Left ventricular diastolic dysfunction (LVDD) occurs at initial stage of heart failure. Electrocardiographic (ECG) criteria and machine learning for ECG features with or wihtout biological features have been applied successfully to predict LVDD in middle- and old-aged individuals. The purpose of this study is to clarify the performance of machine learning in young adults. In a large sample of 2,206 military males in Hualien, Taiwan, aged 17-43 years, the prevalence of LVDD is 4.26%. Five machine learning classifiers including random forest (RF), support vector machine (SVM), gradient boosting decision tree (GBDT), multi-layer perceptron (MLP) and logistic regression (LR) for the input of 26 ECG features with or without other 6 biological features (age, anthropometrics, and blood pressures) to link the output of LVDD are compared with the corrected QT interval (QTc) calculated by the Bazett’s formula, a traditional ECG criterion for LVDD. The definition of LVDD is based on either one of the echocardiographic criteria: (1) the E/A ratio of the mitral inflow <0.8; (2) the lateral mitral annulus velocity, e’ <10 cm/s; and (3) the E/e’ ratio >14. The area under the receiver operating characteristic (ROC) curve in machine learning of the RF, SVM, GBDT, MLP and LR for ECG only are 84.1%, 78.7%, 77.9%, 77.6% and 75.4%, which are all superior to 64.6% in the QTc interval. If the specificity is fixed around 70-80%, the sensitivity of these mahine learning classifiers for ECG only are 81.0%, 76.2%, 71.4%, 71.4% and 71.4%, which are all higher than 47.6% in the QTc interval. This study suggests that using machine learning for ECG features only to predict LVDD in Asian young adults is reliable and thereby it is possible for people to take an early preventive action on heart failure.
Diffuse myocardial fibrosis in hypertrophic cardiomyopathy can be identified by cardiovascular magnetic resonance, and is associated with left ventricular diastolic dysfunction
