[A clinical prediction model for failure to attain standard body weight in the early postoperative period after sleeve gastrectomy: construction and validation based on feature screening via the Boruta algorithm].

<b>Objective:</b> Significant individual variability exists in weight loss outcomes following sleeve gastrectomy (SG). To enable a more comprehensive and multi-scenario individualized assessment of postoperative weight loss, this study aims to construct and validate a model to predict the failure to achieve weight normalization in the early postoperative period. <b>Methods:</b> A retrospective study was conducted involving patients with obesity who underwent SG at the Xinjiang Uygur Autonomous Region People's Hospital between January 2021 and August 2024. Patients were classified into an early weight normalization group [body mass index (BMI) <25 kg/m²] and a non-normalization group (BMI ≥ 25 kg/m²) based on their weight status within one year post-surgery. Feature variables were selected using the Boruta algorithm. A prediction model was constructed using multivariate logistic regression, and its performance of the model was evaluated separately using the validation dataset, the Bootstrap method and 10-fold cross-validation. Model validity was further assessed in the validation cohort using calibration curves, Receiver Operating Characteristic (ROC) curves, and Decision Curve Analysis (DCA). <b>Results:</b> A total of 257 patients were enrolled and randomly divided into a training cohort (<i>n</i>=193) and a validation cohort (<i>n</i>=64) at a 3∶1 ratio. There were no statistically significant differences between the two groups in terms of postoperative early body weight achievement, sex, marital status, age, physical measurement indicators, visceral fat area (VFA), and laboratory indicators including routine blood test, thyroid function, fasting insulin, fasting C-peptide, blood lipid, electrolytes, liver and renal function, albumin (all <i>P</i>>0.05), indicating that the two groups were comparable. In the training cohort, compared to the non-normalization group (<i>n</i>=137), the weight normalization group (<i>n</i>=56) had a higher proportion of female (<i>χ</i>²=6.588, <i>P</i>=0.010). Conversely, preoperative weight, height, VFA, white blood cell count, red blood cell count, lymphocyte count, fasting insulin, fasting C-peptide, alanine aminotransferase (ALT), gamma-glutamyl transferase (GGT) were significantly lower in the normalization group (all <i>P</i><0.05). The key predictors identified by the Boruta algorithm included preoperative body weight, VFA, fasting insulin, and fasting C-peptide. The Area Under the Curve (AUC) was 0.857 (95%<i>CI:</i> 0.800-0.914) in the training cohort and 0.779 (95%<i>CI</i>: 0.655-0.904) in the validation cohort. The 10-fold cross-validation yielded an AUC of 0.843 (95%<i>CI</i>: 0.830-0.857), and Bootstrap validation showed an AUC of 0.857 (95%<i>CI</i>: 0.853-0.862). Calibration curves demonstrated excellent consistency between predicted and actual outcomes, and DCA indicated significant net clinical benefit across a wide range of threshold probabilities. <b>Conclusion:</b> A prediction model was successfully constructed and validated based on four preoperative indicators: body weight, VFA, fasting insulin and fasting C-peptide. It provides an auxiliary predictive tool for the clinical assessment of the risk of failing to attain standard body weight in the early postoperative period among obese patients undergoing SG.