Prognostic Role of Electrocardiographic Alternans in Heart Failure Patients with Implanted Cardioverter Defibrillator: Comparison of Machine Learning Methods

Implantable cardioverter defibrillator (ICD) is often indicated for the primary prevention of sudden cardiac death in heart-failure (HF) patients, but sometimes, the device remains always inactive, highlighting that the implantation criterium is not specific. The aim of this study is to assess if electrocardiographic alternans (ECGA; an index of cardiac instability) can have a useful prognostic role in improving the identification of HF patients who will experience serious ventricular arrhythmias and truly benefit from the ICD. We analyzed the Leiden University Medical Center database of primary prevention ICD patients by computing ECGA using the enhanced adaptive matched filter (EAMF) method. Patients were categorized into those who needed ICD therapy (40 cases) and those who did not (82 controls) based on their follow-up. ECGA features were used to train and test five machine learning methods (i.e., Decision Tree-DT, Logistic Regression-LR, Naïve Bayes-NB, Linear Discriminant Analysis-LDA, Support Vector Machine-SVM), whose performance was assessed by computing sensitivity (SE), specificity (SP), F1 score (F1) and accuracy (ACC). Results indicated that SVM was the most suitable algorithm (SE=98%; SP=83%; F1=96%; ACC=94%), followed by DT and LR, and ECGA appeared to be a potentially useful tool to improve identification of patients benefiting from ICD.