AI Discovers New Method to Detect Sudden Cardiac Death Risk

Researchers from the University of California, Berkeley developed an AI system that identifies a new electrocardiogram (ECG) biomarker to predict sudden cardiac death (SCD) risk. According to a study published in Nature, this deep learning tool detects high-risk patients who would be missed by current clinical standards, potentially identifying thousands of additional candidates for preventative treatment annually.

The system analyzes standard ECG readings to find patterns that are invisible to human cardiologists. This discovery addresses a long-standing gap in cardiac care where patients suffer fatal arrhythmias despite having ECGs that appear normal under traditional medical review, according to reporting from Medical Xpress.

How does the AI detect sudden cardiac death risk?

The AI utilizes deep learning to scan thousands of ECG waveforms, identifying subtle electrical signatures that correlate with a high risk of SCD. While human doctors look for specific, known markers like QT interval prolongation or specific ST-segment changes, the AI identifies a complex biomarker that does not have a previous name or established visual pattern in medical textbooks, according to the University of California, Berkeley.

This approach allows the system to categorize patients based on their actual risk of an event rather than relying on the presence of a known disease. Stat News reports that this method helps solve a medical mystery by finding “hidden” risks in patients who otherwise meet all current criteria for being low-risk.

Why is this discovery significant for patient care?

The primary clinical value lies in the identification of patients who qualify for implantable cardioverter-defibrillators (ICDs). These devices can shock a heart back into a normal rhythm during a lethal arrhythmia, but they are typically only implanted in patients with a proven high risk of SCD. Current screening methods miss a significant portion of the at-risk population, according to Nature.

Medical Xpress notes that the implementation of this AI system could identify thousands more patients annually who are eligible for these life-saving devices. By expanding the pool of identified high-risk patients, clinicians can intervene before a cardiac event occurs rather than reacting to one.

How does this compare to traditional ECG screening?

Traditional ECG screening relies on human interpretation of specific, visible abnormalities. This manual process is limited by the known patterns that physicians are trained to recognize. In contrast, the UC Berkeley AI system identifies non-linear patterns across the entire ECG signal.

The difference in outcomes is stark. While traditional methods might clear a patient as “low risk” based on the absence of known markers, the AI can flag that same patient as “high risk” based on the new biomarker. Stat News highlights that this contrast reveals how much risk currently goes undetected in standard clinical practice.

What are the limitations and next steps?

Despite the potential, the researchers emphasize that the AI is a screening tool and not a standalone diagnostic. The system identifies risk, but clinical decisions regarding ICD implantation still require a comprehensive medical evaluation by a cardiologist, according to the University of California, Berkeley.

Further validation is required to determine the specificity of the biomarker across diverse patient populations. The researchers aim to refine the algorithm to ensure that it does not lead to an over-prescription of ICDs in patients who may not actually need them, which would introduce unnecessary surgical risks.

The findings were formally detailed in the research published on June 24, 2026, marking a shift toward using machine learning to discover new biological markers rather than simply automating existing human tasks.