Abstract
PURPOSE: The reappearance of gastric cancer, a frequent postoperative complication following radical gastric cancer surgery, substantially impacts the near-term and far-reaching medical outlook of patients. The objective of this research was to create a machine learning algorithm that could recognize high-risk factors for gastric cancer recurrence and anticipate the correlation between gastric cancer recurrence and Helicobacter pylori (H. pylori) infection. PATIENTS AND METHODS: This investigation comprised 1234 patients diagnosed with gastric cancer, and 37 characteristic variables were obtained. Four machine learning algorithms, namely, extreme gradient boosting (XGBoost), random forest (RF), k-nearest neighbor algorithm (KNN), and multilayer perceptron (MLP), were implemented to develop the models. The k-fold cross-validation technique was utilized to perform internal validation of the four models, while independent datasets were employed for external validation of the models. RESULTS: In contrast to the other machine learning models, the XGBoost algorithm demonstrated superior predictive ability regarding high-risk factors for gastric cancer recurrence. The outcomes of Shapley additive explanation (SHAP) analysis revealed that tumor invasion depth, tumor lymph node metastasis, H. pylori infection, postoperative carcinoembryonic antigen (CEA), tumor size, and tumor number were risk elements for gastric cancer recurrence in patients, with H. pylori infection being the primary high-risk factor. CONCLUSION: Out of the four machine learning models, the XGBoost algorithm exhibited superior performance in predicting the recurrence of gastric cancer. In addition, machine learning models can help clinicians identify key prognostic factors that are clinically meaningful for the application of personalized patient monitoring and immunotherapy.