A dose of optimization for rising healthcare costs

The growing costs of healthcare and health insurance, says Aurélie Thiele, have opened up opportunities for engineers who are well-versed in the art of optimization.

Thiele, associate professor of industrial and systems engineering, is responding to the challenge by developing models that optimize the relationships among costs, risks and the options offered in employer insurance plans.

Despite vaults full of data and decades of experience processing claims, she says, the health insurance industry has done relatively little to optimize data to control costs.

“Human resource analysts develop plans that are relevant to their patient populations,” says Thiele, “but without state-of-the-art quantitative tools there has been a lot of guessing.”

Thiele is on sabbatical at MIT working with Dimitris Bertsimas, the Boeing Professor of Operations Research at the Sloan School of Management, to optimize the design of health insurance and reinsurance policies from the perspective of employers offering a menu of insurance policies for workers to choose from.

Based on previous claims data and the number of plans an employer wishes to offer, the researchers are developing a mathematical model that assesses parameters such as deductible size and coinsurance. They are examining whether reinsurance, in which an employer shifts the risk of inordinately large claims to a reinsurance company for a fee, can help decrease the premiums paid by employees.

Their goal is to help companies control costs, maintain fairness for lower-paid workers and protect against rare but expensive medical conditions.

Optimization uses big data and computational finesse to look at problems with many variables and achieve the best possible outcome, says Thiele, who co-directs Lehigh’s M.S. program in analytical finance.

Choosing a healthcare plan can be daunting. Employers can select from as many plans as they want but usually offer workers only a handful of options to choose from. Some public health exchanges, on the other hand, make as many as 80 plans available.

“The literature says that too many choices paralyze customers,” Thiele says. “There is a moment when it becomes too much information for the average person.”

The proof-of-concept model that Thiele and Bertsimas are developing slices through data to address key questions: What options can be offered given the expenditures made by employers? What expensive conditions affecting only a few patients can be reinsured to spread risk over larger pools?

These questions involve large amounts of nonlinear data and are well-suited to optimization techniques, Thiele says. But engineers should have a thorough understanding of a question before running a data-analysis program.

“If you aren’t clear about what you are going for,” she says, “the data-crunching can go on forever.”

Thiele and Bertsimas’s preliminary findings show that companies’ allocation of expenditures—between preventive and sick care, or pharmaceutical and surgical intervention—is critical.

“High-level decision-making is more important than the precise values of policy parameters,” she says, “because it drives employee behavior.” For example, helping employees feel secure seeking preventive care can reduce sick days, costlier treatments or hospitalization down the line.

“It turns out that deductibles and reinsurance are the most important policy parameters,” Thiele says. If deductibles are too low, employees don’t have enough of a personal investment. Maximum out-of-pocket limits, however, while reached by only a few, make a real difference in the lives of patients who are already sick.

Thiele’s model also helps employers determine how much to reinsure against expensive cases that rarely occur. Reinsurance spreads these low-probability risks over a larger pool, she says, and evidence shows that it keeps premiums down for everyone in a group.

So far, Thiele and Bertsimas are working with data from known care and cost trends. Their model is open, allowing companies to understand how it will use their sensitive data. The researchers are also breaking their model into small parts so that practitioners with different experience in insurance, surgery or family practice can see how it is relevant to them.

This approach is new for Thiele’s branch of engineering. “A lot of industrial engineers look at healthcare as if patients were products in a warehouse,” she says, “because they understand those flows very well.”

Addressing the nation’s health insurance challenges, however, requires more than a simple application of existing technology.

“Optimization improves what we are able to do by a significant factor. We have to create new models and frameworks to be able to help the health insurance industry and patients.”

Story by Robert W. Fisher ’79