Predictive modeling (PM), by basic definition, is the process of developing health care programs for at-risk populations based on statistical probabilities obtained through advanced software tools. The idea is to target patients who have a good possibility of developing a disease somewhere down the road, and then administer treatment early on in an effort to stop the illness, or at least limit its severity.

Think of the Tom Cruise futuristic movie "Minority Report" - in which he plays a cop in a dystopian society where murders are predicted and prevented before they occur - and you have predictive modeling in a nutshell.

"The purpose of predictive modeling is to identify the patient who is at risk for incurring higher health care costs in the next 12 to 18 months, or who is at risk for hospitalization, but does not otherwise know that they are at risk for such," comments Joel Brill, chief medical officer at Predictive Health, a Phoenix-based company that performs PM analysis of medical and pharmaceutical claims data.

"Think about it," Brill continues. "If you're in the hospital or know you have a condition, you will be generally receptive to a health professional contacting you about the next steps you need to take to improve your health."

However, getting patients to participate in medical programs after they've been identified by predictive modeling can be tricky, since many feel healthy or are unaware of their risks.

"It can be difficult because you're contacting someone who may not know they're on this slippery slope," remarks Leslee Budge, assistant director of population care technology at Kaiser Permanente.

Benefits for employers

Despite participation challenges to overcome, PM has many potential uses for employers and could serve as an aid in underwriting, premium rate setting, benefit design, provider reimbursement, disease management, case management, resource needs assessment and provider profiling.

It requires a data repository (where data is mined) and the selection of a tool that can spin the data and predict outcomes. Data can come from several sources, including individual demographics, pharmacy and medical claims and health risk assessments. Tools are developed by companies like DxCG, a Boston-based tech firm that makes PM software, and Ingenix, a health care IT solutions company based in Eden Prairie, Minn., among others.

Employers interested in testing PM on their employee population should ask their broker or insurance company if they offer the service. If they're self-insured, they should ask their third-party administrator and/or benefits consultant.

First, employers need to know all the players involved, remarks Clive Riddle, president of Managed Care On-Line, a health management resource company based in Modesto, Calif. "If an employer has open enrollment periods with multiple vendors, then there's going to be a lot of data flowing around in a lot of different places. The first thing an employer needs to do is assess where the data resides and how many parties are involved."

Employers should realize that they most likely won't see an early return on their investment in PM, as money spent in the first year is used to identify employees who are not complying with their prescription regimens, experts say.

"The employer must recognize that, for the most part, it can take several years to achieve an ROI. So, the employer needs to have a multiyear commitment to these processes," says Brill.

However, the benefits of PM are real, Brill adds. "When integrating true predictive modeling with other medical management programs, the employer has the opportunity to improve patient care while controlling cost trends over time."

Consulting firm Mercer recently used Ingenix's software as part of a massive study that analyzed claims data from a consortium of 14 employer-clients involving more than 400,000 lives. The purpose was to use PM to find members who had diabetes and other comorbidities. Two Ingenix tools, running on various algorithms, sifted through participants' records and analyzed a dizzying amount of information, including Social Security numbers, date of birth, gender and diagnosis codes on claims. One way to think about how it works is to envision crime television shows where a fingerprint is used to try and find an ID match.

The model determined that there were 9,090 high-risk diabetics with comorbid conditions. Another 13,793 were deemed low-risk, meaning their diabetes was under control and did not shown signs of comorbidities.

The main point of the study was to check the accuracy of claims data received by health management companies from disease management vendors. Russell Robbins, principal and senior clinical consultant with Mercer, says the results were very positive, and the performance of the PM model was impressive.

"This study validates the ability of predictive models to identify people with medical conditions, in this case diabetes, and assign them a risk score," says Robbins. "It also validates the findings of the disease management vendors in terms of their ability to identify the diabetic patients in the employee population."

Another Mercer client looked at employee data generated through PM, predicted costs, and waived copays on generic drugs and reduced by 50% copays on brand-name drugs used to treat diabetes, asthma and heart disease. The result: First-year savings from reduced non-pharmacy medical costs were equal to the copay reductions.

"It's the new thinking now," says Robbins.

"Total health management is more than just wellness. Predictive modeling helps determine not only that employees are physically healthy, but also that they're in work and productive. It's a win for everybody."

Visit http://ebn.podhoster.com/ to download a "Five Minutes With ..." podcast with Mercer's Russell Robbins.