Prognostic Prediction Model For Postoperative Patients With Stage Ⅲ Gastric Cancer Based On SEER Database

Objective:In this study,the clinical data of postoperative patients with stage Ⅲ gastric cancer were included based on the Surveillance,epidemiology,and end results(SEER)database of the National Cancer Institute of the United States,and the independent prognostic factors were explored and a nomogram was constructed.The aim of this study is to establish a more individualized prognostic prediction model and provide reference for clinical diagnosis and treatment.Methods:Case data of patients diagnosed with postoperative stage Ⅲ gastric cancer from 2010 to 2017 in the SEER database were included and randomly divided into training and internal validation cohorts.In addition,the data of patients diagnosed as postoperative stage Ⅲ gastric cancer in Hebei General hospital from January 2016 to December 2020 were retrospectively collected,and an external validation cohort was constructed based on the inclusion and exclusion criteria.A nomogram prediction model for overall survival(OS)at 1-year,3-year,and 5-year was constructed and survival curves were plotted using machine learning and Cox proportional hazards models.Then the prediction model was validated internally and externally,and the discrimination,consistency and clinical utility of the constructed prediction model were evaluated by area under the ROC curve(AUC),calibration plot and decision curve analysis(DCA),respectively.Finally,the temporal AUC curve,Akaike information criterion(AIC),Bayesian information criterion(BIC),net reclassification improvement(NRI)and integrated discrimination improvement(IDI)were used to compare predictive ability of the nomogram prediction model and the traditional AJCC eighth edition TNM staging system.Results:1.Baseline comparison between cohorts and nomogram model constructionThe final 4030 patients in the SEER database of this study were included in the analysis(training cohort: 2821;validation cohort: 1209),and231 patients were finally included in the Hebei General hospital.There were no significant differences in baseline characteristics between the training cohort and the internal validation cohort(P>0.05).There were significant differences in age,primary site,tumor size,pathological type,radiotherapy and chemotherapy between the training cohort and the external validation cohort(P<0.05).The results of machine learning method and multivariate Cox regression model showed that age,primary site,pathological grade,T stage,N stage,radiotherapy and chemotherapy were independent prognostic factors for postoperative patients with stage Ⅲ gastric cancer.A nomogram was constructed based on the independent prognostic factors in the multivariate Cox regression analysis.2.Validation of nomogram prediction modelsVariables in the internal and external validation cohorts were included in nomogram prediction models for model assessment,respectively.The AUC at 1-year,3-year,and 5-year was 0.753,0.706,0.708;0.712,0.692,0.681;0.728,0.693,and 0.646 for the training,internal validation,and external validation cohorts,respectively,indicating that the prediction model had good discrimination;the slopes of the calibration curves were close to the ideal slopes;and DCA showed good net benefit.3.Comparison of nomogram prediction models with AJCC eighth edition TNM staging systemThe AIC values of the nomogram prediction model and the AJCC staging system were 30391.00 and 30749.97,respectively,and the BIC values were 30481.50 and 30783.91,respectively,and the lower AIC and BIC values of the nomogram indicated a better goodness of fit.The 1-year,3-year,and 5-year NRI values were 0.278,0.058,and 0.038,respectively,and the IDI values were 0.091,0.062,and 0.045,respectively,indicating that the nomogram prediction performance was significantly improved.The 1-year,3-year,5-year AUC of nomogram and AJCC staging system were0.753,0.706,0.708,and 0.632,0.643,0.659,respectively.The change trend of time AUC curve of nomogram model was not inferior to that of AJCC staging system,indicating that the established nomogram prediction model had good discrimination.Conclusions:1.Age,primary site,pathological grade,T stage,N stage,radiotherapy and chemotherapy are independent prognostic factors for postoperative patients with stage Ⅲ gastric cancer.2.The internal and external validation results confirm that the nomogram prediction model has good prediction efficacy and has certain reference significance in clinical practice application.3.Compared with the AJCC eighth edition TNM staging system,the nomogram prediction model has better goodness of fit and differentiation,which may be used as a supplement to TNM staging.