Construction and validation of a prognostic prediction model for poorly cohesive gastric adenocarcinoma based on inflammatory indices and machine learning

This study aims to construct a survival prognostic prediction model for poorly cohesive gastric adenocarcinoma patients by combining inflammatory indices and machine learning methods, and to evaluate its clinical application value. A total of 722 patients with poorly cohesive gastric adenocarcinoma were recruited, and their clinical data and inflammatory indices (including prognostic nutritional index, lymphocyte-to-monocyte ratio [LMR], glucose-to-lymphocyte ratio [GLR], and fibrinogen-to-lymphocyte ratio [FLR]) were collected. Machine learning models (including XGBoost) were used to construct and validate the survival prognostic model, and the model’s performance was evaluated by AUC, accuracy, precision, recall, and F1 score. Single and multivariate Cox regression models were used to assess the independent prognostic value of each variable. Univariate and multivariate Cox regression analyses showed that pTNM stage, tumor size, and the XGBoost model were independent prognostic factors for gastric cancer patients’ survival. The XGBoost model showed a hazard ratio (HR) of 3.27 in the training set and 1.83 in the test set, demonstrating good prognostic prediction ability. Kaplan–Meier survival curve analysis indicated that the XGBoost model could significantly distinguish between high-risk and low-risk patients’ survival. The combination of inflammatory indices and machine learning methods enabled the model to provide accurate survival predictions for clinical physicians. This study constructed a survival prognostic model for patients with poorly cohesive gastric adenocarcinoma by integrating inflammatory indices and machine learning algorithms. The model showed promising predictive performance and may serve as a useful supplementary prognostic tool. It assists clinicians in reasonable risk stratification and individualized prognostic assessment, and provides reference evidence for optimizing clinical management strategies.