Identifying Ventricular Arrhythmia Cases And Their Predictors By Applying Machine Learning Methods To Electronic Health Records (Ehr) Of Hypertrophic Cardiomyopathy (Hcm) Patients

Introduction: Ventricular arrhythmias (VA) are a major cause of death in hypertrophic cardiomyopathy (HCM) patients. We develop a machine learning method to effectively identify VA in HCM patients using a small set of clinical variables. Methods: We scanned the EHR of 788 HCM patients, who underwent detailed clinical phenotyping, for sustained ventricular tachycardia/fibrillation. Patients with VA (61) were tagged as Arrhythmia cases and the remaining (727) as non-Arrhythmia. To identify the most informative variables for separating arrhythmia from non-arrhythmia we used the 2-sample t-test and power analysis. Patient records were reduced to include only these variables. Notably, the dataset is highly imbalanced, resulting in poor performance of conventional classifiers. While imbalance is often addressed by either over- or under-sampling of one of the classes, we apply a combination of both. We trained and tested multiple classifiers (including random forest and logistic regression), under this sampling ...