Prediction of all-cause mortality in cardiovascular patients by using machine learning models

Abstract

Abstract Background Cardiovascular disease (CVD) is the leading cause of death worldwide. The prediction of all-cause death in patients with chronic cardiovascular risk will enable improved preventive strategies. Purpose This study aims to predict death of patients, who suffering from cardiovascular disease, within a follow-up period of 20 years. For this purpose was utilized the Ludwigshafen Risk and Cardiovascular Health (LURIC) study was utilized and employed machine learning models. Methods The Ludwigshafen Risk and Cardiovascular Health (LURIC) study was used in this analysis, which includes 3,316 patients (mean age 61.98 years, 69.2% male, 1722 deceased patients and 1594 alive or censored patients). After performing feature selection, 23 clinical and laboratory markers were included in our analysis. Pre-processing techniques such as SelectKbest and KNNImputer, were applied to the dataset and then it was splitted into a train and test (30%) set. The prediction of 20-year-risk of death was accomplished using the following machine learning algorithms: Logistic Regression (LR), Support Vector Machine (SVM), eXtreme Gradient Boosting (XGB) and Adaptive Boosting (AdaB). 10-fold cross validation was applied to verify the results of the models. Results The efficiency of each machine learning algorithm has been evaluated. XGB outperformed the others, achieving the highest accuracy (76.00%) and Area under (AUC) the Receiver Operating Characteristic Curve (ROC curve) value (0.776). The AUC value of the LR, SVM and AdaB are 0.761, 0.746 and 0.751, respectively. Figure 1 presents the ROC curve and AUC values for all models. Furthermore, the accuracy (ACC), precision, specificity, recall, F1-score and standard deviation were estimated for each classifier (Table 1). Conclusion We predicted all-cause mortality in patients with 20 years cardiovascular risk, using machine learning models and data of 3,316 patients. The most accurate predictive model was achieved by the eXtreme Gradient Boosting, with an accuracy equal to 76.00% and an AUC value of 0.776. Funding Acknowledgement Type of funding sources: Public grant(s) – EU funding. Main funding source(s): HORIZON2020