The Application Of Statistical Models In The Prediction Of Difficult Intubation

Objectives Evaluate the effects of methods which are most often used in the prediction of difficult intubation in clinical practice and discuss how to establish multivariate statistical models for the prediction of difficult intubation, so that we can improve the accuracy of the preoperative prediction for tracheal intubation. Compare the differences and similarities among these methods, give advices on choosing the proper method in the prediction of difficult intubation and give some suggestions for the relative study.Methods The data of the patients in West China Hospital undergoing tracheal intubation were analyzed in detail. According to the Cormack and Lenhane scale, we analyzed the accuracy of the methods that most frequently used in clinical practice and established Fisher's discriminant analysis model and decision tree models. Compare the accuracy of these methods by the areas under the ROC curve and discuss the advantages and disadvantages of these methods.Results The Mallampati Test and the thyromental distance, methods used inclinical practice most frequently, were used to predict the difficult intubation. The total accuracy of these two methods is 69.4% and 58.1% respectively for all patients, but to the patients of difficult intubation the accuracy is only 38.3% and 44.2%. Establish the multivariate statistical models, i.e. the Fisher's discriminant analysis model. It's total accuracy is 77.4% and the accuracy to the patients of difficult intubation is 79.6%. Then we establish the decision tree models by CART algorithm and C4.5 algorithm respectively. CART model contains 7 variables and 10 leaves while C4.5 model contains 6 variables and 8 leaves. Finally we extract 5 rules from CART model and 4 rules from C4.5 model. The accuracy of these two models is 81.1% and 80.1% respectively. Use ROC curve to evaluate the effect of these methods, the thyromental distance was poor at identifying difficult intubation and the Mallampati test was not very good as well, whereas the discriminant analysis model and the decision tree were good at identifying difficult intubation.Conclusions The accuracy of the most frequently used methods in the prediction of difficult intubation in the clinical practice was low relatively. Established multivariate statistical models, the discriminant analysis model and the decision tree model, for difficult tracheal intubation, and both of them could improve the accuracy conspicuously, especially the decision tree. The decision tree was more valuable in clinical practice because of its concise results and higher accuracy.