Apple caused quite a stir in 2018 when it unveiled electrocardiogram (ECG) on the Apple Watch 4, with some critics pointing out that it wasn't much different from what AliveCor had been doing for years.
Four years later, researchers at the Mayo Clinic in Rochester, MN have not only accepted the role that the Apple Watch ECG app plays in the market, but they've developed an artificial intelligence (AI) algorithm that's programmed to interpret single-lead ECG tracings from an Apple Watch to more effectively identify patients who are living with a weak heart pump, or left ventricular systolic dysfunction.
A condition that affects 2% to 3% of people worldwide and up to 9% of people over age 60, left ventricular systolic dysfunction sometimes produces no symptoms. Earlier detection can improve outcomes, says Paul Friedman, MD, chair of the department of cardiovascular medicine at Mayo Clinic.
“On average, the mortality at five years is approximately 5% for stage B, no symptoms, and 25% for stage C, overtly symptomatic heart failure,” Friedman said. “What is important, is that once we know a weak heart pump is present there are many life-saving and symptom-preventing treatments available. It is absolutely remarkable that AI transforms a consumer watch ECG signal into a detector of this condition, which would normally require an expensive, sophisticated imaging test, such as an echocardiogram, CT scan, or MRI."
Through this technology, it’s possible that the number of patients diagnosed will increase.
“Epidemiologic studies in the United States and Europe have generally shown that approximately half of left ventricular systolic dysfunction has no symptoms, so that likely a substantial number of new cases will be uncovered,” Friedman said. “But whether the proportion of the total population will change and the extent of regional variation is not clear.”
How the researchers translated Apple Watch ECG readings into signals for the algorithm
More than 2,400 patients participated in a recent decentralized, prospective study. A standard ECG placed on the chest, arms, and legs was used to create a tracing to evaluate the heart's electrical signals. To interpret signals generated from the single lead on the watch, researchers modified an established 12-lead algorithm for low ventricular ejection fraction (i.e., the weak heart pump). To adapt the 12-lead algorithm to work with a single-lead watch signal, an adaptation technique was created to translate the single-lead readings into understandable signals for the algorithm.
Participants were required to download an app that securely transferred watch ECGs. Participants from 46 states and 11 countries securely transmitted 125,610 ECGs during a six-month period. The average app use was about two times per month and overall participation was high, as 92% of patients used the app more than once. Researchers chose the cleanest ECG readings. Study participation demonstrated the possibility for a scalable tool to be developed to screen and monitor heart patients for this condition wherever they are located, according to researchers.
"Approximately 420 patients had a watch ECG recorded within 30 days of a clinically ordered echocardiogram,” said Itzhak Zachi Attia, PhD, lead AI scientist in the department of cardiovascular medicine at Mayo Clinic. “We took advantage of those data to see whether we could identify a weak heart pump with AI analysis of the watch ECG. While our data are early, the test had an area under the curve of 0.88, meaning it is as good as or slightly better than a medical treadmill test.”
Researchers worked with the Mayo Clinic's Center for Digital Health to develop the app for the study, which securely sent all ECGs as they were recorded by patients to a data platform where they were analyzed.
"Our next steps include global prospective studies to test this prospectively in more diverse populations and demonstrate medical benefit,” Friedman said. “This is what the transformation of medicine looks like: inexpensively diagnosing serious disease from your sofa."
While the research did not compare the use of the technology by age group, Friedman said the data suggest there could be an increased sense of trust and engagement with digital technology among older generations. The mean age of participants was 53 years, with the oldest patient being 94.