Predictive Analytics in Health Care: Opportunity or Risk?

The approach taps data mining, stat modeling and machine learning to transform data into predictions. It’s changing medicine, and it’s a force physician leaders cannot ignore.

Medicine has always revolved around probabilities. Whether a doctor is diagnosing a patient or deciding what drug or treatment path to pursue, the decision-making process is based on the best evidence available.

What: National Health IT Week

When: Oct. 8-12.

About: An awareness week focusing on the value of technology in health care.

Online: healthitweek.org .

However, in the digital age, there’s a new doctor in town: predictive analytics. The approach taps data mining, statistical modeling and machine learning to transform historical data into predictions. According to a 2017 survey conducted by the Society of Actuaries, 93 percent of health payers and providers believe that predictive analytics is important to the future of their business. In addition, 89 percent said they plan to use predictive analytics within the next five years.

To be sure, the technology offers clinical promise — particularly in identifying symptoms and steering physicians toward the most effective approach. Yet it also raises questions about the accuracy of predictions and the role of medical professionals. Says Craig A. Umscheid, MD, MS, associate professor of medicine and epidemiology at the University of Pennsylvania Health System: “It’s no magic bullet. It can be extremely helpful or completely useless, depending on how you use it.”

By the Numbers

Organizations are increasingly turning to predictive analytics to address questions about patient satisfaction, costs, patient risks, readmissions , profitability, diagnosis, clinical outcomes and mortality. The Society of Actuaries study found that 47 percent of health care providers now use the technology in some shape or form.

RELATED: Mining Gold in Data to Boost Quality of Care, Trim Costs

In June 2018, Google raised the stakes by reporting that its artificial intelligence algorithm could outperform doctors in predicting survival and death rates among patients. While this information could prove valuable, it also introduces troubling questions. For example, how likely is a physician to avoid live-saving measures if a computer indicates it’s pointless? How does the role of a physician change if computers deliver better outcomes?

Chad Mather III, MD, MBA, assistant professor of orthopedic surgery at Duke University School of Medicine in North Carolina, believes predictive analytics is critical to the future of medicine but it’s not a replacement for doctors and human thinking.

“A physician should always be able to override the recommendations of a system,” he says. Ideally, he says, there are systems in place for peer input and continuous feedback. “Using ongoing data and machine learning, it’s possible over time to continue to improve systems and outcomes.”

It’s a sentiment shared by Umscheid, who believes any computer-generated prediction is simply a starting point for a discussion.

“A prediction is only as good as the data that’s entered into a system. … A prediction of high mortality may simply be a cue to the medical team to assess a patient’s goals or wishes about end of life.”

At the University of Pennsylvania Health System, mortality analytics trigger an “advanced care plan discussion,” he says. “It has proven a helpful tool in guiding care for high risk patients.”

Deep Insights

Not surprisingly, many physicians bristle at the notion of a computer providing input and direction. “In some cases, there’s a belief that it undermines their skills,” Mather says. However, he points out that the growing complexity of medicine demands more data and an ability to learn from data. Medical professionals should view predictive analytics as a valuable tool. “It’s really all about defining and redefining the physician’s role with the focus on arriving at the best possible outcome,” he says.

Umscheid has found that predictive analytics can produce benefits, problems and unintended consequences. This includes teams not knowing what to do with information, misusing information and ignoring information. For instance, when he helped introduce predictive analytics to aid in the diagnosis of sepsis, he discovered that teams often ignored recommendations. The software was successful in identifying the condition, but this didn’t solve the problem.

RELATED: Developing Algorithms That Can Prevent, Diagnose Illnesses

“Teams didn’t know what to do because the software wasn’t necessarily detecting active clinical deterioration. It was simply predicting that sepsis could occur at some point in the future,” he explains.

It’s possible to take aim at the challenge by focusing on both analytics and implementation. For instance, Duke University School of Medicine is designing an “appropriateness calculator” that uses predictive data to display green, yellow and red indicators for guiding teams and their discussions surrounding joint replacements.

“The goal is to review the individual patient’s situation and seek additional input from other peer medical professionals when the system identifies certain conditions or criteria,” Mather says.

Make no mistake, predictive analytics will change medicine. Mather says physician leaders and medical practitioners must explore the technology, create test cases, pursue a rigorous implementation strategy, and integrate it into workflows and processes when predictive analytics demonstrates results.

“There is a lot of hype surrounding predictive analytics, but it’s a legitimate tool. It’s important to experiment with it and find the use cases where it can drive improvements,” Umscheid concludes.

Samuel Greengard is a freelance business and technology journalist based in Oregon. This article is part of a series of stories AAPL is posting to bring awareness to U.S. National Health IT Week .