Interpretable machine learning model for predicting refeeding syndrome after colorectal cancer surgery.

Area of Science:

• Oncology
• Medical Informatics
• Nutritional Science

Background:

• Refeeding syndrome (RFS) is a critical complication in postoperative nutritional support for colorectal cancer (CRC) patients.
• Early and accurate prediction of RFS is essential for effective patient management.

Purpose of the Study:

• To develop an explainable machine learning (ML) model for early risk prediction of RFS in CRC surgery patients.
• To evaluate the predictive performance and clinical utility of the developed ML model.

Main Methods:

• Retrospective analysis of 446 CRC surgery patients.
• Development and comparison of four ML models: logistic regression, random forest, SVM, and XGBoost.
• XGBoost model interpretation using SHapley Additive Explanations (SHAP) for feature importance and individual predictions.

Main Results:

• XGBoost model achieved the highest predictive performance (AUC = 0.872) in the validation set.
• Key predictors identified: preoperative serum phosphate, time to bowel sound recovery, preoperative albumin, and time to first flatus.
• SHAP analysis provided global and individual-level insights into RFS risk factors.

Conclusions:

• The explainable XGBoost-SHAP model offers accurate and interpretable prediction of postoperative RFS risk.
• This approach can guide individualized nutritional management strategies for CRC surgery patients.