The discrepancy between machine-estimated age from transthoracic echocardiography and chronological age as a key predictor of increased risk for new-onset atrial fibrillation

Background Age estimation utilizing artificial intelligence can be derived from a variety of sources. Objectives This study aimed to assess the efficacy of convolutional neural networks (CNNs) in estimating age through standard transthoracic echocardiography (TTE) and to evaluate its predictive capability regarding the onset of atrial fibrillation (AF) while in sinus rhythm. Methods The algorithm underwent training with a cohort of 76,342 patients, was validated with an additional 22,825 patients, and tested on 20,960 patients. A multivariate Cox regression model was utilized to examine the association between machine-estimated age and chronological age in relation to the diagnosis of new-onset atrial fibrillation. Patients with prior diagnosis of atrial fibrillation were excluded. Results The model demonstrated a mean average error of 4.9 years for age estimation, root-mean-square-error (RMSE) of 6.33 and achieved a Pearson correlation coefficient of 0.922 (Figure 1). Furthermore, multivariate analysis indicated that a machine-based age prediction exceeding five years of chronological age correlates with an independent 18% increased risk of developing new-onset atrial fibrillation during follow-up (p<0.001) (Figure 2). Consistent results were obtained in a multivariate analysis after adjustment for age, sex, mortality predictors, left ventricular ejection fraction (LVEF), left atrial dimensions and echocardiographic diastolic parameters [HR=1.17, 95% CI 1.02-1.34, p=0.02]. Conclusions The integration of artificial intelligence into standard TTE facilitates the estimation of biological age. Machine-based age estimation serves as an independent predictor of the emergence of new-onset atrial fibrillation. Such predictive models can be instrumental for risk stratification and suggesting its potential utility to identify individuals for AF prevention.Model description and age estimation Figure 2.Cumulative AFib incidence