Beyond the abdomen: an interpretable machine learning model for predicting postoperative ileus in non-abdominal...

Area of Science:

• Medical Informatics
• Surgical Outcomes
• Neurogastroenterology

Background:

• Postoperative ileus (POI) is an underestimated complication after non-abdominal surgery.
• The brain-gut axis, involving factors like depression and SSRI use, may influence POI development.

Purpose of the Study:

• To develop and validate a machine learning model for predicting POI risk.
• To integrate brain-gut axis variables into POI prediction.
• To assess the model's generalizability across different patient cohorts.

Main Methods:

• A multicenter retrospective study of 2000 patients undergoing non-abdominal surgery.
• Development of eight machine learning algorithms using a dual-algorithm feature selection strategy (LASSO and Boruta).
• Performance evaluation using Area Under the Curve (AUC), calibration plots, and Decision Curve Analysis.

Main Results:

• Seven predictors identified: chronic SSRI use, depression history, intraoperative opioids, surgery duration, neutrophil-to-lymphocyte ratio, serum albumin, and fluid balance.
• The Random Forest model achieved high AUCs (0.942 training, 0.917 internal, 0.895 external validation).
• The model significantly outperformed logistic regression and demonstrated excellent calibration and clinical utility.

Conclusions:

• The Random Forest model is a robust tool for predicting POI in non-abdominal surgery.
• Brain-gut axis factors significantly contribute to delayed bowel recovery.
• The model facilitates improved risk stratification and personalized perioperative management.