A Risk Probability Prediction Model for Sudden Cardiac Death Based on Heart Rate Variability Metrics

The estimated annual number of sudden cardiac death (SCD) is approximately 4 million cases worldwide and approximately 50% of SCDs are unexpected first manifestations of cardiac disease. The identification of subjects at high risk for SCD is of great importance as the prevention of SCD events would be possible with the implantable cardioverter defibrillator (ICD). However, there is no reliable method to estimate individualized SCD risk for prevention. In this paper, we introduced a novel approach to predict individualized SCD risk probability based on heart rate variability metrics (HRV). First, we selected 11 commonly used HRV metrics. The heart rate (HR) corrected HRV metrics (HRVC) and ventricular beat rate (VBR) from 1 h electrocardiogram(ECG) segments were extract as candidate features. Then, the feature dimension was reduced by recursive feature elimination. Finally, training set of normal control and SCD victims was employed to build multi-layer perceptron (MLP) model. To evaluate the model’s predictive ability, best cut-off threshold of high and low SCD risk was determined using Youden index. The SCD risk probability based on 1 h HRVC was 0.00 ± 0.01 for normal control and 0.99 ± 0.01 for SCD victims. The near-perfect results were achieved for discriminating SCD from normal control. Our method not only estimated individualized SCD risk probability reliably, but also had higher prediction accuracy than the existing methods.