Machine learning in the prediction of in-hospital mortality in patients with first acute myocardial infarction
Xiaoli Zhu,Bojian Xie,Yijun Chen,Hanqian Zeng,Jinxi Hu
DOI: https://doi.org/10.1016/j.cca.2024.117776
IF: 6.314
2024-02-01
Clinica Chimica Acta
Abstract:BACKGROUND: Persistent efforts are required to further reduce the in-hospital mortality of patients suffering from acute myocardial infarction (AMI), even in the face of a global trend of declining AMI-related fatalities. We investigated deep machine learning models for in-hospital death prediction in patients on their first AMI.METHOD: In this 2-center retrospective analysis, first AMI patients from Hospital I and Hospital II were included; 4783 patients from Hospital 1 were split in a 7:3 ratio between the training and testing sets. Data from 1053 AMI patients in Hospital II was used for further validation. 70 clinical information and laboratory examination parameters as predictive indicators were included. Logistic Regression Classifier (LR), Random Forests Classifier (RF), eXtreme Gradient Boosting (XGB), Support Vector Machine Classifier (SVM), Multilayer Perceptron (MLP), Gradient Boosting Machine (GBM), Bootstrapped Aggregation (Bagging) models with AMI patients were developed. The importance of selected variables was obtained through the Shapley Additive exPlanations (SHAP) method. The performance of each model was shown using the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (Average Precision; AP).RESULT: The in-hospital mortality for AMI in the training, testing, and validation sets were 5.7 %, 5.6 %, and 6.0 %, respectively. The top 8 predictors were D-dimer, brain natriuretic peptide, cardiogenic shock, neutrophil, prothrombin time, blood urea nitrogen, cardiac arrest, and phosphorus. In the testing cohort, the models of LR, RF, XGB, SVM, MLP, GBM, and Bagging yielded AUROC values of 0.929, 0.931, 0.907, 0.868, 0.907, 0.923, and 0.932, respectively. Bagging has good predictive value and certain clinical value in external validation with AUROC 0.893.CONCLUSION: In order to improve the forecasting accuracy of the risk of AMI patients, guide clinical nursing practice, and lower AMI inpatient mortality, this study looked into significant indicators and the optimal models for predicting AMI inpatient mortality.
medical laboratory technology