Wearable AI-ECG Age and Its Link to Atrial Fibrillation
Original Title: Wearable device derived electrocardiographic age and its association with atrial fibrillation
Journal: NPJ digital medicine
DOI: 10.1038/s41746-026-02344-8
Overview
This research introduces the PROPHECG-Age Single model, a deep-learning framework designed to estimate electrocardiographic age from single-lead wearable recordings. This approach enables continuous monitoring of cardiovascular aging in real-world settings. The study utilized a dataset of one million 12-lead electrocardiograms, which were converted into synthetic single-lead signals using a Cycle-Consistent Generative Adversarial Network. Validation in two independent wearable cohorts, S-Patch and Memo Patch, demonstrated mean absolute errors of 10.01 and 11.88 years, respectively. Statistical analysis revealed that the gap between predicted and chronological age is significantly associated with the presence of atrial fibrillation, showing a pooled adjusted odds ratio of 1.03 for every one-year increase in the age gap. Furthermore, the model established a positive correlation with atrial fibrillation burden, where each additional year in the age gap corresponded to a 0.8 percentage point increase in the time spent in an arrhythmic state.
Novelty
A primary challenge in developing wearable artificial intelligence models is the scarcity of large-scale single-lead datasets. This study addresses this by adopting a domain adaptation strategy that leverages existing hospital records. Instead of reconstructing 12-lead information from single-lead inputs, the researchers transformed 12-lead data into a single-lead format to train the model. This preserves the information content of established archives while ensuring compatibility with wearable hardware. The study also identifies a structural error floor associated with single-lead inputs, proving that the higher error rate compared to 12-lead models is a consequence of reduced spatial information rather than environmental noise. The research demonstrates high temporal consistency, with an intraclass correlation coefficient of 0.93 across multiple days, confirming that the derived age gap functions as a stable, personalized biomarker.
Potential Clinical / Research Applications
The ability to derive aging biomarkers from wearable devices offers several applications in clinical and research settings. In preventive cardiology, this model could facilitate risk management by identifying individuals with an accelerated cardiac aging profile before they develop symptomatic atrial fibrillation. Because the concept of an electrocardiographic age is intuitive, it may improve patient engagement by providing an understandable metric of heart health. In clinical trials, the model could serve as a digital endpoint to evaluate the impact of therapies on the rate of electrophysiological aging. Furthermore, the high reproducibility of the age gap makes it suitable for long-term population health studies aimed at understanding how environmental factors influence cardiovascular decline. By making the model weights publicly accessible, the researchers have provided a foundation for integrating personalized aging metrics into remote monitoring platforms, supporting continuous cardiovascular care.