Development and validation of machine learning-based prediction models for adult bowel necrosis for patients presenting with acute abdominal pain at the emergency department: a multicenter retrospective cohort study.
Authors
Affiliations (6)
Affiliations (6)
- Department of Gastrointestinal Surgery, Xuzhou Central Hospital, Southeast University, Xuzhou, Jiangsu Province, 221000, China.
- Department of Thyroid Surgery, The Third Affiliated Hospital of Soochow University, Changzhou First People's Hospital, Changzhou, Jiangsu Province, 213000, China.
- Department of General Surgery, The People's Hospital of Sishui, Jining, Shandong Province, 273200, China.
- Department of Breast Surgery, Jinan Maternity and Child Care Hospital, Jinan, Shandong Province, 250001, China.
- Department of Gastrointestinal Surgery, Xuzhou Central Hospital, Southeast University, Xuzhou, Jiangsu Province, 221000, China. [email protected].
- Department of Oncology, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu Province, 221000, China. [email protected].
Abstract
Bowel necrosis is a life-threatening condition requiring urgent surgical intervention. Early identification remains challenging due to nonspecific clinical presentations. This study aimed to develop and validate machine learning models for predicting bowel necrosis in adult patients. We conducted a multicenter retrospective cohort study including adult patients (≥ 18 years) presenting with acute abdominal pain between January 2021 and December 2023. Data were collected from electronic health records including demographics, clinical symptoms, laboratory parameters, and imaging findings. Six machine learning algorithms were developed and compared: logistic regression (LR), random forest (RF), support vector machine (SVM), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), and artificial neural network (ANN). Model performance was evaluated using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and calibration metrics. A total of 8,247 patients were included, with 623 (7.6%) diagnosed with bowel necrosis. The XGBoost model demonstrated superior performance with an AUC of 0.912 (95% CI: 0.887-0.936), sensitivity of 87.3%, and specificity of 84.7% in the validation cohort. Key predictive features included lactate level (importance score: 0.185), white blood cell count (0.142), abdominal CT findings (0.138), age (0.108), and C-reactive protein (0.095). The model showed adequate calibration (Hosmer-Lemeshow p = 0.324) and clinical utility with a net benefit of 0.142 at a 10% risk threshold. The machine learning-based prediction model demonstrates good discriminative performance in identifying patients at risk for bowel necrosis. Implementation of this model as a clinical decision-support tool may inform early triage and reduce diagnostic delays, potentially improving patient outcomes through timely surgical intervention. Prospective external validation is required before widespread clinical adoption.