Company uses Mayo data to teach digital tool to detect heart problems.
As you read these words, your heart’s left ventricle is pumping out as much blood as it can. Hopefully it’s enough.
If your heart is working normally, its left ventricle is pushing out 55 to 70 percent of its total blood volume with each beat. If this “ejection fraction” falls below 40 percent, you may have at least mild heart failure and may feel winded walking up stairs. But cardiologists have long sought a quick and inexpensive way to detect people whose heart is performing between those ranges and can be treated before their condition worsens.
Now the Mayo Clinic is teaming up with the California medical device company Eko to conduct a clinical trial to see whether a new kind of digital stethoscope can be used as an early screening tool to detect compromised hearts. Heart failure affects 6.5 million Americans today, and many of them have asymptomatic left-ventricular dysfunction that could potentially be detected with the system.
If it works, this digital stethoscope would join a fast-growing list of medical devices that are producing medical insights more quickly and cheaply than traditional technologies. Artificial intelligence, tiny sensors and cellphone communications are rapidly being incorporated into devices and applications for uses in diabetes care, radiology, mental health, cardiology and many others.
The U.S. Food and Drug Administration has cleared many digital mobile medical technologies, and health care innovation centers like Minnesota and Silicon Valley are teeming with digital health care startups whose medical offerings may look more like phone apps to the untrained eye.
The Eko medical device company in Berkeley already has FDA approval for its digital stethoscope, the Duo, which gathers traditional acoustic data from the heart and also incorporates a single-lead electrocardiogram to expand a physician’s insights into heart function.
Mayo is working with Eko to put the device to a new use, which will eventually require its own FDA clearance.
“The core concept is, can we detect silent disease, or someone who is going to develop disease, in a way that is actionable so we can improve health?” said Dr. Paul Friedman, head of cardiac medicine at Mayo. “That’s our goal.”
Studies show that patients with asymptomatic left-ventricular dysfunction have significantly increased risk of symptomatic heart failure and mortality. The longer the condition remains undiagnosed, the more time the heart has to change its shape in ways that are associated with future health problems, especially after a heart attack. The goal is to halt or even reverse those changes as soon as possible.
The idea of detecting low ejection fraction with a fancy stethoscope combines technologies already familiar in digital medicine, including a scope that transmits data, an artificial intelligence algorithm that learns from old health data, and a smartphone app that relays data to the AI.
The most novel element in the project may be the addition of Mayo’s expansive electronic health record system.
Mayo has been keeping comprehensive patient records in standardized digital formats for decades, along with signed consent forms from patients allowing the data to be used in future research. Those data are invaluable sources for training artificial intelligence, it turns out.
For heart failure patients, the electronic records often include matched pairs of traditional electrocardiogram (ECG) readings showing the electrical activity in the heart, plus the definitive echocardiograms (“echos”) that are traditionally used to diagnose ejection-fraction problems.
“You have these pairs of ECGs and echos,” Friedman said. “The echo is the expensive test and the ‘right’ answer, so to speak. And then we have the ECG, which is the inexpensive and ubiquitous test. And we want to train the computer to read the ECG. So we say, here’s the ECG [from the old medical record] and here’s the right answer. And each time we do that, it is learning the little individual characteristics.”
According to a presentation at a cardiology conference in March, the system was trained using 45,000 matched sets of ECGs and echos from Mayo’s data vault, and then tested on data from another 52,000 patients. The full study data haven’t been published in a peer-reviewed journal yet, but Friedman said the system worked surprisingly well, with an 86 percent accuracy reported in the March presentation.
The system flagged 1,300 cases that at first appeared to be “false positives” — cases where the AI thought the ECG suggested low ejection fraction, but the echo from the same time did not confirm it. But researchers then looked at Mayo’s long-term follow-up data and discovered that these “false positive” patients had a fivefold increase in risk of having an abnormal ejection fraction reading within five years.
“So the heart muscles are having some electrical problem that is subtle that the ECG is picking up, but it’s so early that the heart pump is still strong,” Friedman said. “It’s detecting [heightened risk] before the echo even shows a weak heart pump.”
Despite its name, Eko’s stethoscopes do not take echos. The hope is that, if the ejection-fraction algorithm is accurate enough to gain FDA clearance, the device can use its single-lead ECG reader and the Mayo-educated AI to find patients whose hearts deserve a closer look with a comprehensive echocardiogram.
“What Mayo Clinic was able to do was connect these two data sets that were otherwise unconnected and find the hidden patterns,” said Connor Landgraf, co-founder and CEO of Eko. “This doesn’t supersede the physicians’ judgment. This is just a way for us to provide additional information to a physician when they are thinking about making a referral to echocardiography.”
The Eko Duo stethoscope is a prescription device, which means patients can use it at home only under active care of a physician. The Eko Duo system currently retails for $349 on the Eko website and pairs with iOS and Android devices. Last week, a study found that the same device could be paired with an AI algorithm to detect pediatric heart murmurs as accurately as a cardiologist; that use will also be submitted to the FDA for clearance.
Landgraf said Eko agreed to pay licensing royalties to Mayo for the ejection-fraction detection system, in addition to Mayo’s ownership stake in the company. Private tech investors, traditional medical device companies, and health systems have also invested. Asked whether the stethoscope’s price can generate gross margins seen in traditional med-tech, Landgraf said Eko’s business model isn’t focused on high-margin hardware.
“We focus really heavily on just getting the hardware into the market, to help us build data sets and then also to help us build software and algorithms in the future,” he said. “Having ubiquity and access to the device is far more important for us to be able to build the data set, build the AI, and then leverage that as a software-as-a-service.”