Abstract:Objective: To establish an automatic early-warning model for predicting positive chyle test in pleural effusion of patients accepted lung cancer surgery and evaluate its early warning efficiency. Methods: 700 patients who underwent lung cancer surgery in the Department of Thoracic Surgery, the First Affiliated Hospital of Nanjing Medical University from January to December 2018 were enrolled for a prospective analysis. The general data, postoperative physiological indicators and drainage of pleural fluid were collected,independent risk factors of positive chyle test in postoperative pleural effusion were screened out by univariate and multivariate analysis. The binary Logistic regression model, neural network model and decision tree model were constructed by using SPSS 19.0 software,and the values of their area under curve (AUC) were compared by Medcalc software to select an optimal model. Finally, 131 patients accepted lung cancer surgery from the same hospital were included to test the optimal model. In the validation stage, the early-warning efficiency of the optimal model was judged by calculating sensitivity, specificity, positive predictive value, negative predictive value and overall accuracy. Results: The AUC for the binary logistic regression model,BP neural network model and decision tree model was 0.854,0.980,0.835, respectively. The neural network model was selected as the optimal model, and the difference was statistically significant after pairwise comparison. A total of 8 variables were included, in descending order of importance: pleural triglyceride, BMI, pulmonary resection range, N stage of tumor, postoperative albumin, age, occurrence of lymph node dissection, and T stage of tumor. In the validation group, its sensitivity was 89.9%, specificity was 91.9%, positive predictive value was 92.5%, negative predictive value was 89.1%, and the total accuracy was 90.8%. Conclusion: The early-warning model constructed by neural network for predicting positive chyle test in pleural effusion is scientific and effective, which can be used for screening high-risk patients with chylothorax early. Our study lays a foundation for the construction of perioperative nutrition management for lung cancer patients.