Single-lead ECG biomarkers for cardiovascular and mortality risk prediction
Paper: Lisa Attali, Yosef Solewicz, Shany Brimer Biton, Hagai Hamami, Eran Zvuloni, Ilan Green, Izhar Laufer, Pablo Laguna, Alba Martín-Yebra, Juan Pablo Martínez, Ronit Almog and Joachim A. Behar. Single-lead ECG biomarkers for cardiovascular and mortality risk prediction. Physiological Measurement, published September 18, 2026.
What Can a Single-Lead ECG Tell Us About Future Cardiovascular Risk?
We are pleased to share our latest work, “Single-lead ECG biomarkers for cardiovascular and mortality risk prediction,” published in Physiological Measurement.
Cardiovascular disease remains one of the leading causes of morbidity and mortality worldwide. Yet much of the information contained in routinely collected cardiac signals is still used primarily for diagnosing what is happening now. In this study, we asked a different question:
Can a simple single-lead ECG also tell us something about a patient's future cardiovascular risk?
Our results suggest that it can.
From rhythm monitoring to digital biomarkers
Long-term ECG recordings, such as Holter monitoring, contain much more information than individual rhythm abnormalities. They provide a continuous view of cardiac electrical activity over many hours and allow us to quantify physiological patterns that may reflect underlying cardiovascular health.
We focused on three ECG-derived digital biomarkers:
Atrial fibrillation burden (AFB): the proportion of time a patient spends in atrial fibrillation.
Premature ventricular contraction burden (PVCB): the frequency of premature ventricular beats during the recording.
T-wave alternans (TWA): subtle beat-to-beat changes in ventricular repolarization.
Each of these biomarkers has previously been associated with cardiovascular outcomes. Our hypothesis was that they capture different and complementary aspects of cardiovascular risk, and that combining them could provide more information than any one biomarker alone.
A large real-world primary-care cohort
To test this, we analyzed 81,362 Holter recordings from 54,395 individuals collected across 20 primary-care centers in Israel.
Rather than focusing on a highly selected hospital population, this cohort represents patients undergoing ambulatory cardiac monitoring in routine clinical practice.
We combined AFB, PVCB and TWA with one of the most powerful predictors of cardiovascular disease, age, and trained machine-learning models to estimate five-year risk for three clinically important outcomes: heart failure, ischemic stroke and all-cause mortality.
Complementary information hidden in the ECG
The main finding was that the three ECG biomarkers did not simply provide redundant information.
Instead, they contributed complementary prognostic signals.
On the held-out test set, the best-performing models achieved an AUROC of 0.75 for heart failure, 0.69 for ischemic stroke, and 0.79 for all-cause mortality.
Importantly, combining the ECG-derived biomarkers improved discrimination compared with using age alone, with improvements of up to 10% among individuals younger than 75 years.
This age-dependent result is particularly interesting. Age itself is such a dominant cardiovascular risk factor that it can be difficult for additional biomarkers to add substantial predictive information in older populations. In younger patients, however, physiological signals extracted from the ECG may help identify individuals whose cardiovascular risk is higher than their chronological age alone would suggest.
Why single-lead ECG matters
Another important aspect of this work is that the biomarkers were derived from a single ECG lead.
This is increasingly relevant because single-lead ECG signals can be acquired using relatively simple ambulatory and wearable devices. The potential value of such signals therefore extends beyond traditional Holter analysis.
Instead of viewing a wearable ECG only as a tool for detecting an arrhythmia when it occurs, we can begin to think of longitudinal ECG monitoring as a source of digital cardiovascular biomarkers.
A continuously or repeatedly recorded physiological signal could potentially contribute to risk assessment long before a major cardiovascular event occurs.
Toward interpretable ECG-based risk prediction
Much recent work in artificial intelligence has focused on training deep neural networks to predict clinical outcomes directly from raw ECG waveforms.
Our approach is somewhat different.
Rather than relying entirely on a black-box representation, we deliberately used a small number of physiologically interpretable ECG biomarkers whose relationship with cardiac electrophysiology is already understood.
This makes it possible not only to generate a risk estimate, but also to investigate which aspects of the cardiac signal are contributing to that risk.
The results suggest that relatively simple, interpretable features extracted from long-term ECG monitoring still contain substantial prognostic information, particularly when several complementary biomarkers are considered together.
What comes next?
This study is an important step, but not the final one.
The models were developed using a large retrospective real-world cohort, and further work will be needed to determine how these biomarkers generalize across healthcare systems, recording devices and patient populations, and ultimately how they might be incorporated into clinical decision-making.
There is also considerable room to expand the information extracted from ambulatory ECGs. Modern machine-learning and foundation-model approaches can characterize cardiac signals at much greater depth than conventional ECG measurements alone.
Our broader goal is therefore to understand how the enormous amount of physiological information collected by ECG monitoring can be transformed into meaningful, clinically useful measures of future health.
A single-lead ECG may look like a simple signal.
But recorded over time, it may contain a surprisingly rich picture of cardiovascular risk.
Comments