A machine learning-based study of serological biomarkers for predicting intestinal necrosis in patients with adhesive small bowel obstruction

ObjectiveTo explore the value of machine learning-based serological markers in predicting irreversible transmural intestinal necrosis (ITIN) in surgical patients with adhesive small bowel obstruction (ASBO). MethodsA total of 133 ASBO patients who underwent surgical treatment at Xuzhou Central Hospital from February 2023 to February 2025 were prospectively enrolled. According to intraoperative exploration and pathological results, patients were divided into necrosis group (n = 68) and non-necrosis group (n = 65). Fourteen indicators were assessed, including serum homocysteine (HCY), endotoxin, procalcitonin (PCT), C-reactive protein (CRP), interleukin-6 (IL-6), IL-1β, IL-5, neutrophil gelatinase-associated lipocalin (NGAL), lactate dehydrogenase (LDH), vitamin B12 (VB12), folate, and age, gender, and body mass index (BMI). Twenty machine learning models were constructed. The dataset was randomly divided into a training set (n = 106) and a test set (n = 27) at an 8:2 ratio. Model performance was evaluated on the test set using ROC curves, decision curve analysis (DCA), calibration curves, and SHAP feature importance analysis was performed. ResultsLevels of HCY, endotoxin, PCT, and CRP were higher in the necrosis group than in the non-necrosis group (all P ＜ 0.05). The Extra Trees model demonstrated optimal performance with an AUC of 0.977 (95% CI: 0.955-0.999), sensitivity of 92.6% (95% CI: 83.9%-96.8%), and specificity of 95.4% (95% CI: 87.3%-98.4%). SHAP analysis identified HCY as the most important predictor (mean |SHAP value| = 0.119 6), followed by endotoxin (0.1008) and CRP (0.0557). Decision curve analysis showed that within a threshold probability range of 0.2-0.8, the net benefit of the Extra Trees model was significantly higher than that of the “treat-all” or “treat-none” strategy. The calibration curve demonstrated good agreement (Brier Score = 0.098). ConclusionsA machine learning-based multi-biomarker models can accurately predict the risk of intestinal necrosis in surgical ASBO patients, with the Extra Trees model showing the best performance. HCY is the most important predictor, providing an objective basis for preoperative clinical risk assessment. Future development of a comprehensive prediction model applicable to conservatively treated ASBO patients is needed.