with the problems of Medicaid fraud, because they can very effectively filter huge amounts of data,'' he says.
Neural networks mimic the workings of the human brain. But instead of neurons, they rely on a series of processing units, which are trained through repetition to recognize patterns in streams of data. Gradually, the system learns to categorize patterns and call them to the attention of system operators.
Neural networks, which were first developed in 1949, did not gain widespread attention until the Gulf war, when they were designed to recognize the heat exhaust from a tank or building and guide rockets and bombs toward the targets.
Mr. Sharp says neural networks will provide a ''smart bomb for Medicaid fraud.'' Over the past five years, neural networks have found a wide range of applications and are now used in handwriting, fingerprint, and speech identification.
Fraudulent Medicaid claims are costly, yet few are detected. Texas and other states are required by the federal government to use a program called SURS, for Surveillance Utilization Review Subsystem. But since 1977, the system has not uncovered a single case of fraud in Texas that has led to prosecution.
Yet in some instances, fraudulent activity is glaringly apparent. Acting on a tip, the Texas attorney general's office recently began investigating an Austin-based doctor who submitted Medicaid claims 361 days out of the year. Texas officials say that kind of activity should be easy to spot using a neural network.
''I think we can detect 25 to 50 percent of the existing Medicaid fraud,'' says Joe Brown, the chief technical officer for Berkom USA, the Austin-based company that is developing the technology for the state of Texas.
Sharp, who developed a system that replaced food stamps with an electronic benefits-transfer card the US Department of Agriculture plans to implement nationwide over the next four years, says the Medicaid neural network should be used by the federal government as well.
''Hopefully, we will be able to convince the federal government to change what they are doing,'' Sharp says. Referring to the SURS program, he adds, ''if they are going to mandate a system, they ought to mandate one that works.''