The unexpected ways AI is impacting the delivery of care, including for COVID-19

Is it possible for healthcare to predict the future?

With the help of artificial intelligence, it just may be, according to Paul Friedman, M.D., the chairman of the cardiovascular department at Mayo Clinic. 

Friedman’s team has trained an AI-algorithm embedded into standard electrocardiogram tests to detect which patients have weak heart pump, Friedman said speaking during a recent Fierce AI Week event.

When researchers went back and performed additional tests, it appeared the the heart pump was normal and that the AI had “false positive.” 

But, Friedman said, when the researchers looked at those same patients patients five years later, they saw a five-fold increase among those patients for having a weak heart pump, he said.

WATCH THE ON-DEMAND PLAYBACK: Unexpected ways AI is transforming the delivery of care

“The physiological signals are affecting the electrocardiogram in subtle ways,” Friedman said. “If human beings could remember the patterns from 100,000 ECGs, we would be able to identify those and say: ‘Oh, there’s a problem coming.’ So it looks like it’s looking into the future.”

The Food and Drug Administration recently granted Emergency Use Authorization to Eko to use the technology to search for physiological problems in the heart associated with cases of COVID-19, Friedman said. 

It's just one example experts pointed to during the recent virtual Fierce Healthcare event about how machines are beginning to outshine their human counterparts as AI is applied to an increasing number of healthcare tasks.

Researchers at UPMC are using AI and natural language processing to examine questions around how to reduce the amount of effort that goes on in healthcare, such a measuring healthcare quality, said Rebecca Jacobson, vice president of analytics at UPMC.

This creates some powerful opportunities, she said.

But it also raises one of the major challenges that has emerged in the AI space: validating how effective AI algorithms are compared to having a human dig through medical records by hand.

“We find mistakes that look like disagreements where it turns out actually the machine found things the humans did not. That happens a lot,” Jacobson said.