Abstract: Objective To analyze the risk factors of upper gastrointestinal cancer and precancerous lesions, to construct a prediction model for high⁃risk population of upper gastrointestinal cancer, and to identify high⁃risk groups of upper gastrointestinal cancer. Methods The population group at age 40-69 who participated in the "Rural Upper Gastrointestinal Cancer Early Diagnosis and Treatment Project " in Yangcheng County, Shanxi Province from June 2020 to December 2021 were selected, and randomly divided into a training set (n=1,997) and a validation set (n=852), according to the inclusion and exclusion criteria with a ratio of 7 to 3, used for training and validation of the model, respectively. The chi⁃square test was used for univariate analysis, the factors with P < 0.2 were screened for the optimal subset of variables, and the combination of variables with the lowest akaike information criterion (AIC) was selected to construct a logistic regression model and establish a scoring scale. The receiver operating characteristic curve (ROC) was drawn and the discrimination of the model was evaluated according to the area under the curve (AUC). The calibration of the model was evaluated by the Hosmer⁃Lemeshow (H⁃L) test and the calibration curve. The clinical decision curve analysis (DCA) was used to evaluate the clinical applicability. Results A logistic regression prediction model for the high⁃risk population of upper gastrointestinal cancer and precancerous lesions was established based on five risk factors, including age, sex, smoking, hot food intake and family history of cancer. The AUCs of the training set and the validation set of the model were 0.759 (95%CI: 0.689-0.830) and 0.743 (95%CI: 0.606-0.880), respectively. The calibration curve combined with H⁃L test proved that the model had good calibration (P>0.05). According to the logistic regression model, a scoring scale model was established with a score ranging from 0 to 27 points. The higher the score, the higher the risk of disease. The AUCs of the training set and the validation set were 0.760 (95%CI: 0.690-0.829) and 0.748 (95%CI: 0.612-0.884), respectively. The calibration curve combined with the H⁃L test proved that the model had good calibration (P>0.05).The DCA indicated that the model had good clinical applicability. Conclusions The prediction model and scoring scale model of the high⁃risk population of upper gastrointestinal cancer and precancerous lesions, based on the five risk factors including age, sex, smoking, hot food intake and family history of cancer, have good predictive value, which are helpful for the screening of the upper gastrointestinal cancer population.

Key words:  , Upper gastrointestinal cancer, Prediction model, Screening, Early diagnosis and treatment