Improving Mortality Prediction in Cardiovascular Risk Patients by Balancing Classes

The objective of this work is to develop a prediction system for mortality risk based on ESCARVAL-RISK study, a large cardiovascular cohort (54,678 patients) with a follow-up starting in January 2008 through December 2012. The main challenge to face in this problem is the highly unbalanced classes that may lead to a poor performance in the classification, this work proposes a way to balance classes in order to circumvent that problem. Achieved results show that several factors emerged as the main factors of mortality in this population with a high cardiovascular (CV) risk, besides age and gender, treatments for hypertension, diabetes and dyslipidemia are the most relevant factors present in the cohort although the weight of each of them varies with the model. Despite the variation among the models, treatment for hypertension seems to be the most relevant factor. The presence of chronic kidney disease (CKD) defined by an estimated glomerular filtration rate.