Prediction of Airway Management of Trauma Patients Based on Machine Learning

Objective: To explore the application of different machine learning models to predict whether patients need endotracheal intubation, and to screen the key indicators by the classifier XGBoost algorithm based on decision tree algorithm. Methods A total of 158 patients with endotracheal intubation and 1003 patients without endotracheal intubation were collected from the first aid database of the General Hospital of the People's Liberation Army of China, and labeled as experimental group and control group respectively. After screening, 53 physiological indicators were obtained, and four classifiers, namely Logistic regression, support vector machine and Adaboost and XGBoost based on weak decision tree classifier, were used to predict the results. Results The average F-score of XGBoost model was 0.8414, and the area under receiver operating characteristic curve (AUROC) was 0.9103, which was the best among the four classifiers. The Adaboost model showed the best performance next, with an F-score of 0.8222 and the AUROC of 0.8935. The two algorithms based on decision tree are superior to Logistic Regression and SVM in this experiment. Conclusion Compared with the prediction model based on Logistic Regression, SVM or Adaboost, the prediction model based on XGBoost algorithm performed better, and could more effectively assist clinicians in the decision of airway management of injured patients.