Abstract / Summary
Preoperative anemia is common in gastric cancer and may adversely affect patient outcomes, particularly among women. This study aimed to evaluate the prognostic significance of anemia and to construct a machine learning model based on inflammation–nutrition-related indices for individualized survival prediction. A total of 1,178 female gastric cancer patients undergoing curative R0 resection were retrospectively analyzed. Kaplan–Meier and Cox regression analyses assessed the prognostic role of anemia. Inflammatory and nutritional indices were selected using recursive feature elimination, and multiple machine learning algorithms were developed to predict 5-year overall survival. Model performance and interpretability were evaluated using AUC metrics and SHAP analysis. Anemia was identified as an independent adverse prognostic factor. Distinct prognostic patterns of inflammation–nutrition indices were observed between anemic and non-anemic groups. The XGBoost model achieved the best predictive performance (AUC = 0.735 and 0.773, respectively) and provided interpretable risk stratification, with high-risk patients showing significantly poorer survival across subgroups. The XGBoost model based on routine hematologic indicators showed promising predictive performance in this single-center proof-of-concept study. With further external validation and prospective evaluation, it may serve as a simple and readily available tool for preliminary risk assessment. However, its current clinical utility remains to be confirmed in broader patient populations.