Early Prediction of Surgical Intervention in Neonates with Necrotizing Enterocolitis Using Machine Learning: A Retrospective Cohort Study

Abstract

Background: Necrotizing enterocolitis (NEC) is a devastating gastrointestinal emergency in neonates, frequently requiring surgical intervention. Early prediction of surgical necessity remains a major clinical challenge due to the rapid progression and heterogeneity of NEC presentations.

Methods: This study aims to develop and validate a machine learning (ML) model to predict the need for surgical intervention in neonates diagnosed with NEC using routine clinical and laboratory data available within the first 48 hours of diagnosis.

Results: A retrospective cohort of 298 neonates diagnosed with NEC (Bell Stage II or higher) between 2015 and 2024 was analyzed. Thirty-two clinical and biochemical parameters were extracted. Four ML algorithms—Logistic Regression (LR), Random Forest (RF), XGBoost, and Support Vector Machine (SVM)—were trained and evaluated. Model performance was assessed using area under the ROC curve (AUC), sensitivity, specificity, and F1-score. SHAP (SHapley Additive exPlanations) was used to enhance interpretability.

Conclusion: Of 298 neonates, 102 (34.2%) required surgery. XGBoost achieved the best performance (AUC=0.91, sensitivity=87%, specificity=84%, F1-score=0.86). Key predictors included serum lactate, CRP, platelet count, abdominal distension, and oxygen requirement.

The proposed ML-based framework demonstrates high predictive accuracy for early surgical intervention in NEC. Its integration into clinical workflows could support timely decision-making and improve neonatal outcomes