Participants with AI-derived vascular age gap >9 years had 2.37-fold higher MACCE risk; each 1-year increase raised in-hospital mortality odds by 2% in external validation.
Does AI-derived vascular age from PPG signals predict the risk of MACCE and secondary cardiovascular outcomes?
AI-derived vascular age from PPG signals is a scalable digital biomarker that significantly predicts the risk of MACCE and other cardiovascular outcomes.
Abstract Introduction With the increasing availability of wearable devices, photoplethysmography (PPG) has emerged as a promising non-invasive tool for monitoring human hemodynamics. Vascular aging is a critical factor in cardiovascular health, yet existing assessment methods remain limited in accessibility and scalability. Deep learning offers new opportunities to enhance vascular age estimation from PPG signals, potentially improving risk stratification and early intervention. Purpose This study aims to develop a deep learning-based framework to estimate vascular age (AI-vascular age) from PPG signals and assess its potential as a digital biomarker for cardiovascular health. Additionally, we introduce a distribution-aware loss function to address biases associated with imbalanced data distributions. Methods The model was developed and validated using data from the UK Biobank (UKB) cohort. The development cohort comprised 98,672 data pairs from 98,672 participants without a history of circulatory disorders, diabetes, or dyslipidemia, while 144,683 data pairs from 113,559 participants were reserved for clinical evaluation. We assessed the association between the vascular age gap (AI-vascular age minus calendar age) and the risk of major adverse cardiovascular and cerebrovascular events (MACCE) and secondary outcomes. Longitudinal applicability was examined using serial PPG data, and external validation was conducted on a MIMIC-III-derived cohort (n = 2,343). Results After adjusting for confounding factors, participants with a vascular age gap exceeding 9 years exhibited a significantly higher risk of MACCE (HR = 2.37, p 0.005) and secondary outcomes, including diabetes (HR = 2.69, p 0.005), hypertension (HR = 2.88, p 0.005), coronary heart disease (HR = 2.20, p 0.005), heart failure (HR = 2.15, p 0.005), myocardial infarction (HR = 2.51, p 0.005), stroke (HR = 2.55, p 0.005), and all-cause mortality (HR = 2.51, p 0.005). Conversely, participants with a vascular age gap below -9 years had a significantly lower incidence of these outcomes. Longitudinal analysis demonstrated the value of AI-vascular age in risk stratification, as measurements at two distinct time points improved MACCE prediction. External validation in the MIMIC-III cohort confirmed that each one-year increase in vascular age gap was significantly associated with elevated in-hospital mortality risk (OR = 1.02, p 0.005). Conclusion AI-vascular age represents a novel, non-invasive digital biomarker for cardiovascular health assessment. By transforming PPG signals into actionable insights, this approach holds significant potential for scalable risk stratification, personalized health monitoring, and early intervention in both general and high-risk populations.Development and application of the model Cumulative incidence of the outcomes
Zhao et al. (2025) studied this question. Participants with AI-derived vascular age gap >9 years had 2.37-fold higher MACCE risk; each 1-year increase raised in-hospital mortality odds by 2% in external validation.