A new deep learning model has uncovered a hidden electrocardiogram (ECG) biomarker that strongly predicts sudden cardiac death, according to two studies published in Nature. The biomarker, invisible to standard analysis, was identified by training the algorithm on thousands of routine ECG recordings. Researchers say the finding could transform how doctors assess risk for a condition that claims millions of lives annually.
The discovery addresses a critical gap in cardiac care: sudden cardiac death often strikes people with no prior symptoms or known risk factors. Current screening tools, such as ejection fraction measurements, miss many at-risk individuals. The deep learning approach offers a low-cost, noninvasive way to flag high-risk patients from a standard ECG, a test already widely used in clinics.
In the studies, the algorithm analyzed ECG data from large patient cohorts, correctly identifying those who later experienced sudden cardiac death with notable accuracy. The precise sensitivity and specificity figures were not detailed in the available sources. The model focused on subtle waveform patterns that escape human interpretation, leveraging millions of data points to learn the signature.
If validated in prospective trials, this biomarker could be integrated into routine ECG interpretation software, alerting clinicians in real time. Patients flagged as high-risk might then be referred for further evaluation or preventive therapies, such as implantable defibrillators. However, widespread adoption would require regulatory clearance and integration into existing medical systems.
Experts caution that the model's performance in diverse populations remains untested, and overdiagnosis could lead to unnecessary procedures. Prospective studies are needed to confirm clinical utility before the tool reaches practice.