To Predict The Occurrence Of Anastomotic Leakage After Esophagogastric Junction Tumor Operation By The Comparison Of Logistic Regression Model And Artificial Neural Network

Background:The incidence of esophageal and gastric junction tumors is very high in China,especially in gastrointestinal disease.The tumors located at the junction of the stomach and esophagus play an important role in digestive function.Therefore,both thoracic surgery and general surgery have studied them to a certain extent,the number of patients with esophageal-gastric junction tumor also increased significantly,so it is considered that esophageal-gastric junction tumor should be distinguished from esophageal cancer and gastric cancer,can not be generalized.Surgery is still the main treatment for patients with tumors at the esophageal-gastric junction.Because of the relative complexity of surgery for tumors at the esophageal-gastric junction,various complications often occur after surgery,and the complications are closely related to some risk factors,there was a significant correlation between the quality of life and the presence or absence of complications.Objective:Through the implementation of this project,it can quickly predict the probability of anastomotic leakage in patients with esophageal-gastric junction tumor after surgery,so that the clinician can not only predict by experience,it is a theoretical basis for early intervention,so as to reduce the incidence of anastomotic leakage after esophageal-gastric junction tumor surgery,change the status of the high incidence of anastomotic leakage after esophageal-gastric junction tumor surgery and improve the poor prognosis of patients.Methods:Retrospective analysis was conducted to collect basic information of postoperative patients with esophageal-gastric junction tumors in the Third Affiliated Hospital of Southern Medical University from 2013 to 2020,to clarify the types of postoperative complications and the related factors of postoperative anastomotic leakage at the junction of esophagus and stomach,and to establish the artificial neural network model and logistic regression model through statistical analysis,to compare the accuracy and sensitivity of the two models,an artificial neural network depth learning software related to anastomotic fistula was developed for clinical prediction.Results:(1)A total of 202 patients with esophageal-gastric junction tumors were treated by operation in the same treatment group of General Surgery in our hospital.20 patients(9.9%)suffered from esophageal-gastric anastomotic fistula after operation.(2)The risk factors of postoperative anastomotic leakage(hypoproteinemia,etc.)were obtained by literature investigation and the results were analyzed by single factor analysis,which were in accordance with P<0.25.(3)Using logistic regression model to carry on the multi-factor analysis,we can obtain the adherence sample accuracy:86.9%,the sensitivity:14.3%,the specificity:96.2%.(4)Using artificial neural network to build three-layer model,we can get the accuracy of stick sample:91.8%,the sensitivity:71.4%,the specificity:94.4%.(5)By plotting the ROC curve,we found that the artificial neural network was slightly better in predicting anastomotic leakage after the esophageal and gastric junction.Conclusion:First,sex,hypoproteinemia,diabetes,cardiovascular disease,and distant metastasis.Second,hypoalbuminemia was an independent risk factor for anastomotic leakage after esophagogastrostomy.Third,artificial neural network was superior to logistic regression model.