Prediction of Sex from a 12-Lead ECG Using Deep Learning: External Validation and Interpretation

Sex-related information might be encoded in the ECG given the inherent biological distinctions between sexes.The aim of this study was to predict sex from the ECG and to uncover the representation of sex within the morphology of the ECG waveform using saliency maps.In a secondary analysis, different ECG rhythms and age groups were compared to examine the effect of variation in the ECG morphology on the outcome.150052 patients were included with a mean age of 59.2 16.5 years of which 48% were females.The overall model resulted in an accuracy of 0.89.Saliency maps highlighted different parts of the P-wave and the ST-segment across leads as the most predictive regions.Additionally, subgroup analysis revealed decreased performance in subgroups with an altered P-wave and/or ST-segment.This study demonstrates that sex can be predicted from a 12 lead ECG with high accuracy and indicates that sex information is concealed in the P-wave and the ST-segment.