Time-dependent S-wave areas by 24-hour ECG are correlated with a high risk of sudden cardiac death: ECG prediction model development and validation for SCD risk.

Abstract

Background: Sudden cardiac death (SCD) is associated with severe electrocardiogram (ECG) abnormalities. Current prediction relies heavily on static ECG parameters, limiting accuracy. This study aimed to explore dynamic ECG parameters, particularly the S-wave area and its circadian variations, as novel markers for SCD risk prediction.

Methods: All participants were divided into three different SCD risk groups based on their disease status at the time of enrollment. Dynamic single-lead ECG data was collected continuously for 24 hours and segmented into 1,440 one-minute intervals with time information tags from 0:00 to 24:00. Forty-two ECG parameters, including the S-wave area, were analyzed. Randomly selected 70% of the samples from Sun Yat-sen Memorial Hospital to construct training set and remaining samples to construct independent test set. Student's t-test was used to compare the expression differences of ECG parameters in different SCD risks patients at different time points within a day. Repeatedly attempted to establish multivariate logistics regression models combining different time points and ECG parameters and performed five-fold cross validation sequentially. Selected time point-ECG parameter combined model with the highest AUC to conduct further univariate logistic regression and calculate odds ratio (OR) of each time point-ECG parameter combination.

Results: From September 2017 to December 2020, 289 participants were enrolled: 43 at high risk of SCD (SCDHR), 138 with heart failure (HF), and 108 healthy controls (HC). Significant circadian variations in ECG parameters were observed. In the SCDHR group, key parameters significantly increased during 16:00-22:00, while the HF group showed distinct changes from 21:00-06:00. Logistic regression achieved robust performance in distinguishing groups: SCDHR vs. HC (AUC =0.887 training; AUC =0.747, accuracy =0.755, precision =0.800 test), SCDHR vs. HF (AUC =0.857 training; AUC =0.714, accuracy =0.681, precision =0.280 test) and HF vs. HC (AUC =0.965 training; AUC =0.842, accuracy =0.704, precision =0.867 test). Decision curve analysis and calibration curve showed good clinical performance of three logistics models for each comparison pair.

Conclusions: Dynamic ECG parameters, especially time-dependent variations in the S-wave area, were strongly associated with the SCD risk. They may develop into promising markers enhancing predictive accuracy for SCD stratification after further large-scale and prospective validation.