Risk prediction for chronic atrophic gastritis using a random forest model: A multicenter study.

Area of Science:

• Gastroenterology and Oncology
• Biostatistics and Machine Learning
• Preventive Medicine

Background:

• Chronic atrophic gastritis (CAG) is a precancerous condition for gastric cancer (GC).
• Current diagnostic methods (gastroscopy, biopsy) are invasive and have limitations.
• Noninvasive tools are needed for early CAG detection.

Purpose of the Study:

• To develop and validate a noninvasive risk prediction model for CAG.
• To identify high-risk individuals for gastric cancer screening.

Main Methods:

• A random forest model was built using multimodal data from 1268 subjects.
• Features included demographics, lifestyle, H. pylori status, and serological markers (PGR).
• Model performance was evaluated using AUC and validated in an external cohort; SHAP analysis was used for interpretation.

Main Results:

• The model identified six key predictors: age, H. pylori infection, PGR, smoking, alcohol, and family history.
• The model achieved high AUCs (0.8542 training, 0.8073 testing, 0.8505 external validation).
• H. pylori infection, age, and PGR were the most influential factors.

Conclusions:

• The random forest model accurately predicts CAG risk noninvasively.
• The model is interpretable and has significant clinical utility for early screening.
• A web-based platform facilitates clinical application.