Abstract
INTRODUCTION: Osteoporosis that emerges subsequent to gastrectomy poses a significant threat to the long-term health of patients. The primary objective of this investigation was to formulate a machine learning algorithm capable of identifying substantial preoperative, intraoperative, and postoperative risk factors. This algorithm, in turn, would enable the anticipation of osteoporosis occurrence after gastrectomy. METHODS: This research encompassed a cohort of 1125 patients diagnosed with gastric cancer, including 108 individuals with low bone density or osteoporosis. A total of 40 distinct variables were collected, comprising patient demographics, pertinent medical history, medication records, preoperative examination attributes, surgical procedure specifics, and intraoperative details. Four distinct machine learning algorithms-extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM), and k-nearest neighbor algorithm (KNN)-were employed to establish the predictive model. Evaluation of the models involved receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Shapley additive explanation (SHAP) was employed for visualization and analysis. RESULTS: Among the four prediction models employed, the XGBoost algorithm demonstrated exceptional performance. The ROC analysis yielded excellent predictive accuracy, showcasing area under the curve (AUC) values of 0.957 and 0.896 for training and validation sets, respectively. The calibration curve further confirmed the robust predictive capacity of the XGBoost model. The DCA demonstrated a notably higher benefit rate for patients undergoing intervention based on the XGBoost model. Moreover, the AUC value of 0.73 for the external validation set indicated favorable extrapolation of the XGBoost prediction model. SHAP analysis outcomes unveiled numerous high-risk factors for osteoporosis development after gastrectomy, including a history of chronic obstructive pulmonary disease (COPD), inflammatory bowel disease (IBD), hypoproteinemia, postoperative neutrophil-to-lymphocyte ratio (NLR) exceeding 3, steroid usage history, advanced age, and absence of calcitonin use. CONCLUSION: The osteoporosis prediction model derived through the XGBoost machine learning algorithm in this study displays remarkable predictive precision and carries significant clinical applicability.