P2437Screening for cardiac relaxation abnormalities using surface ECG wavelets for identifying high-risk cardiac phenotypic abnormalities

Abstract

Abstract Background The impairment of myocardial relaxation is a strong predictor of all-cause mortality and has been proposed to be a potential tool for cardiovascular (CV) risk stratification. We investigated a novel signal-processed electrocardiography (spECG) technique to extract the features of abnormal myocardial relaxation as a screening tool to identify high-risk CV patients. Methods Time–frequency-energy features extracted from continuous wavelet-transformed spECG (Fig. A) were obtained in 1,006 cases. A machine learning model was trained for predicting abnormal myocardial relaxation as a screening tool for detecting high-risk CV patients. High-risk CV phenotype was defined as presence of LV hypertrophy, advanced LV diastolic dysfunction (grade 2 or 3), LV ejection fraction <50%, and/or significant valvular heart disease. Results After training with 5-fold cross validation using data from 810 patients, the machine learning model when tested in an independent hold-out validation set of 180 cases, showed an area under receiver-operating characteristic curve (AUC) of 0.83 (p<0.001) for prediction of myocardial relaxation impairment (Fig. B). A prediction of abnormal relaxation was associated with older patients (64±11 vs. 45±16 years old, p<0.001) with a higher prevalence of coronary artery disease (23% vs. 7%, p=0.004), hypertension (70% vs. 40%, p<0.001), and diabetes (30% vs. 9%, p=0.001). Furthermore, a prediction of abnormal myocardial relaxation was associated with increased likelihood of high-risk CV phenotypes (Odds ratio: 3.93, p<0.001) including LV hypertrophy (Odds ratio: 2.62, p=0.028), advanced LV diastolic dysfunction (Odds ratio: 11.4, p=0.020), and LV ejection fraction <50% (Odds ratio 10.5, p=0.025). Age and gender modified the prediction of abnormal relaxation with higher diagnostic value seen for patients under 60 years (Fig C, AUC: 0.88, p<0.001) and in male patients (Fig D, AUC: 0.87, p<0.001). The algorithm for abnormal relaxation also showed robust prediction of LV ejection fraction <50% (Fig E, AUC: 0.91, p<0.001) in male patients. spECG showed significant net reclassification improvement (0.47, p<0.001) and integrated discrimination improvement (0.16, p<0.001) over traditional surface ECG interpretation using Glasgow score for predicting abnormal relaxation and other high-risk phenotypic presentations. ROC curves Conclusion The spECG provided a robust prediction of abnormal myocardial relaxation and may be a valuable screening strategy for early detection of high-risk cardiac structural and functional abnormalities.