IBM's predictive analytics software enabled North Carolina to detect $6.2 million in potentially fraudulent Medicaid payments.
North Carolina's Department of Health and Human Services (DHHS) has conducted an investigation of Medicaid fraud by using IBM predictive-analytics software
similar to that of IBM's Watson supercomputer.
For its fraud detection, the state is using two IBM applications: IBM Fraud and Abuse Management System (FAMS) and IBM InfoSphere Identity Insight
to analyze claims of nearly 2 million Medicaid patients and 60,000 providers to detect suspicious billing patterns.
FAMS mines data to detect patterns of fraud and abuse using modules configured for North Carolina Medicaid. "The algorithms and models are used across the health care industry (both public and private payers) as well as cell phone companies, property and casualty insurers," Shaun Barry, IBM's fraud and abuse leader, wrote in an email to eWEEK.
The program can detect billing behaviors that suggest fraudulent activity.
"The beauty of this system is that it recognizes patterns of billing behavior that don't fit in with the normand then takes it a step further by looking for relationships among providers that can point us to a Web of suspicious accounts," Al Delia, North Carolina DHHS acting secretary, wrote in an IBM blog post
Meanwhile, InfoSphere Identity Insight allows organizations to resolve identity conflicts and determine if providers are using different names on billing statements, said Barry.
"It resolves many distinct provider numbers into single entities based on shared attributes, characteristics and numbers," Barry explained.
DHHS announced the results of its Medicaid fraud research on May 22.
North Carolina Governor Beverly Purdue announced in 2010 that the state would use IBM software to analyze medical claims.
FAMS and InfoSphere offer machine-learning capabilities similar to Watson
, according to Barry. The more data fed into the software, the smarter it becomes, he said.
"What Watson did on "Jeopardy" is it learned to understand the categories and their context as it played," said Barry. "This helped it build a better profile of answers for the category."
North Carolina's fraud detection is following a similar process, he said.
"As cases and data are processed, the system is storing certain patterns of behavior that have a high probability of fraud," said Barry. "This improves its ability to find similar cases as new data is fed to the system."
Using the IBM applications, North Carolina discovered that providers billed for 44 hours of service in a 24-hour period, DHHS noted. In addition, providers billed for both group rates and individual group members at the same time.
DHHS searched three years of claims data during the first phase of its inquiry, the state reported. It has completed 10 investigations of outpatient behavioral health, and it discovered $6.2 million in potentially fraudulent payments, according to DHHS.
In addition, 206 outpatient behavioral health providers across the state may have charged "unusual" Medicaid billing worth up to $191 million, the state reported.
DHHS has referred the case to the Medicaid Investigations Unit of the North Carolina attorney general's office.
As part of the investigation, DHHS looked at providers with unusual billing, compared with peer groups of similar providers, geographic areas and procedure types, according to DHHS. The department will research whether providers intended to bill as they did.
"Leveraging both technical and deep human expertise, the state can enhance the standard of care delivered via the Medicaid program for all residents," said Barry. "North Carolina's efforts show the power of advanced analytics can be applied to multiple facets of government, improving efficiency and building fairer systems for everyone," he said.
In just our first phase of data analysis, weve flagged hundreds of providers who have submitted bills for hundreds of millions of dollars in suspicious claims," said Delia. "Analytics makes it possible to sift through tens of thousands of documents and hundreds of millions of pieces of data to identify strange behavior in a matter of minutes."
Using the information gained from the data mining thus far, DHHS investigators will look for suspicious activity during unannounced visits to health providers this summer.