Identification Of The Risk Factors Of Postperative Complications And Construction Of Prediction Model For Gastric Cancer

Objective:Identification of the risk factors of postoperative complications based on univariate and multivariate analysis for gastric cancer.Construction of a classification model based on risk factors using support vector machines,aims to guide clinical prevention and treatment of postoperative complications.Methods:Collection of patient data from January 2015 to June 2017,449patients with gastric cancer from the First Affiliated Hospital of Guangxi Medical University were selected,and 426 patients who had undergone radical resection and complete indexes were included in this study.Univariate(χ~2 test for count data,rank sum test for measurement data using)were performed to identify the differential distribution of factors between groups with or without complications.Furthermore,univariate and multivariate analyses(COX proportional hazard model)were used to investigate which clinical factors were correlated with postoperative complications.Then,support vector machines were used to construct the classification model based on identified risk factors(based on function“svm”obtained from R package“e1071”),and predict the statuses of complication for each patient and test the accuracy of the classification using leave one out cross validation.Results:(1)Univariate analysis showed that gender,age,ADL,type of incision,operation time,bleeding/weight ratio had significant differences in the distribution of presence/absence of complications(P<0.05);(2)Multivariate analysis showed that age,incision type,operation time,and bleeding/weight ratio were risk factors for postoperative complications(P<0.05);(3)Four postoperative complications risk factors were used as the classification features to construct a classifier using support vector machine,and a leave-one-out method was used to verify the accuracy of the classification results(AUC is 0.883),suggesting good classification efficiency.Conclusion:Patient age,incision type,operative time,and bleeding/weight ratio were identified as risk factors for postoperative complications.Prediction models based on risk factors can effectively distinguish between patients with and without complications.