Prediction of sudden cardiac death for chagasic patients

Abstract

Objective: Identify the risk of patients with Chronic Chagas Cardiomyopathy (CCC) to prevent them from having Sudden Cardiac Death (SCD). Methods: We developed an SCD prediction system using a heterogeneous dataset of chagasic patients evaluated in 9 state-of-the-art machine learning algorithms to select the most critical clinical variables and predict SCD in chagasic patients even when the interval between the most recent exams and the SCD event is months or years. Results: 310 patients were analyzed, being 81 (14,7%) suffering from SCD. In the study, Balanced Random Forest showed the best performance, with AUC:80.03 and F1:75.12. Due to their high weights in the machine learning classifiers, we suggest Holter - Non-Sustained Ventricular Tachycardia, Total Ventricular Extrasystoles, Left Ventricular Systolic Diameter, Syncope, and Left Ventricular Diastolic Diameter as essential features to identify SCD. Conclusion: The high-risk pattern of SCD in patients with CCC can be identified and prevented based on clinical and laboratory variables.