Risk Factors And Separated Prediction Models Construction Of Patients With Gastric Neuroendocrine Tumors

【Objective】This study aimed to establish effective nomograms to estimate the prognosis of gastric neuroendocrine carcinoma（GNEC）and gastric neuroendocrine tumor（GNET）respectively.【Methods】The clinical data of patients who were pathologically diagnosed as GNENs（including GNEC and GNET）with complete follow-up data from 2004 to 2015were extracted from the SEER*Stat software,demographic and clinical pathological data collected,randomly assigned to training and validation sets.Kaplan-Meier survival curves were plotted,Log-rank test was used for Univariate analysis,COX regression was used for multivariate analysis,and nomogram was used to visualize the Prognostic Prediction Model construction of the COX regression scoring system,identifing potential predictors of OS.The patients were randomly assigned to training and validation cohorts with a ratio of 7:3.OS was set as the study endpoint.Univariate and multivariate Cox regression analysis were performed to determine independent predictors and construct nomograms.Based on the above results,we developed nomograms to predict the OS of GNEC and GNET patients at 3 and 5 years respectively,and verified them.It was by means of the calibration curve and receiver operating characteristic（ROC）curve to prove the reliability and accuracy of the clinical prediction model.【Result】Combining with Multivariate analysis,prognostic factors including age（P<0.05）,surgery and TNM stage were included for the nomograms of both GNEC and GNET.In addition,Gender,Histologic grade（P<0.05）,N stage were selected as predictors for the prognostic nomograms prediction model construction of GNEC.In the calibration plots of GNEC and GNET in both training and validation sets,we achieved an optimal agreement between the 3-and 5-year survival rates predicted by nomograms and the actual survival rates.Among them,GNEC:training set C Index 0.832（95%CI=0.803-0.862）,validation set C Index 0.797（95%CI=0.738-0.857）;GNET:training set C Index 0.749（95%CI=0.667-0.832）,validation set C Index 0.807（95%CI:0.699-0.915）.The AUCs of the nomograms for predicting the OS of validation set were0.913 for 3-year and and 0.887 for 5-year,which were higher than those of the 8^thAJCC staging system（0.816 and 0.789）.Excellent discrimination was observed in the GNET validation cohorts（AUCs of nomogram vs AJCC staging for 3-year OS:0.78 vs 0.589;5-year OS:0.779 vs 0.602）.Based on the results,visual nomograms were separately developed and validated for predicting 3-,and 5-year OS in GNEC and GNET patients^[3].The calibration,receiver operating characteristic（ROC）curve also demonstrated the reliability and accuracy of the clinical prediction model.【Conclusion】Comparing with the 8^thedition of the AJCC staging system,the nomogram prediction models of gastric neuroendocrine tumors and gastric neuroendocrine cancers that we established basing on the SEER database had better performance.Our predictive model is expected to be a personalized and easily applicable tool for evaluating the prognosis of GNEC patients,and may contribute toward making an accurate judgment in clinical practice.