Opera Uses Big Data To Fight US Health Insurance Fraud
Opera Solutions is working with the US federal government to detect fraud in the country’s new health insurance exchanges using big data techniques
As the US federal government prepares to step in and build health insurance exchanges (HIXes) for those states that decline to implement them under the Affordable Care Act, preventing fraud in the exchanges will be a challenge for the American health care IT industry.
Following the re-election of President Obama, either the states or the Centres for Medicare and Medicaid Services (CMS) will be moving ahead in 2013 to set up web-based HIXes that enable individuals to purchase health insurance.
Opera Solutions is building a software platform for CMS that will detect fraud in the exchanges. CMS will begin rolling out the HIXes in January 2014. The software company will use its expertise in health care claims data detection to bolster operational controls in the exchanges to detect suspicious patterns.
“We are poised to put our advanced machine-learning techniques and pattern-detection science to work in building critical operational safeguards from the inception of these exchanges,” Arnab Gupta, chief executive of Opera Solutions, said in a statement.
A feedback mechanism in the software will be one key component to prevent fraud, Pieter Schouten, general manager of health care solutions at Opera, told eWEEK.
CMS will use big data analytics to protect the exchanges from fraud through continuous monitoring of the web services.
“Any sort of fraud model needs to have a feedback mechanism,” said Schouten. A learning system evolves and allows IT managers to adjust the pattern-detection components for the next day, he explained.
With the exchanges’ potential ability to process millions of applications in a short period of time, the learning has to be “instant and rapid”, said Schouten.
“You could go awry if you build a static fraud model that relies heavily on historical data but doesn’t adjust every day, or is rules-based and doesn’t adjust to changing patterns or is without human input,” he explained. “That could be a real problem.”
Opera’s lab environments will allow the company to simulate fraudulent behavior that could occur. If-then logic in the company’s software flags suspicious patterns, or outliers.
“You can take a pool of data and say this is what the data is going to look like,” said Schouten. Opera will create outliers in synthetic data streams and train software models to catch them, he said.
The company’s R&D group constructs artificial anomalies to anticipate fraud that may occur in actual data. “Since you don’t have any live data until you launch, that’s important in order to get ready,” said Schouten.
Analytics software could also check for anomalies in transaction data, such as risk adjustment.
“Learning will be very, very rapid, and you have to be quick on your feet to incorporate any learning, and that gets fed back into the models,” said Schouten.
Fraud in HIXes might consist of identity theft in which perpetrators receive tax credits for the insured person’s premiums, said Schouten.
“There’s some attempt to make up data to qualify for benefits,” he said, drawing comparisons to fraud that occurs with unemployment benefits.
Opera’s software can also detect unusual charges that appear for medical procedures, Schouten said. “We’ve developed a proprietary approach to identify any unusual discrepancies or claims patterns,” he said.
CMS appears ahead of the states in recognising a need for big data analytics to identify fraud in HIXes, said Schouten.
“At the state level, they are building exchanges and have purchased exchanges, but I don’t yet see the specific analytic approach,” he said. “Maybe the states are a little behind; they have different processes.”
Although suspicious activity may or may not exist on the eligibility and application side, the presence of medical claims as part of insurance exchanges makes fraud a potential threat, Schouten noted.
Are you a security pro? Try our quiz!
Originally published on eWeek.