IDDF2024-ABS-0008 Using machine learning algorithms establish the prediction model of acute bowel injury in patients with heart failure

<h3>Background</h3> The objective of this study was to develop a prognostic model for in-hospital acute bowel injury among patients with heart failure. <h3>Methods</h3> The present investigation involved an analysis of the clinical data pertaining to heart failure patients derived from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database. The primary objective of this study was to employ machine learning algorithms, including decision tree, random forest (RF), support vector machine (SVM), K-nearest neighbor (KNN), and logistic regression (LR), to predict the occurrence of acute bowel injury (ABI) in heart failure patients. These algorithms considered various factors, including demographic characteristics, past medical history, laboratory test results, and treatment attributes. The RF algorithm, employing the Gini index, was utilized to identify the features correlated with the development of ABI. The effectiveness of the predictive model was assessed by evaluating metrics such as accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUROC). <h3>Results</h3> In this study, 778 patients with heart failure were included, and out of these, 319 developed ABI within the first 48 hours of admission to the intensive care unit. Compared to patients in the non-ABI group, those in the ABI group were older. Additionally, the ABI group had higher levels of serum creatinine, urea nitrogen, and sequential organ failure assessment (SOFA) scores. Among 5 machine learning algorithms, the RF algorithm exhibited the highest performance with the AUROC of 0.89, which was better than 0.83 of decision tree, 0.90 of SVM, 0.84 of KNN, and 0.90 of LR. In addition, the Gini index showed that SOFA score, partial pressure of oxygen (PaO2) were highly relevant to ABI development. <h3>Conclusions</h3> The utilization of machine learning algorithms enabled accurate prediction of ABI occurrence among heart failure patients.