Latest News

Mayo May Soon Bring AI-Enabled Electrocardiograms To The Clinic

By Deborah Borfitz

June 21, 2021 | Researchers at the Mayo Clinic have demonstrated that a standard, 12-lead electrocardiogram (ECG) can be a biomarker of biological age, and the gap relative to chronological age is an independent risk predictor of both all-cause and cardiovascular mortality. Artificial intelligence (AI) was used to predict biological age from ECGs, one of a growing number of algorithms being developed by Mayo’s AI in Cardiology Work Group to detect heart conditions such as low ejection fraction and aortic stenosis, according to Francisco Lopez-Jimenez, M.D., co-director of the group and chair of the division of preventive cardiology.

Medicine needs neither another assumption-based risk prediction formula nor a complex genetic test directly measuring biological age, he says. The utility of the novel age-gap tool is that it could be easily deployed in clinical practice where ECG testing is commonplace, as well as reveal who is at advanced biological age even in the absence of factors like smoking and high cholesterol known to cause heart disease and premature death.

As envisioned by Lopez-Jimenez, AI-enabled ECGs could help motivate patients to make sustained, positive lifestyle changes and remind healthcare providers to screen for pre-symptomatic disease. He also thinks similar tools could be developed to pick up concerning, age-related changes that go beyond the heart, possibly including dementia and frailty.

The biology that can be captured on ECGs is potentially quite broad, he says. His Mayo colleagues, for example, recently found that AI could analyze ECGs to identify people with liver cirrhosis with a high level of accuracy. “Many things that are not related to the heart affect the heart one way or the other.”

Results of the age-gap study, recently published in the European Heart Journal – Digital Health (DOI: 10.1093/ehjdh/ztab043), validated and expanded on prior observations that the model could detect accelerated aging by showing that those whose predicted age was older than their actual age died sooner, particularly from heart disease. Adjusting for multiple standard risk factors only made the association between the age gap and cardiovascular mortality more pronounced.

More than 25,000 subjects, 95% Caucasian, were tested and followed for more than 12 years, says Lopez-Jimenez. Subgroup analysis on the other 5% (over 1,200 individuals) suggests the tool would work well across racial and ethnic groups but this needs to be confirmed with additional studies.

Previous algorithms developed by the AI in Cardiology Work Group to estimate the gap between actual and biological age proved to be broadly applicable across large patient cohorts from Russia and Scandinavia, he notes. The predictive tool for low ejection fraction (a measure of the heart’s pumping ability), also created using ECGs on a disproportionately Caucasian population, was subsequently found to have identical accuracy in Asians, Blacks, and Hispanics.   

Benefits And Risks

Use of the age-gap tool by Mayo Clinic practitioners is not far off, Lopez-Jimenez says. Only a couple more validation studies still need to be done to attest to its value in different types of people and settings.

What is uncertain is the ideal time to implement a new algorithm. The U.S. Food and Drug Administration applies the same rigor to the approval process for AI software as it does for traditional diagnostic tests, although Lopez-Jimenez says it is doubtful the agency will apply its software as a medical device rules to an algorithm doing age and risk prediction. The Framingham Risk Score and American College of Cardiology/American Heart Association Pooled Cohort Equations for estimating risk of cardiovascular events are both widely used and publicly available on the web.

Risks associated with using the new age-gap estimator are presumably small and, for patients, might include the inconvenience of getting an ECG or being unhappy at hearing they are biologically older than their years on earth, he says. On the other hand, the information might well be actionable.

As is soon to be published in Wellcome Open Research (DOI: 10.12688/wellcomeopenres.16499.1), delta age is closely related to established risk factors for cardiovascular disease (blood pressure, body mass index, total cholesterol, and smoking). If the age-gap measure started appearing on ECG reports, patients might be more inclined to do what doctors have been preaching for years—adopt a healthy lifestyle, undertake appropriate screening, and take anti-aging measures.

In most cases, people are biologically older than their chronological age as a function of stress and lifestyle issues, Lopez-Jimenez says. He is confident that new information provided by AI-enabled ECGs would facilitate physician-patient dialogue about the benefits of a more intense lifestyle modification or weight loss program.

Licensing Strategy

The most likely scenario for broader rollout of the predictive algorithm is by licensing the technology to hospitals or companies like Apple that produce smartwatches, says Lopez-Jimenez. It is not clear how a traditional payer would even be billed for “repurposing” an ECG with AI generating new outputs.

If Apple wants to validate the new algorithm with its ECG app, and the model is found to be as predictive as when a full ECG is used, the company could theoretically decide to buy the license and start offering age-gap readouts as a service to Apple Watch users, he says. Any such deal-making will be happening through Anumana Inc., which the Mayo Clinic recently launched with health technology company nference to develop and commercialize its AI-enabled algorithms.

The focus initially is on designing neural network algorithms based on billions of relevant pieces of heart health data in Mayo Clinic's Clinical Data Analytics Platform, including ECG signals, to enable early detection and faster treatment of heart disease. Since its creation tree years ago, the AI in Cardiology Work Group has developed a handful of algorithms that mine ECGs for biomedical knowledge. The AI tool for predicting low ejection fraction has been extensively validated in various populations and devices, including in combination with the ECG-enabled stethoscope of digital health company Eko.

Among other Mayo projects undertaken by the team are an algorithm specific to detecting amyloidosis, a rare but serious heart condition, and Long QT syndrome, a heart rhythm disorder causing fast, chaotic heartbeats, Lopez-Jimenez says. An algorithm for diagnosing atrial fibrillation, which can lead to a fatal stroke, can read the warning signs in an ECG even during periods when the heart is beating normally.