Analysis of Age-Related Variations in Photoplethysmography: A Machine Learning Approach

Vascular aging is an important indicator in cardiovascular risk assessment. In this study, we used a machine learning approach to estimate the cardiovascular age of subjects using the photoplethysmographic signal (PPG). From PPG, acquired in 115 healthy subjects aged 18 to 66 years, we extracted a set of morphological features and Heart Rate Variability parameters. These parameters were used in a cross-validation approach to predict the cardiovascular age of the subjects using the GradientBoostingRegressor algorithm. Quantitative performance evaluation showed promising results, yielding a mean absolute error of ( \(6.81\,\pm \,0.86\) ) and a coefficient of determination equal to ( \(0.44\,\pm \,0.22\) ). Using the SHAP method, we determined the impact of features on model performance by identifying heart rate change, low signal frequencies, and systolic phase velocity as the most significant parameters. These findings improve our understanding about the influence of age on the PPG signal, offering potential insights for future clinical applications in cardiovascular risk prevention.