Introduction

Surface electrocardiograms (ECGs) are used frequently in routine clinical care, but also in investigational studies examining the effects of pharmacological and non-pharmacological treatments on the heart. Readout measures include the RR interval, PR interval, QRS duration and (corrected) QT interval. Typically, the pharmacological treatment effects are mediated by recognized channels on the cardiac surface. [1]However, there are cardiac effects that require a longer period of time to become visible on the surface ECG, such as aging induced cardiac fibrosis, and it is largely unknown if these subtle effects can be visualized on a surface ECG. [2, 3]
There has been a number of recent investigations regarding the prediction of physiological age using medical records, vital signs and laboratory data, or epigenetic changes. [4-6] These investigations indicated the existence of a gap between predicted physiological age and actual chronological age. Exploration of this gap is clinically important as a serious gap difference has been shown to be associated with higher risks of all-cause mortality, cardiovascular disease, obesity, earlier menopause, and frailty. [5, 7-11] Various previous studies have already shown that the 12-lead ECG can be a reliable tool to estimate physiological aging. [5, 7-16]
Previous studies have applied artificial intelligence to the raw ECG data, allowing estimation of physiologic ECG age, which was found to reflect aging and comorbidities. [17] However, these algorithms were based on large hospital datasets, thus including patients that may have disease-induced abnormalities in their ECGs, which makes the outcome difficult to interpret when applied to a healthy volunteer. Therefore, the aim of the present analysis was to develop a neural network in healthy volunteers to characterize the effect of aging on the ECG.