Construction and Application of Early Warning Model for Ischemic Colitis in Emergency Patients Based on Machine Learning

Introduction: Ischemic colitis (IC), caused by reduced blood flow to the intestines, often presents with nonspecific early symptoms, leading to diagnostic delays and severe complications like necrosis or perforation. Current diagnostics (clinical evaluation, lab tests, imaging) lack sensitivity and specificity in early stages, highlighting the need for new predictive tools. This study proposes a machine learning model integrating clinical data, blood tests, and imaging descriptors to enable early IC detection at initial medical contact. Methods: Data from IC patients’ initial visits (Oct 2015–Jun 2022, Wenzhou Medical University) were analyzed. Mutual information selected key features; six models (e.g., random forest, logistic regression) were built. The top-performing model was streamlined and externally validated using first-contact data from Ningbo Second Hospital. Results: The random forest model, derived from first medical contact data of 427 IC patients and 507 control patients, demonstrated the highest performance, achieving an area under the curve (AUC) of 0.9251 and an accuracy of 0.8936 in the test data set. The model, optimized with 21 critical features, showed an AUC of 0.9191 and an accuracy of 0.8510. External validation yielded an AUC of 0.9963 and an accuracy of 0.9369. Conclusions: The RF-based IC model achieved superior diagnostic accuracy. Post-optimization, it maintained performance and demonstrated strong generalizability in external validation, underscoring its clinical utility.