Patients:
1002 patients with gastric cancer undergoing surgery without blood diseases, rheumatic diseases, acute cardiopulmonary diseases, severe infection and other tumor diseases were enrolled at Xinhua hospital affiliated to Shanghai Jiaotong University School of Medicine between January 1st, 2013 and January 1st, 2020. No patients received chemotherapy, radiotherapy and endoscopic therapy before surgery. No patients had received gastrectomy and had blood disease 1 month before surgery. All patients were classified according to the International Union against Cancer TNM Classification (Eight Edition): 272 patients had stage I disease, 172 stage II, 435 stage III and 123 stage IV. Patients with stage IV had no organ metastasis but PM. PM was diagnosed by laparotomy when metastasis nodules were found in the peritoneal cavity, postoperative pathology which confirmed metastasis nodules in the peritoneum, PET-CT or CT.
Objective clinical factors:
The patients’ data recorded included sex, age, clinical features (including hemorrhage of alimentary tract or not, hypertension or not and diabetes or not), primary tumor data (including macroscopic type, tumor location, degree of tumor differentiation, tumor size and serosal invasion or not) and preoperative laboratory data of blood (including blood routine examination, liver and kidney function, coagulation function and tumor markers). Macroscopic type was classified into ulcer type, protrusion type and infiltration type. Tumor location was categorized into cardia/fundus, body/antrum and pylorus. Degree of tumor differentiation was classified into poorly, moderate and well differentiation.
Statistical analysis and variables selection:
The correlation between variables (including clinicopathological features and preoperative laboratory data of blood) and PM was analyzed by chi-square test for categorical variables, t test for normally distributed variables and Mann-Whitney U test for non-normally distributed variables. The variables with significant statistical difference were further analyzed by univariate logistic regression to select variables as group 1. The analysis above were carried out via SPSS Statistics Software version 19.0. Next, extract complete data of variables of group 1 excluding the missing value, where LASSO (the least absolute shrinkage and selection operator) was utilized to select variables as group 2. Then the complete data of variables of group 2 extracted from original data set was divided into train set (n=782,PM positive rate was 11.1%) and test set (n=100) by stratified sampling according to peritoneal metastasis or not. The final variables as group 3 were selected by LASSO and 10-folds cross validation in train set to be used for nomogram. The analysis above were carried out via R software version 4.0.2. All p-values were two-sided and p<0.05 was considered statistically significant.
LASSO is a penalized regression approach that estimates the regression coefficients by maximizing the log-likelihood function (or the sum of square residuals) with the constraint that the sum of the absolute values of the regression coefficients proposed by Tibshirani in 1996[7]. LASSO can be applied to delete unnecessary covariates and select variables as it can constrain many components to exactly 0 automatically.
Development and evaluation of nomogram:
A nomogram model was developed on the basis of variables of group 3 and evaluated by receiver operating characteristic (ROC) curve and the area under the curve (AUC) values via R software version 4.0.2. In the nomogram, the regression coefficients of variables were proportionally converted to a specific number within 0 to 100 point scale. Internal validation, external validation and 10-folds cross validation were performed respectively to evaluate the accuracy and generalization capability of the nomogram model.