Improved Prediction of Atrial Fibrillation by Identifying ECG Patient Trajectories over Time

Atrial fibrillation (AF) risk can be predicted from a single-point ECG one month before onset.This study investigates whether this AF risk score can be improved by combining the risk score of multiple consecutive ECGs over time.We compare the predictive performance of a single-point risk score with the average and maximum ECG risk scores within one year prior to AF onset as well as trajectories of one-year interval averaged and maximized scores up to five years before onset.K-means clustering was used to visualize trajectory patterns among patients.We included 13036 patients with an AF prevalence of 22.6%.The results showed a slight increase in performance for the one-year interval maximized approach, but no further improvement was observed using the trajectory approach.Trajectory clustering revealed substantial variability in AF onset patterns, suggesting diversification in the development of AF across patients.