AI May Detect Hidden Heart Risk in Routine ECGs, Study Finds
New research suggests artificial intelligence can identify patterns linked to sudden cardiac death that are missed by current screening methods.
Artificial intelligence may soon help doctors identify a hidden risk of sudden cardiac death by analyzing electrocardiograms (ECGs), according to new research from the University of California, Berkeley. The study, published in the journal Nature, trained an AI model to detect subtle waveform patterns in ECGs that are associated with sudden cardiac arrest, a condition that can strike individuals without prior warning or known heart problems.
Sudden cardiac arrest claims hundreds of thousands of lives annually in the United States, with survival rates dropping rapidly once it occurs outside a hospital setting. While CPR and defibrillators can be life-saving, timely intervention is critical. This AI tool could potentially flag at-risk patients earlier, even when their hearts appear normal on conventional tests.
How AI Identified Hidden Heart Risk
ECGs record the heart's electrical activity, displaying it as a series of spikes and waves that clinicians review. For this study, researchers utilized over 440,000 ECGs from Sweden, correlating them with death certificates and health records. The AI model was trained on this data to identify patterns linked to sudden cardiac death.
Crucially, the model's performance was then evaluated on separate patient data from the U.S. and Taiwan, demonstrating its robustness across different healthcare systems. This validation is a critical step, as AI models often perform well on initial datasets but falter in real-world applications.
Limitations of Current Heart Screening
Physicians often rely on measurements such as left ventricular ejection fraction (LVEF) to assess cardiac risk. A low LVEF can indicate that a patient may benefit from an implantable defibrillator, a device designed to restore a normal heart rhythm during a dangerous event. However, this method has significant gaps.
Many individuals who experience sudden cardiac death do not undergo the deeper cardiac evaluations that might reveal a low LVEF. Furthermore, some patients with a normally functioning heart may still be susceptible to dangerous rhythm disturbances. The UC Berkeley model identified a high-risk group with an annual sudden cardiac death rate of 7%, surpassing the 4.6% rate found in the standard reduced LVEF group. Notably, the majority of patients flagged by the AI were not identified by the LVEF method, suggesting that routine ECGs may contain overlooked warning signs.
Understanding the AI's Findings
To prevent the AI from becoming a "black box," researchers also employed another AI system to analyze the differences between low-risk and high-risk ECG patterns. This analysis pointed to a previously undescribed feature within the QRS complex—the main electrical signal during a heartbeat—in a specific ECG view known as aVL. This signal appears to be a strong predictor of sudden cardiac death and had not been previously documented in medical literature.
Implications for Defibrillator Decisions and Future Care
The potential for this AI tool to refine decisions about implantable defibrillators is significant. While these devices can save lives, their implantation carries procedural risks and costs, and many devices are never activated. By flagging patients who warrant closer monitoring, the AI could help prevent unnecessary procedures while ensuring those at high risk receive appropriate attention.
Researchers are currently conducting further testing with health systems in Sweden, Taiwan, and the U.S. If the AI tool consistently flags high-risk scans, patients could be contacted for closer monitoring, potentially with heart-monitoring patches, to detect dangerous rhythms before they become fatal.
Data Privacy Considerations
The development of medical AI relies on vast datasets, raising important questions about data privacy. Ensuring clear guidelines for how patient data is controlled, protected, and utilized by hospitals, researchers, and AI companies is paramount for public trust. Patients should be informed about how their health records are used, especially as AI tools become more integrated into healthcare.
What This Means for the Public
While this AI tool is not yet available for public use, its potential is substantial. It suggests that a routine test many individuals have already undergone could one day reveal hidden risks. In the meantime, individuals should not ignore symptoms like fainting, dizziness, or a racing heartbeat, and should discuss family history of sudden cardiac death with their doctor. While wearable devices and blood pressure monitors can offer some health insights, they do not replace professional medical evaluation. Learning CPR and being aware of Automated External Defibrillator (AED) locations remain vital emergency preparedness measures.