Estimation of calendar age based on autonomic cardiovascular function by applying machine learning techniques

Abstract

Aging is accompanied by changes in the cardiovascular physiology that promote the development of age-related diseases. This paper presents a modern approach to quantify the physiological effects of age on the cardiovascular system by applying modern machine learning techniques to several indicators of autonomic cardiovascular function. In 885 healthy subjects, 33 different indices were calculated on resting state electrocardiogram and continuous blood pressure recordings. Based on those parameters, five different approaches were applied in order to reconstruct the calendar age of healthy subjects, i.e., linear regression (LR), neural network (NN), Gaussian process regression (GPR), support vector regression (SVR), and relevance vector regression (RVR). Hyper parameters of machine learning methods were optimized via grid search. After 20 repetitions of a five-fold cross-validation, the mean absolute error (MAE) was computed between the calendar and estimated age to assess the accuracy of each method. The results show that the lowest error for age estimation was achieved using SVR with a MAE of 5.49 years. GPR performed comparably well to SVR with a MAE of 5.55 years, while NN led to a MAE of 5.72 years. RVR and LR revealed MAE of more than six years (6.21 and 6.34 years). The error of age estimation could be further reduced by applying outlier correction to the input data leading to a minimum MAE of 4.53 years with GPR. In conclusion, our results suggest that machine learning can be used to quantify the effects of healthy aging on cardio-vascular function.

Keywords

• linear regression
• neural network
• gaussian process regression
• support vector regression
• relevance vector regression