As rising healthcare costs comprise an ever-increasing share of employers' expense ratios, predictive modeling is being touted as an effective equalizer. By using formulas to identify the biggest cost factors among health plan populations, employers and payers can work with providers to establish interventions and treatments that keep rising expenses at bay, predictive modeling authorities say.

A statistical decision support tool, the predictive modeling application takes existing population data and processes it through an engine to predict future trends. The financial sector first used it to identify risk factors stratified across various demographic groups, and actuaries have reportedly found predictive modeling to be a valuable tool in determining pricing and underwriting. The results generated by predictive modeling programs are said to accurately reflect the old 80/20 rule – that 20 percent of the population will utilize 80 percent of the resources. Applied to healthcare, it means that the greatest cost will come from a relatively small group most likely to need services: patients with chronic conditions such as diabetes, asthma, congestive heart failure and chronic obstructive pulmonary disease. The Center for Studying Health System Change estimates that about 57 million working-age Americans between the ages of 18 and 64 are beset with a chronic condition of some type.

The key to rolling back healthcare costs is in identifying those with a proclivity toward chronic conditions before the diseases become full blown, said Wendy Wilson, MD, senior manager with New York-based Capgemini Health.

"By predicting which plan members are most likely to get into trouble, you're more likely to help them," she said. "If you wait until they're high cost, it's too late."

Julie Meeks, Ph.D., whose Indianapolis consulting firm provides predictive modeling software to self-funded employers, says that by the time customers seek predictive modeling for cost relief, they want it immediately. The Haelen Group CEO contends that the cost consumption rate is probably more like 70/10 – that 10 percent of the population consumes 70 percent of the costs.

"If the name of the game is to impact cost trends in the first year, it sets a very different standard in identifying the 10 percent," Meeks said. "We've been perfecting each phase of that model. We find with a high degree of accuracy who those 10 percent are."

Experts agree that the root causes of chronic conditions are behavioral in nature and that prevention is the best way to stem the problem.

"It comes down to personal decisions on how to care for ourselves," Meeks said. "Poor lifestyle choices, lack of recognition and early warning signs pose tremendous costs for the payer."

Predictive modeling programs with advanced functionality can also direct successful patient interventions, said Michael Cousins, Ph.D., vice president of health informatics for Richmond, Va.-based Health Management Corp.

"The first generation of tools picks out the people who are likely to cost more than average, but it doesn't tell the case manager whether the patient is going to be affected by a particular treatment regimen," he said. "For example, if John Smith has diabetes, has been to the hospital several times, has lost a leg and has congestive heart failure, it is unlikely a program like ours is going to have much impact on him. By contrast, Jane Doe is also a diabetic, but is younger and without the co-morbidities. Our program can figure out a care plan for her."

While acknowledging that prevention is a laudable goal, Cousins maintains that intervention offers a chance to make an even greater impact by reducing the number of patient ER visits and re-admission, which ultimately add up to major dollars.

"We're giving the nurse an opportunity to change the future," he said. "It's one thing to identify problems, but quite another to find out who in the near future is likely to develop a serious complication due to the condition. We're giving nurses a heads-up on who to look out for."

Getting a good read on a population usually takes three years, Wilson said, with the first sample yielding cursory data, followed by more detailed information added during the second and third years.

As the concept gains favor in healthcare, predictive modeling companies are being asked to do more, Cousins said.

"There have been predictive modeling ‘bake offs' where they demonstrate how their models perform," he said. "One major health plan brought in six companies together and told them to figure out who their most expensive members were. They are to be commended for that."

The provider community, while not a fiscal beneficiary of predictive modeling, still plays an integral part in the equation because modeling programs foster collaboration and encourage smarter consumer healthcare, Wilson said.

"The reason why predictive modeling is a good tool is that it is in everyone's interest to come together," she said.Link 1 Text: Center for Studying Health System Change tracks healthcare costs.