Application of Machine Learning to Predict 1-Year Weight Loss Outcomes After Laparoscopic Sleeve Gastrectomy.

BACKGROUND: Laparoscopic sleeve gastrectomy (LSG) is an effective treatment for obesity. This study aimed to develop and internally validate machine-learning models using routinely available preoperative clinical variables to predict optimal clinical response at 1 year after LSG. METHODS: This single-center retrospective study included 621 individuals with obesity who underwent LSG between September 2021 and May 2025 and completed 1-year follow-up. Feature selection was performed using the Boruta algorithm. Five machine-learning models were developed and evaluated in a Training set and an Internal test set (7:3). Model performance was assessed using the area under the receiver operating characteristic curve (AUC), and model interpretability was examined using Shapley Additive Explanations (SHAP). RESULTS: The Boruta algorithm identified seven key predictors, including body mass index (BMI), fasting plasma glucose (FPG), waist circumference (WC), age, low-density lipoprotein cholesterol (LDL), white blood cell count (WBC), and estradiol (E2). Among the five models, the random forest (RF) model achieved the best predictive performance, with an AUC of 0.804 (95% CI: 0.728-0.881) and an accuracy of 0.796 in the Internal test set. SHAP analysis demonstrated that BMI, FPG, age, and WC were the most influential predictors. A web-based prediction tool was subsequently developed based on the RF model. CONCLUSIONS: Machine-learning models based on routinely available preoperative variables showed potential for predicting optimal clinical response at 1 year after LSG. The RF model demonstrated the best overall performance and may assist preoperative risk assessment and patient stratification. Further external validation is warranted.