New Studies: AI Captures Electrocardiogram Patterns That Could Signal a Future Sudden Cardiac Arrest

Two new studies by Cedars-Sinai investigators support using artificial intelligence (AI) to predict sudden cardiac arrest—a health emergency that in 90% of cases leads to death within minutes.

“Sudden cardiac arrest is a mostly lethal condition, and prevention will make the biggest impact, but we need to find novel clinical tools to make that possible,” said Sumeet Chugh, MD, director of the Division of Artificial Intelligence in Medicine at Cedars-Sinai and senior author of both studies. “Using AI algorithms to improve prediction of sudden cardiac arrest could help doctors identify which patients might be at higher risk of experiencing this devastating condition.”

More than 350,000 people have an out-of-hospital sudden cardiac arrest in the United States every year, according to the Centers for Disease Control and Prevention.

During sudden cardiac arrest, a change in the heart’s electrical activity causes it to suddenly stop beating. Having a heart condition can make a person more likely to experience sudden cardiac arrest, but it also can occur in people with no known heart condition.  

In a study published in Communications Medicine, David Ouyang, MD, assistant professor of Cardiology and Medicine at Cedars-Sinai, along with Chugh and fellow investigators trained a deep learning algorithm to study patterns in electrocardiograms, also known as ECGs, which are recordings of the heart’s electrical activity.

The model studied electrocardiograms from people who experienced sudden cardiac arrest and people who did not. The study included 1,827 pre-cardiac arrest electrocardiograms from 1,796 people who later experienced sudden cardiac arrest. It also included 1,342 electrocardiograms taken from 1,325 people who did not experience sudden cardiac arrest. 

The investigators found the Cedars-Sinai-developed AI model more accurately predicted who would experience out-of-hospital sudden cardiac arrest than did the more conventional method, called the ECG risk score. This is a way for doctors to calculate a person’s risk for sudden cardiac arrest that incorporates information from electrocardiogram readings.

“The entire digital electrocardiogram signal performed significantly better than a few of its components,” said Chugh, who is also the Pauline and Harold Price Chair in Cardiac Electrophysiology Research and associate director in the Smidt Heart Institute. “We plan to continue to study this AI method to learn how it could be used in a clinical setting.”

In another study, published in Circulation: Arrhythmia and Electrophysiology, Chugh along with fellow investigator Piotr Slomka, PhD, director of Innovation in Imaging at Cedars-Sinai and a research scientist in the Division of Artificial Intelligence in Medicine and the Smidt Heart Institute; and other colleagues trained an AI model to differentiate between two underlying causes of sudden cardiac arrest: pulseless electrical activity and ventricular fibrillation.

Pulseless electrical activity means that the heart’s electrical signals are too weak to produce a heartbeat. It cannot be treated with a defibrillator and often leads to death. Ventricular fibrillation is a type of irregular heartbeat that can cause the heart to stop beating, but an electric shock from a defibrillator can trigger the beating again. 

After the AI model reviewed patterns in electrocardiogram readings as well as patient characteristics, investigators were able to determine risk factors for both types of sudden cardiac arrest.

People who had pulseless electrical activity sudden cardiac arrest, for example, were more likely to have been older, been overweight, have had anemia, or experienced shortness of breath as a warning symptom. Those who had ventricular fibrillation were more likely to be younger, have had coronary artery disease or experienced chest pain as a warning symptom.

“We have ways of preventing sudden cardiac arrest through technologies like a defibrillator, but the challenge is knowing who is most likely to benefit from this intervention,” said Lauri Holmstrom, MD, PhD, a visiting postdoctoral scientist at Cedars-Sinai and first author of both studies. “These findings could help cardiologists identify which patients are likely to have a pulseless electrical activity sudden cardiac arrest or ventricular fibrillation sudden cardiac arrest, and help them prevent these events from occurring.”

The AI models used in both studies were trained, tested, and validated using data from two ongoing studies of sudden cardiac arrest founded and led by Chugh: the Oregon Sudden Unexpected Death Study and the Ventura Prediction of Sudden Death in Multi-Ethnic Communities (PRESTO) study.

“These studies exemplify the potential for AI to detect patterns in the body that the human eye and standard medical tests cannot,” said Paul Noble, MD, the Vera and Paul Guerin Family Distinguished Chair in Pulmonary Medicine and chair of the Department of Medicine at Cedars-Sinai, who was not involved in the studies. “We are getting closer to being able to use AI to prevent dangerous events such as sudden cardiac arrest.”

Other Cedars-Sinai investigators who worked on the Communications Medicine study include Harpriya Chugh; Kotoka Nakamura, PhD; Ziana Bhanji; Madison Seifer; Audrey Uy-Evanado, MD; and Kyndaron Reinier, PhD.

Other Cedars-Sinai investigators who worked on the Circulation: Arrhythmia and Electrophysiology study include Bryan Bednarski; Harpriya Chugh; Habiba Aziz; Hoang Nhat Pham, MD; Arayik Sargsyan, MD; Audrey Uy-Evanado, MD; Damini Dey, PhD; and Kyndaron Reinier, PhD.

Funding: Both studies were funded, in part, by the National Heart, Lung, and Blood Institute.

Read more on the Cedars-Sinai Blog: Heart Attack, Cardiac Arrest, Heart Failure—What's the Difference?