Abstract:Objective To construct a warning model of lower respiratory tract infection in patients with glottic carcinoma based on SMOTE algorithm, so as to provide reference for preventing postoperative lower respiratory tract infection of these patients in clinic. Methods The clinical data of 596 glottic carcinoma patients undergoing tumor radical operation were investigated, and single factor comparison and logistic regression analysis were utilized to screen the influencing factors of postoperative lower respiratory tract infection, then the original data set of influencing factors was reconstructed using the SMOTE algorithm to obtain a warning model of postoperative lower respiratory tract infection for glottic carcinoma patients. Results Totally 38 patients (6.38%) suffered from postoperative lower respiratory tract infection. Logistic regression analysis revealed that, age, BMI, smoking history, diabetes, lung disease, operation type, postoperative sputum aspiration, intraoperative bleeding volume, postoperative extubation time and use of artificial nose were the influencing factors of postoperative lower respiratory tract infection for glottic carcinoma patients (P<0.01 for all). When compared the origin warning model with the warning model based on SMOTE algorithm, there was no significant difference in TPR, while the PPV, F-score and AUC were significant (P<0.05, P<0.01). Conclusion The warning model based on SMOTE algorithm is better than the original one, which can accurately predict postoperative lower respiratory tract infection for laryngeal cancer patients, then take targeted intervention to prevent postoperative lower respiratory tract infection.