Health Tech, Artificial Intelligence

Mayo Clinic finds algorithm helped clinicians detect heart disease, as part of broader AI diagnostics push

Mayo Clinic shared results of a study showing that an AI tool developed by the system could be used to improve diagnosis of low ejection fraction, a type of heart disease. It’s part of a broader push by Mayo Clinic to commercialize AI-based diagnostic tools, starting with spinout company Anumana.

After spinning up a pair of companies to commercialize remote diagnostic tools, Mayo Clinic shared the results of a study indicating an AI algorithm could be used to improve diagnosis of a certain type of heart disease.

The algorithm, which was developed internally at Mayo Clinic using a type of neural network, was designed to screen for low ejection fraction, a condition where the heart can’t contract strongly enough to pump out at least half of the blood from its chamber.

While an echocardiogram can be used to diagnose this condition, it can be costly and time-consuming. Prices can run as high as $1,830, according to an analysis conducted by UnitedHealth Group. The algorithm would use an electrocardiogram (EKG), a common type of screening tool that measures the heart’s electric signal, to screen for potential cases of low ejection fraction.

Although the pricier echocardiogram would ultimately still be used to diagnose patients, the goal was to identify patients who would have otherwise slipped through the cracks, said Dr. Peter Noseworthy, a cardiac electrophysiologist at Mayo Clinic, in a news release. 

Primary care clinicians at 45 facilities across Minnesota and Wisconsin used the algorithm in their practice as part of the eight-month trial. During that time, more than 22,000 patients received an EKG as part of their care. They were randomized into two groups, where patients’ clinicians had access to the AI results, or where they continued with usual care.

In total, the AI flagged 6% of patients, and increased the overall diagnosis of low ejection fraction by 32% compared to usual care, according to results published in Nature. In other words, it yielded five new diagnoses for every 1,000 patients compared to the standard of care, stated Xiaoxi Yao, the study’s lead author and a health outcomes researcher in cardiovascular diseases.

Notably, the study didn’t just consider the effectiveness of the algorithm itself, but how well it worked in medical practice. It also evaluated which care teams used the algorithm most and who used it the least, a feedback loop that isn’t always available when diagnostic algorithms are implemented.

“The takeaway is that we are likely to see more AI use in the practice of medicine as time goes on,” Noseworthy said in a news release. “ It’s up to us to figure how to use this in a way that improves care and health outcomes but does not overburden front-line clinicians.”

Mayo Clinic licensed out the algorithm to Anumana, a startup it spun out to commercialize algorithms using patient data from remote monitoring devices. It was created as a joint venture between Mayo Clinic and longtime partner Nference. They recently poured $25.7 million in funding into the startup.

Although the algorithm can’t be used in routine practice without the FDA’s blessing, the company has submitted it to the agency for review. In the future, Mayo Clinic plans to develop additional AI-based diagnostic tools for cardiology through Anumana.

Photo credit: ismagilov, Getty Images