General Business Issues

Featured Health Business Daily Story January 30, 2008

More Health Insurers Are Using Predictive Modeling to Set Premiums, Not Merely to Predict Costs

Reprinted from HEALTH PLAN WEEK, the industry's leading source of business, financial and regulatory news of health plans, PPOs and POS plans.

More health insurers are adding predictive modeling tools to their underwriting arsenal, using the software to help set premiums for new and renewing employer groups. Experts say it's a fairly easy transition for health plans that already use predictive modeling applications as a tool to aid disease and case managers in identifying members at high risk of future medical costs. Despite that, many insurers limit the tool's use to midsized groups, such as those with between 50 and 500 employees.

Predictive modeling software programs analyze a few years of past medical claims data, as well as lab, pharmacy and other records where available, to predict how expensive an enrollee is likely to be in the future. Often, the most expensive patients in terms of past claims history aren't those most likely to cost a lot in the future. For example, a patient who was in a car accident might have high costs one year but low costs the next, while a patient with diabetes and heart disease might incur only pharmacy and physician office visits one year, but have multiple hospital stays the next.

LifeWise Health Plan of Oregon and Arizona, a subsidiary of Mountlake Terrace, Wash.-based Premera Blue Cross, started using predictive modeling as an underwriting tool for its association business in 2001. Late last year, the insurer expanded the software's use on an ad hoc basis to employer groups with 51 or more workers, says Sharon Howe, the plan's director of underwriting.

Excellus BlueCross BlueShield also has begun testing the use of predictive modeling software for underwriting on a case-by-case basis, says Martin Presberg, director of that insurer's rating support team.

Health plans most often start by using predictive modeling in their disease and case management departments, says Swati Abbott, president of Orlando, Fla.-based MEDai, Inc., a predictive analytics company. As companies grow more comfortable with the tool, underwriters start pushing to access it as well.

Presberg says the predictive modeling tool makes underwriting analysts' jobs easier. Previously, "analysts would have had to do a number of queries into a number of sources [to access medical and pharmacy data] for high-cost claimants…that would have impacted decisions on a ratings basis," he explains. But Excellus's predictive modeling vendor, MEDai, analyzes the insurer's claims and returns a Web-based tool in which claims experience already has been aggregated and analyzed.

Abbott estimates that fewer than half of insurers use predictive modeling for underwriting applications. Typically, insurers focus on small to midsized groups, such as those with between 10 and 500 workers. "That's a small enough population that one or two [patients] could drive significant variation" in medical costs, she says. Health plans generally already have a proven underwriting formula for large groups, in which medical spending is less affected by the health status of individual enrollees. And insurers typically do not use the programs to set rates for very small groups or individuals.

Since many insurers already use predictive modeling for medical applications, an underwriting application can be added "with minimal additional work," Abbott says. She estimates that a 100,000-member health plan might spend between $30,000 and $100,000 to purchase and install a predictive modeling system, assuming that it already has a system installed for disease management functions. The cost range is quite broad because there is wide variability in the services offered by predictive modeling companies, Abbott adds.

LifeWise Manages Association Risk

Predictive modeling is "just one more risk factor that's used in pricing the group," Howe says. "You have to weigh all three of those models [demographic data, past claims experience and predictive modeling] to get a good blend of what your projection is. It's that kind of 'art' part of underwriting."

LifeWise started using predictive modeling to "manage risk within associations," Howe says. "Associations really are small-group pools…with an average group size that's usually between five and 10 employees." But associations do not have premium rates regulated by the state, Howe notes, unlike LifeWise's actual pool of small groups with state-regulated rates, which might total 35,000 members. "So we want to make sure we price them [i.e., associations] appropriately so that you don't get poor risk coming out of the small-group pool into the association pool."

Rate-setting practices vary by the association and how it's set up to offer insurance, Howe says. But LifeWise typically begins by setting a starting rate for the association using demographic data, Howe says. "Then we look at the [medical] loss ratio and large claims" to adjust the rate, and finally incorporate predictive risk scores to establish a rate range for the entire association.

She adds that because Oregon is making changes to its regulation of association insurance rates, LifeWise will stop using the tool to price associations in that state.

LifeWise in November 2006 started using predictive modeling "on a discretionary basis" for groups of 51 or more workers. "We actually use it on a blended basis up to about 200-plus lives," she says. "We've definitely used it on all [case] sizes in considering an adjustment." But, she adds, "it's just one of the pieces of information" the plan relies upon.

The insurer uses Impact Pro, an episode-based predictive modeling and care management program sold by IHCIS, a unit of UnitedHealth Group subsidiary Ingenix. "We use the same program as our nurses use to decide who they're going to case manage," she says. Sometimes, LifeWise goes a step further in making underwriting decisions, actually calling the nurse case manager to see if current interventions under way with the patient are likely to lower costs beyond what Impact Pro has projected. Of course, "you can't do that on every case," Howe adds.

Howe says LifeWise has noticed an improvement in the predicted risk for its entire book of business from last year to this year. "I don't know if it's [predictive modeling or] a combination of everything we've done in underwriting," she says.

Excellus Dips a Toe in the Water

Predictive modeling tools are "well-established in the health care affairs, medical benefit management, case and disease management kinds of areas," Presberg says. "We've been using the tool in that area primarily, for case identification, for several years."

But in terms of using predictive modeling as an underwriting tool, "it's very early on in the process," he says. Excellus now "is putting in place the processes and analysis that will tell us whether it's going to give us better data, and will tell us whether or not it's going to improve our workflow."

For example, the insurer prepared a proposal for a large employer "on how we might choose to administer benefits and how we might structure the financing. One of the tools we used internally was the predictive modeling tool," which Excellus used to test assumptions made using more traditional underwriting methods, Presberg says.

At this point, however, Excellus still is using the tool only in special cases. "We've not really automated that kind of analysis so we can do it on a broad basis for many groups.…At this point, because we are early in the process, it's more about giving us a comfort level with decisions that we already made."

Presberg adds that New York state rate regulations bar the insurer from using any form of medical underwriting on groups of 50 or fewer employees. So Excellus is limited to using the tool on groups of 51 or more workers.

Advertise With AIS

Privacy

Site Map