A Risk Prediction Model For Peritoneal Metastasis Of Early Gastric Cancer Based On Weighted Gene Co-expression Network Analysis

Purpose:Gastric cancer is one of the most common malignant diseases in our country.Peritoneal metastasis is one of the main reasons for poor prognosis of gastric cancer,and early diagnosis is difficult and the curative effect is poor.However,there is no risk prediction model of peritoneal metastasis for gastric cancer patients in early stage（I-III）has been established and widely used.The purpose of this study is to establish a model for predicting the risk of peritoneal metastasis in patients with early gastric cancer and to study the prognostic value of the model.Methods:In this study,84 patients with gastric cancer with complete clinical information of stage I-III from 2013.5 to 2023.2 in the Department of Oncology,the First affiliated Hospital of China Medical University were collected retrospectively.Multivariate survival analysis was used to determine independent prognostic factors and establish a clinical factor model,and GSE62254 data set was used to verify the clinical factor model.For more comprehensive and accurate prediction,using the data of GSE62254,early patients were randomly divided into training set and verification set according to 60%vs 40%.In the training set,peritoneal metastasis-related genes were screened by WGCNA and lasso analysis,and a combined model of clinical factors and gene expression characteristics was established to predict the risk of peritoneal metastasis.Use validation sets for validation.The combination model was evaluated and compared with the clinical factor model.The difference between the two models was determined by self-weight sampling and t-test.Draw KM curve and ROC curve to explore the prediction effect of combined model on OS.Results:Sex,age,Borrmann type,local infiltration,T stage,N stage and clinical stage were analyzed,and clinical stage and sex were obtained as independent prognostic factors of peritoneal metastasis.a clinical factor model based on clinical stage and sex was established,and the predictive efficiency of peritoneal metastasis was higher than that of clinical staging alone(AUC,66%vs 55%;P=2.2e^-16)and validated in the external validation set GSE62254.In order to make a more comprehensive and accurate prediction,four peritoneal metastasis related genes（ANKRD35,PDE1A,PLAC9 and TCEAL2）were screened by WGCNA combined with lasso analysis in the training set of GSE62254,and a combined model for predicting the risk of peritoneal metastasis was established.The high and low risk groups of the combination model risk score observed significant differences in peritoneal metastasis（HR=30.93;95%CI,4.05-236.47）,and were verified in the verification set.The predictive efficiency of the combined model for peritoneal metastasis was higher than that of the simple clinical factor model（AUC,86%vs 66%;P=8.5e-16）.The combined model also has a certain predictive efficiency for OS（AUC 66%）.Conclusion:This study established a combined model that can predict the risk of peritoneal metastasis in patients with I-III gastric cancer,including clinical and gene expression characteristics.This model will contribute to the more accurate stratification of patients with gastric cancer and promote the realization of accurate medical care for patients with stratification.