XGBoost model predicts acute lung injury after acute pancreatitis

Abstract

To develop an XGBoost model to predict the occurrence of acute lung injury (ALI) in patients with acute pancreatitis (AP). Using the case database of Xinhua Hospital affiliated to Shanghai Jiaotong University School of Medicine, 1231 cases suffering from AP were screened, and after 137 variables were identified, the clinical characteristics of the samples were statistically analyzed, and the data were randomly divided into a training set (75%) to build the XGBoost model and a test set (25%) for validation. Finally, the performance of the model was evaluated based on accuracy, specificity, sensitivity, and subject characteristics working characteristic curves. The model performance is also compared with that of three other commonly used machine learning algorithms (support vector machine (SVM), logistic regression, and random forest). The age and laboratory tests of patients with AP combined with ALI differed from those of patients without combined acute lung injury. The area under the receiver operating characteristic (ROC) curve of the test set after model evaluation was 0.9534, the specificity was 0.7333, and the sensitivity was 0.7857, with arterial partial pressure of oxygen, bile acid, aspartate transaminase, urea nitrogen, and arterial blood pH as its most important influencing factors. In this study, the XGBoost model has advantages compared with other three machine learning algorithms. The XGBoost model has potential in the application of predicting acute lung injury after acute pancreatitis.