Machine Learning

The next wave of artificial intelligence is making critical decisions in health care
Photograph by Chris Hedly

Machine learning. It’s the branch of artificial intelligence capable of identifying who is likely to be a no-show for their next clinic appointment or who is at risk for fatal medical conditions. Based on the idea that computer systems can learn from data, identify patterns, and make sound decisions with little human intervention, machine learning is shaping the future of health care.

“The best summary is that wherever a human makes an important decision, machine learning is being discussed, and is increasingly being used to improve those decisions,” says Dr. Andrew Rosenberg, chief information officer for Michigan Medicine, which includes the University of Michigan’s health system and medical school. In his role, Rosenberg has managed the integration of multiple enormous data platforms. He has also identified clinical and administrative problems, and matched them with the right type of machine learning to solve them. “What we want is to enrich our ability to make predictions of what will happen using known data points and data points that we don’t know are associated yet, but the machine does.”

Advancements in machine learning are quickly moving across medical specialties. It can identify which patients in an emergency department may have potentially life-threatening conditions such as sepsis, the body’s response to infection, and advise medical staff to test their blood. It has predicted abdominal aneurysms based on genetic abnormalities seen on whole genome sequencing with 69 percent accuracy. Abdominal aneurysms are almost always undetected until fatal — or, detected accidentally during a scan for some other ailment.

Wherever a human makes an important decision, machine learning is discussed.
—Dr. Andrew Rosenberg

In the realm of mental health, machine learning was used in a small study published in 2017 with 79 participants at Carnegie Mellon University and the University of Pittsburgh to analyze functional MRI data sets to look for brain activity signatures among young adults with suicidal thoughts. It was also able to distinguish those who had previously made a suicide attempt from ones who hadn’t with 94 percent accuracy.

The digitization of health records, along with savvy computer scientists and high-performance computing clusters, is important to making the use of machine learning in medicine possible. The money that offers incentives and penalties for hospitals to adopt electronic versions of health care records that were previously largely kept on paper was from the Health Information Technology for Economic and Clinical Health Act, a part of the American Recovery and Reinvestment Act of 2009. Both laws were enacted in response to the Great Recession. Banking, retail, and manufacturing industries, which went digital earlier, have been using machine learning much longer.

“There is much more digital data now that is amenable to machines using it. That’s why we’re seeing advanced analytics like machine learning in medicine,” Rosenberg says. Future uses for machine learning in medical care could potentially include predicting which child is likely to develop obesity, who’s at risk to have a heart attack or stroke, or for personalized medicine identifying the combination of drugs, surgery, and radiation that would be best to treat particular types of cancer.

Rosenberg says machine learning may even have the potential to equalize medical care. “Wouldn’t you like to know that if you’re being taken care of at a small community hospital that the machines, as they get to be a bit more intelligent, are now providing the same level of expertise and care that you would get at the very best health care system in the world?”

Facebook Comments