The team found that a person’s AI-PPG age (an estimate of biological vascular age) often differs from their calendar age. This difference, or “age gap,” turns out to be a powerful predictor of cardiovascular events and mortality, offering a scalable, non-invasive method for population-level health monitoring.
Background
Cardiovascular disease remains the leading cause of death worldwide, yet early detection is often constrained by the need for clinic-based, specialized equipment.
PPG signals, already collected by many wearables, offer a non-invasive and widely accessible alternative. While previous research has used PPG data to detect specific conditions like atrial fibrillation, a more comprehensive indicator of vascular health had yet to be developed.
That’s where AI-PPG age comes in. The researchers built a deep learning model to estimate biological age from raw PPG signals. The gap between this AI-derived age and a person’s calendar age emerges as a novel biomarker that can signal elevated cardiovascular risk and guide proactive health decisions.
Model Development and Validation
The model was primarily trained on data from the large-scale UK Biobank (UKB) and then fine-tuned using the VitalDB subset of the independent PulseDB dataset. External validation was conducted using the MIMIC-III subset of PulseDB, which includes links to clinical outcomes like in-hospital mortality.
A key technical challenge was the age imbalance in the UKB dataset, which skewed toward middle-aged individuals. To address this, researchers introduced Dist Loss, a distribution-aware loss function. This combined traditional mean absolute error (MAE) with a component that aligned the model’s predictions with the actual age distribution, improving both individual accuracy and overall population-level reliability.
The model used a one-dimensional convolutional neural network (Net1D) to extract meaningful features from PPG waveforms. Interpretability was enhanced through saliency maps, which revealed that the model primarily relied on the diastolic peak - aligned with established physiological markers of vascular aging. The primary metric was the AI-PPG age gap, which represents the difference between the predicted biological age and the actual age, and was statistically tested for associations with cardiovascular outcomes and mortality.
Predictive Power of the AI-PPG Age Gap
The AI-PPG age gap showed strong predictive value across multiple health outcomes. In the UKB cohort, each one-year increase in the age gap was linked to a higher risk of major adverse cardiovascular and cerebrovascular events (MACCE), as well as related conditions such as diabetes and hypertension.
Participants whose AI-PPG age exceeded their calendar age by more than nine years faced substantially higher risk across all outcomes. Conversely, those with a gap of more than nine years in the opposite direction had notably lower risk.
Longitudinal analysis also found that participants whose age gap classification worsened over time saw an increased likelihood of future MACCE. External validation in a critical care setting confirmed that higher AI-PPG age was associated with increased in-hospital mortality.
While the model's MAE ranged from 7.6 to 11.8 years depending on the dataset (reflecting some biological and signal variability), its predictive utility remained strong.
Moving the Needle in Preventive Cardiology
The study demonstrates that AI-PPG age is a true and scalable indicator of cardiovascular health. It independently predicts major cardiac events and mortality in both general and critically ill populations. Serial measurements added another layer of value, showing that changes in the age gap over time could refine risk assessment.
Technically, the study tackled common modeling challenges, using a distribution-aware loss function to manage skewed training data and saliency mapping to confirm the model’s focus on physiologically relevant waveform components. Perhaps most importantly, the model’s compatibility with consumer wearables opens the door to large-scale screening and real-time, personalized health monitoring.
That said, the researchers did note certain limitations, including a moderate correlation with calendar age and reliance on data-driven risk thresholds that require further validation. The team also emphasized that AI-PPG age is a supplementary tool, not a substitute for clinical diagnosis, but one with strong potential to enhance early detection and prevention strategies.
Conclusion
This study introduces AI-PPG age as a novel, non-invasive biomarker for assessing vascular health using wearable tech. The gap between AI-estimated biological age and actual age emerged as a reliable indicator of cardiovascular risk and mortality, offering a practical solution for proactive health monitoring.
By harnessing everyday PPG signals, this technology could help shift the focus of cardiovascular care toward early, individualized intervention - and ultimately reduce the burden of heart disease on a global scale.
Journal Reference
Nie et al. (2025). Artificial intelligence-derived photoplethysmography age as a digital biomarker for cardiovascular health. Communications Medicine, 5(1), 481–481. DOI:10.1038/s43856-025-01188-9. https://www.nature.com/articles/s43856-025-01188-9
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.