Tech and tenderness: TPA uses predictive modeling based on artificial intelligence, wrapped with personalized nurse intervention, to lead high-risk patients to medical care

Health Management Technology, May, 2004 by Robin Blair

On the surface, predictive modeling might not light up the sky the way wireless devices or voice recognition systems do. But make no mistake: Its impact is so far-reaching that soon it will influence and possibly direct the healthcare services many of us receive.

Disease management (DM), with its predictive modeling corollary, have of late come under increased scrutiny. The DM concept, of course, made perfect sense in the beginning: Manage those patients with chronic, high-cost conditions to avert admissions or emergency episodes and. hence, reduce costs. But how does a health plan with 500,000 members actively influence 10 percent of its membership with diabetes, congestive heart failure and asthma in such a way that the intervention moderates behavior that, in turn, moderates costs? That's 50,000 people and a lot of intervention. How much smarter might it be to predict who could be in the high-risk population years before they get there, and modify their behavior now?

The advent of more sophisticated predictive modeling technology has turned attention to high-risk health plan members who aren't yet high-cost patients, but soon will be without effective intervention and lifestyle changes. These may be patients in the early throes of utilization, as well as consumers who aren't yet active utilizers. Experts postulate that 2 percent of a health plan's membership may be at very high risk and may, eventually, drive 70, 80 or even 90 percent of the health plan's expenses. Those numbers alone make it a problem worthy of the IT microscope.

Artificial Intelligence

HealthSCOPE Benefits Inc. (HSB), headquartered in Little Rock, Ark., is a full-service health management and claims administration company--a third party administrator. It provides administration, claims and support services to self-funded employers. Its 250 employees manage more than $400 million in annual healthcare expenses, servicing about 140,000 members.

As part of a suite of medical management services that includes disease management and case management, HSB also offers High Impact, a predictive modeling-based program that proactively identifies tomorrow's high-risk. high-cost patients. Today, about 40 percent of HSB's total membership is eligible for the High Impact program.

The technology foundation of High Impact is Risk Navigator Clinical, a forecasting tool from Orlando, Fla.-based MEDai (Medical Artificial Intelligence). The technology foundation of Risk Navigator Clinical is MITCH (Multiple Intelligent Tasking Computer Heuristics), a prediction engine that allows health plans and TPAs to not only predict which members are tomorrow's high-cost patients, even if they are not now heavy utilizers, but also to predict which individuals in the high-risk population are likely to respond to intervention.

MEDai CEO and co-founder Steve Epstein says the technology is third-generation. It started with rules-based systems, he says, then progressed to groupers, and now systems are heuristic. They have become "smart."

"It's a tremendous advantage," says Epstein, "for a health plan to take subsets of a populations and say 'Five years down the road, if you continue your current lifestyle, you run an 80 percent chance of developing heart disease or lung disease. Here is what you can do now to prevent that.' It's an even greater advantage for the health plan to be able to predict which members are likely to respond to its intervention."

Juggling Data

To forecast risk, MITCH uses a varied collection of administrative data--encounter claims, pharmacy claims, lab data and membership files--as well as survey or evaluative data such as health risk assessments. With earlier renditions of predictive modeling programs, accurate predictions couldn't be generated if some of the data were weak or missing but no longer.

MITCH's mission is to spot high risk from miles away. Being able to deal with linear and nonlinear data, "MITCH can compare a consumer sitting at home, with no prior hospital admissions or ER episodes, to his neighbors with similar backgrounds, ages, zip codes and insurance coverage, and can generate an accurate prediction," says Epstein.

Equally important, he says, is the system's ability to identify the drivers. "It's not enough to identify the risks," he continues. "Care management programs must know the drivers of risk. Is it a drug, a lab value, or something in the patient's history at the foundation of his risk score?"

Modeling with artificial intelligence (AI) also makes crystal clear the difference in value of healthcare dollars spent. In their article, "Predictive Modeling in Health Plans," authors Randy Axelrod and David Vogel cite as an example two health plan members that each generate $3,000 of service utilization in a plan year. One member generates those costs in a single episode of hospital treatment; the other, through months of prescription utilization. While the dollar amounts might render similar financial profiles, AI-based predictive modeling would manifest dramatically different risk forecasts for those patients.