Medical Benefits Fund

Turning the Tide Against Limit Surfers in Australia with SAS

Speeding up the claims process can only be a good thing for health insurance customers, but it presents a new headache for the insurer; the time window for detecting fraudulent claims has become very narrow.

In recent years Medical Benefits Fund (MBF), Australia's largest privately owned health insurance company, has introduced automated claims and payment channels that improve the quality and speed of service for its members, while reducing costs.

MBF AutoClaim enables members to make on-the-spot claims for services while at the healthcare practice, through systems such as HICAPS and Optus HealthPoint. All the member has to do is to swipe an MBF card and the payment is automatically credited to the provider.

In financial year 1999-2000, 27 percent of all claims were already being paid via AutoClaim, and 77 percent of all optical claims. However, while these electronic channels offer significant advantages, they can also present new opportunities for fraudulent claims if the transactions are not monitored for authenticity.

A Family Affair
One of the advantages of the AutoClaim system is that all family members are included on a single card, so customers can make claims for their dependants without needing several cards. The downside is that this opens up the temptation for health practitioners and patients to conspire in what MBF calls "limit surfing."

For example, a member might visit an optometrist to get a new set of glasses only to be told that she has already used up her optical limit for the year. Limit surfing is when the optometrist covers the cost of the glasses by claiming on another family member's unused allowance.

According to Jolie Reichel, Research Manager at MBF, limit surfing occurs in all ancillary claims areas but it is particularly common in optical because members have a set limit for the year which is roughly equivalent to a set of spectacles.

"The problem is that while we know such fraudulent behavior is going on, we don't want to upset members or healthcare providers by investigating them without good cause," says Reichel. "That would do nobody any good, least of all MBF. We have to operate in a very competitive marketplace so it is important to win new customers and ensure we retain existing ones.

"In any case, we don't have a huge field staff who can go out and visit all the healthcare providers and check that everything is above board."

In February 2000 MBF started looking for a solution that would enable it to identify aberrant behavior suggesting a high likelihood of fraud. "As a non-profit mutual organization we have to prove the case for any IT solution very thoroughly, since we are spending our members' money. The search for the right solution to invest in was therefore very thorough." says Reichel.

MBF's strategy was to identify test cases of limit surfing through standard investigation methods, extract transaction-level data from the claims database (including data relating to claims that are known to be fraudulent, and data relating to "innocent" claims). Following an extensive vendor selection process, MBF opted for SAS Enterprise Miner and contacted the SAS South Pacific Data Mining Discovery Center.

"We chose to go with SAS because it could scale to our needs and offered greater flexibility than the competitors," says Reichel. "Equally important, SAS could offer us more than a product. They offered us long-term partnership and knowledge transfer, which gives us a real competitive advantage."

Proving it to the Board
The proof of concept project then started in August 2000. "We had a small window of ten working days to devise a model and demonstrate the benefits to the Board.

"Using stratified sampling techniques, we focused on the top one percent of the available data, in which 'rare events' suggestive of fraud were well represented.

We tested several data mining modeling methods. Logistic regression produced the best results, proving 23 times more likely to detect potential fraud than using just a random audit," says Reichel. "This gave us rules that would enable us to detect fraud very quickly."

The Board was very impressed and gave Reichel's team the go-ahead to implement the optical project, and resources for further investment in data mining.

Reichel stresses that this was more than a vote of confidence in a product. It was also a vote of confidence in a significant investment and in the business partnership with SAS.

"The knowledge transfer from the Discovery Center was critically important. Enterprise Miner is so good that it is very difficult to recruit analysts with experience in the software - once any organization has seen what it can do, they are very unwilling to let these people go," she says. "So the partnership was important in helping us to develop our own in-house skills."

MBF domain experts worked closely with SAS technical staff to develop categories and measures that would help identify patterns of fraud. "We need the expert advice of health practitioners to tell us, for example, which combinations of services, products, patient age and gender make logical sense, and which might indicate fraud. Enterprise Miner offers the flexibility to allow us to build this expert advice into the models," says Reichel.

"A great working relationship was built up between my department, the SAS consultants and our domain experts. This is important because with fraud, when you close one loophole you can be pretty sure another one is about to open up. It takes a cohesive team effort to support the investigators. They love what we're doing with data mining because it means they can always keep on top of things."

By applying the model developed in the proof-of-concept to the top five percent of likely fraud cases, MBF found that it could obtain an improvement of 800 percent over random audits. MBF has now applied data mining techniques across other healthcare areas such as dentistry and physiotherapy, and to other areas of the business such as product and system rules and controls.

In the first six months after implementation of the optical project in November 2000, MBF achieved a return on investment of more than 200 percent. "We have saved approximately AUD 1 million (€ 555,828) on initial projects alone, and this is totally attributable to the work with Enterprise Miner," says Reichel. The full-year forecasted savings are more than AUD 2 million (over € 1.1 million) just through fraud detection, but MBF is seeing additional financial benefits through the application of Enterprise Miner in other areas of the business.

"All of that money goes back to our members," says Reichel.

Enterprisewide Intelligence
"The tangible bottom-line benefits are enormous, but the intangible benefits are even greater," says Reichel. "The knowledge we have acquired has been transferred to other departments. For example, some of our marketing people are now also trained in Enterprise Miner and they are now in a position to start their own team, using data mining in-house to learn about our customers, carry out segmentation analysis and drive communications campaigns."

"Fraud prevention is far better than cure. One of the advantages of Enterprise Miner is that we can build models identifying different levels of fraud. Except in clear cases of theft, prosecution is a last resort. But at the same time we have a duty to protect our members' money by sending a clear message to all fraudsters that we know what's going on, we're watching them and that fraud won't be tolerated.

"Since we have been using Enterprise Miner, we've seen much greater control of aberrant behavior at source. In this way we are tackling fraud without upsetting customers or threatening our market share."

Copyright © SAS Institute Inc. All Rights Reserved.