Testing Machine Learning Models to Predict Postoperative Ileus after Colorectal Surgery.

Area of Science:

• Colorectal Surgery
• Surgical Complications
• Machine Learning in Medicine

Background:

• Postoperative ileus (POI) is a frequent complication following colorectal surgery.
• POI significantly increases hospital stay duration and associated healthcare expenses.
• Identifying predictive factors for POI is crucial for optimizing patient care.

Purpose of the Study:

• To investigate patient comorbidities associated with POI development in colorectal surgery.
• To evaluate the accuracy of machine learning (ML) models in predicting POI.
• To compare ML model performance against established risk assessment tools.

Main Methods:

• Retrospective analysis of 316 adult colorectal surgery patients (Jan 2020-Dec 2021).
• Exclusion criteria included complex resections and inadequate follow-up.
• Eight ML models were trained and validated using 29 comorbidities and 4 risk indices.

Main Results:

• The incidence of POI was 6.33%.
• Significant predictors of POI included age, BMI, gender, kidney disease, anemia, arrhythmia, rheumatoid arthritis, and NSQIP score.
• AdaBoost (94.2%) and XG Boost (85.2%) demonstrated the highest predictive accuracy.

Conclusions:

• ML models show high accuracy in predicting POI risk after colorectal surgery.
• These models represent a potential advancement for early POI detection and intervention.
• Improved POI prediction can enhance patient outcomes and reduce healthcare costs.