Cardiovascular Risk Assessment Based on Vascular Age Gap Using Stacking Ensemble Learning
Year: 2026
Type: conference-paper
Abstract: Cardiovascular diseases (CVDs) remain the leading cause of global mortality, necessitating advanced, non-invasive early-screening modalities. This study introduces a sophisticated Stacking Ensemble framework designed to quantify Vascular Age (VA) and the Vascular Age Gap (VAGap) through the analysis of peripheral pulse morphology. By architecting a multi-level paradigm that integrates XGBoost, Random Forest, and Support Vector Regressor (SVR), we achieve a synergistic predictive optimization that significantly outperforms conventional monolithic models. Utilizing a diverse cohort of 226 subjects (1,364 pulse cycles), our methodology employs a rigorous preprocessing pipeline, including Savitzky-Golay denoising and the extraction of 15 multidimensional features from the second (APG) and fourth derivatives of the photoplethysmogram (PPG) signal, alongside clinical biomarkers such as BMI and blood pressure. Critically, to ensure the integrity of biological aging inference and mitigate direct identity leakage, chronological age was strictly excluded from the feature space. Empirical results demonstrate robust and reliable performance, with the coefficient of determination $\left(R^{2}\right)$ reaching 0.93 for males and 0.76 for females under a feature-blinded, pulse-cycle level validation scheme designed to maximize morphological variance extraction. Feature importance analysis identifies Pulse Pressure (PP) and Mean Arterial Pressure (MAP) as the predominant hemodynamic determinants of arterial stiffness. This research underscores the transformative potential of Stacking Ensemble architectures as high-fidelity, cost-effective primary screening tools in the burgeoning landscape of digital cardiology.
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204120159