Interpretable Machine Learning Models for Prognosis in Gastric Cancer Patients

This multicenter, retrospective cohort study aimed to develop and validate an explainable prediction model for prognosis after gastrectomy in patients with gastric cancer. The study included patients who underwent radical gastrectomy for primary gastric or gastroesophageal junction cancer across multiple institutions in China. The primary objective was to create a machine learning-based model to predict postoperative outcomes following gastrectomy, using readily available clinical and pathological parameters. The main outcome of interest was early recurrence within 2 years after surgery, which significantly impacts overall prognosis. The study employed various machine learning algorithms to develop prediction models, which were then compared and validated. Model performance was assessed through measures such as area under the receiver operating characteristic curve (AUC), calibration, and Brier score. The SHapley Additive exPlanations (SHAP) method was used to interpret the model and rank feature importance. This research aims to provide clinicians with a tool for identifying patients at higher risk of poor postoperative outcomes who may benefit from more intensive post-operative monitoring and early intervention strategies, potentially improving prognosis for gastric cancer patients.