Development and evaluation of a machine learning-based risk prediction model for enteral feeding intolerance in sepsis patients.

Background: Early detection and prediction of enteral feeding intolerance (EFI) are essential for effective management of septic patients. This study seeks to develop an interpretable machine learning (ML) model for predicting EFI in septic patients. Methods: Data were collected from septic patients admitted to the intensive care unit and receiving enteral nutrition (EN) at a tertiary hospital in Shandong Province between January 2023 and July 2025. A retrospective cohort was randomly divided into training and validation sets in a 7:3 ratio. Feature selection was performed using univariate analysis and binary logistic regression. 5 independent ML models were developed and evaluated based on the area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, specificity, and F1 score. The Shapley Additive Explanation (SHAP) method was applied to interpret the predictive model. Results: The study included 549 septic patients, with an EFI incidence of 34.6%. 6 key features were selected for model development: age, APACHE II score, albumin levels, receipt of continuous renal replacement therapy, EN start time, and intra-abdominal pressure. Among the models, the RF model demonstrated the best performance, with an AUROC of 0.891, accuracy of 0.830, F1 score of 0.771, specificity of 0.869, and sensitivity of 0.763. SHAP analysis identified albumin levels as a protective factor for EFI in septic patients. Conclusion: This model can serve as a tool to identify high-risk individuals with EFI among septic patients, facilitating clinical healthcare providers in delivering scientific and individualized EN therapy to patients.