Development of A Machine Learning Model for Predicting Unanticipated Difficult Tracheal Intubation

Abstract

The valid prediction of unanticipated difficult tracheal intubation (DTI) is very important in clinic anesthesia. The purpose of this study was to develop a machine learning model for predicting unanticipated DTI. A comprehensive analysis of two prospective observational difficult airway research programs was performed. In total, 3958 patients who underwent tracheal intubation were included in this study. Data were split into a training set and a test set according to 70%:30% randomly. XGBoost machine learning was used to develop a machine learning model for predicting unanticipated DTI. The F1 score was used as the main performance metric because of data imbalance. The model parameter tuning was performed in the training set via pipeline grid search with the aim of optimizing the F1 score. Then, the tuning model were used for unanticipated DTI prediction in the test set. The indicators feature importance and decision rule were analyzed. With the XGBoost machine learning model for unanticipated DTI prediction, the best F1 score of 0.500 ± 0.102 was obtained on the training set with ten-fold cross-validation. The XGBoost model had the area under the precision recall curve (AUPRC) 0.600 and the area under the receiver operating characteristic curve (AUROC) 0.924 with an F1 value of 0.57 in the test set. XGBoost was an effective machine learning model for unanticipated DTI prediction.