Risk stratification of postoperative enteral feeding intolerance using explainable machine learning in oral cancer...

Enteral nutrition (EN) is essential after free flap reconstruction for oral cancer; however, feeding intolerance (FI) frequently limits adequate nutritional delivery. Existing prediction tools are primarily derived from general intensive care unit populations and may not adequately reflect the unique metabolic and inflammatory vulnerabilities of this surgical cohort. We aimed to develop and internally validate an interpretable machine learning model for the early prediction of postoperative FI.

In this single-center retrospective study, 752 patients undergoing radical resection with free flap reconstruction (between 2017 and 2025) were included. FI was defined according to established consensus criteria. A total of 35 perioperative variables were evaluated. Feature selection was performed using least absolute shrinkage and selection operator (LASSO) regression. Seven machine learning algorithms were trained and validated on an independent hold-out set. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), calibration, and decision curve analysis. SHapley Additive exPlanations (SHAP) were used to interpret the model.

Postoperative FI occurred in 36.04% of patients. LASSO identified nine predictors. The random forest model demonstrated the best balance between discrimination and stability, achieving an AUC of 0.889 (95% CI: 0.847-0.925) in the validation cohort, with good calibration and consistent clinical net benefit. SHAP analysis identified elevated fasting glucose, advanced tumor stage, low serum potassium, longer operative time, and a lower Advanced Lung Cancer Inflammation Index (ALI) as the most influential factors, with nonlinear threshold effects observed for key metabolic variables.

This interpretable model enables accurate perioperative risk stratification for FI using routinely available clinical data. The findings highlight the role of metabolic-inflammatory imbalance and support risk-adapted, individualized nutritional management in patients undergoing oral cancer reconstruction. External validation is warranted.