KM challenges fraud

The term "risk management" can mean many things, depending on the type of organization and the activities being conducted. It can relate to anything from governance and financial management to physical safety of workers. One component of risk management that appears in many industries is the detection and prevention of fraud. Financial institutions, healthcare insurance companies and government institutions all have been victims of fraud. Eventually, consumers pay the costs of fraud.

The National Healthcare Anti-Fraud Association (NHCAA) estimates that fraudulent payments for healthcare services accounted for at least 3 percent of the $2.26 trillion in total health expenditures in 2007.

"This number is based on estimates from experienced investigators in the field," says Louis Saccoccio, executive director of the NHCAA. "The problem is challenging, because with 4 billion to 5 billion claims filed each year, it’s impossible to look at each one."

Some estimates place the fraudulent payment rate as high as 10 percent of total expenditures. With the resulting number ranging from a low of nearly $70 billion to a high of over $200 billion, the incentive to prevent the loss of that revenue is evident.

Fraud in healthcare insurance includes a range of actions, from charging patients for services that were never performed, to "upcoding" a diagnosis or service so that it falls into a more expensive category. In some cases, providers and patients conspire to defraud healthcare insurance companies, and in others, the provider bills for services to non-existent patients. Business intelligence tools that can perform complex analytical tasks are often used to detect patterns of abuse. Those patterns include services that do not match typical actions for the providers’ population group, or claims from providers that are higher than average for specific services.

At Highmark, a Blue Cross/Blue Shield affiliate, investigators were working vigorously to analyze claims data to find anomalies that might indicate fraudulent activities. However, they were slowed by analytic tools that consumed large amount of resources to run reports and extract data.

"It took weeks to get results," says Tom Brennan, director of the Special Investigations Unit at Highmark. "We knew we needed to work better and faster."

The unit teamed up with Highmark’s informatics group, which was already using SAS Enterprise Miner and other predictive analytics tools from SAS, and built an application called Financial Investigative Reporting System of Tomorrow (FIRST). FIRST creates more than 40 standard reports based on about five years of data. The reports can be customized by individual analysts to focus on different aspects of the data.

"We can now do in hours or even seconds what used to take days," says Brennan. "In addition, we can archive reports so that they don’t have to be re-run, which is very useful."

The system examines factors such as a patient’s age and diagnosis and, using data mining capabilities from SAS, looks for inconsistencies such as treatments that do not match the diagnosis. Another example is the "impossible day" scenario in which a provider reports more encounters than could be completed in a day.

Although healthcare insurance companies compete with each other to obtain customers, they cooperate to detect and prevent fraud. Therefore, they share information about fraudulent schemes and statistics on providers if problems are indicated. Some frauds end up affecting many carriers. In one recent case, nearly 20 carriers had exposure to a neonatal physician who allegedly billed up to $22 million for services that were not delivered.

One future goal for Highmark is to detect fraudulent claims prior to payment, because recapture of payments is difficult. "By law, we have to pay the claim within a specified amount of time," notes Brennan, "so in order to look for fraud before paying the claim, we need to move very quickly."

Predictive modeling looks for the same types of anomalies as retrospective analysis, but on a near real-time basis. Highmark expects to be able to do that within the year, and is currently focusing on how the analytical steps integrate with the workflow for processing the claim.

The analytic suite from SAS now includes a social network analysis tool that shows how doctors, patients and pharmacies interrelate. For example, a patient might get a prescription for a controlled drug and duplicate it, then fill it at multiple pharmacies. Or the patient might go to multiple doctors for a prescription, to feed an addiction or obtain drugs to sell.

"The SAS solution can aggregate the claims and present a visual representation," says Rick Pro, health plans principal in the Health and Life Science Global Practice at SAS, "so the insurance company sees the network of connections." Such aggregation can also pick up cases in which a single instance might not raise a red flag, but the combined dollar amount would reach a threshold and initiate an investigation.

Detecting unemployment fraud

Although some software is specifically designed for fraud detection, business application software intended primarily for other purposes such as content management can also be used to identify inconsistencies in data that might indicate fraud. This capability stems from the fact that different data elements and documents can be checked nearly instantaneously in electronic systems, as opposed to the laborious process of performing similar crosschecks manually.

Government agencies are the targets of fraud partly because they handle large amounts of money and they interact with thousands of individuals. The state of New York, for example, pays about $2 billion in unemployment benefits per year. In the first nine months of 2008, the state made 600 arrests and recovered more than $16 million in fraudulent payments. Texas recovered $1.4 million as a result of prosecuting unemployment insurance fraud.