Development and validation of a clinical model to predict low-grade intraepithelial neoplasia in chronic atrophic gastritis patients: a retrospective observational multicenter analysis.

BACKGROUND

Chronic atrophic gastritis (CAG), an early stage of gastric cancer, is a major digestive disorder, and the prognosis of CAG is determined by many sociodemographic and clinicopathologic subject characteristics. This retrospective observational multicenter analysis was conducted to explore risk factors and construct a predictive model for low-grade intraepithelial neoplasia (LGIN) in patients with CAG.

METHODS

The training dataset included 317 CAG patients diagnosed and treated in the Second Affiliated Hospital of Anhui University of Chinese Medicine from September 2018 to January 2025. All the baseline characteristics, including gender, age, education, basic diseases, blood indicators, and pathological mechanism during treatment of CAG, were recorded and selected based on both the least absolute shrinkage and selection operator (LASSO) regression analysis with 10-fold cross-validation and logistic regression analysis. After that, the nomogram was established, and its accuracy and predictive performance were evaluated via the area under the receiver operating characteristic (ROC) curves (AUC), calibration curves, Hosmer-Lemeshow goodness-of-fit test, and decision curve analysis (DCA) curves. For the validation dataset, the medical record information of 92 CAG patients diagnosed and treated in the Hefei Second People's Hospital from November 2023 to January 2025 was recorded for subsequent analysis.

RESULTS

Our LASSO regression analysis revealed that family history, HP infection, pepsinogen I, pepsinogen II, bile reflux, and Kimura-Takemoto classification (C3 vs. C1) were significant independent risk factors, and the fitting equation was obtained. A nomogram for predicting LGIN in CAG patients was established. The ROC curve revealed that our predictive model showed good predictive efficacy with an AUC value of 0.838 (95% CI = 0.789-0.887) with a specificity of 0.761 and a sensitivity of 0.791 in the training dataset and an AUC value of 0.941 (95% CI = 0.893-0.989) with a specificity of 0.852 and a sensitivity of 0.908 in the validation dataset. Moreover, calibration and DCA curves demonstrated that our predictive model had a good fit, better net benefit, and predictive efficiency in LGIN in CAG patients.

CONCLUSIONS

Our predictive model demonstrated that family history, HP infection, pepsinogen I, pepsinogen II, bile reflux, and Kimura-Takemoto classification were the independent risk factors of LGIN in CAG patients with high accuracy and good calibration.