¶ The study published in Nature was led by Ziad Obermeyer, associate professor at the University of California, Berkeley. The group flagged by the model accounted for 2.2 percent of the cases examined and had an annual sudden cardiac death rate of 7.0 percent, while people with reduced heart pumping function (LVEF), who made up 1.9 percent of the sample, had an annual rate of 4.6 percent. The current standard measure, left ventricular ejection fraction, or LVEF, did not identify 86.1 percent of these high-risk patients as such.
How the researchers found the signal
The model was based on data from the team led by study leader Ziad Obermeyer from the Swedish region of Halland: 441,614 ECGs from 187,677 patients from 2010 to 2016, linked to death certificates. The model was then validated on 251,858 ECGs from 151,015 patients in a US health system and on 4,107 ECGs from a Taiwanese registry, including 96 cases of subsequent cardiac arrest and 4,011 controls. In the US cohort, the model predicted ventricular arrhythmias that cause sudden death, while in the Taiwanese registry it predicted future arrhythmic cardiac arrests. No patients were specifically recruited for the retrospective analysis, and a sample size was not calculated in advance because the researchers wanted to use as much training data as possible.

H2 A signal in the ECG curve
To show what the prediction model responds to, the authors combined it with a generative model of the ECG curve. They found an easily recognizable biomarker that robustly predicts sudden cardiac death and, to the authors' knowledge, had not previously been described. They linked the shape of this signal to basic electrophysiological principles and conducted a preliminary test of a new hypothesis about the mechanism of sudden cardiac death. According to the study, mortality from sudden cardiac death among patients with an implanted defibrillator whose ECG had been classified as high risk was 54.4 percent lower than expected, which the authors consider a possible indication of a benefit.

What the study does not yet show
Sumeet S. Chugh, who was not involved in the study and leads the Center for Cardiac Arrest Prevention at Cedars-Sinai Medical Center, says that from a patient care perspective, considerably more research is needed before such findings determine the selection of defibrillator candidates for primary prevention. Study leader Obermeyer said he would not recommend having a defibrillator implanted solely because of a high-risk ECG classification; the AI could instead be used to target further examinations. During the peer review process, a reviewer reported that an earlier version of the manuscript identified a high-risk group of 1.8 percent with an annual risk of 6.1 percent, and the authors acknowledged that differences in cohort designs and outcome definitions in the external datasets make the metrics difficult to interpret.
What is planned for use in hospitals
According to ScienceDaily, Obermeyer is already working with health systems in Sweden, Taiwan and the US to introduce the algorithm into hospital ECG databases; patients classified as high risk are to be notified and offered a patch to continuously record the heart's electrical activity. One reason for the effort: About two-thirds of defibrillators implanted in patients identified as high risk never need to deliver a shock, while the widely used LVEF measurement also misses most sudden cardiac deaths. An advantage, according to the report, is that the system works with ordinary ECG signal recordings that are already routinely collected in hospitals; this could improve early detection without requiring an entirely new type of medical examination. The main Swedish dataset is not publicly accessible for privacy reasons, nor is the external US dataset under its data use agreement; the Taiwanese validation dataset, however, is publicly available, and the team also provides code, model architecture and training procedures through a repository. According to Obermeyer, compiling the data took around 10 years and also involved two organizations he co-founded, Dandelion Health and Nightingale Open Science.
