A Machine Learning-Based Clinical Tool for Predicting Inadequate Bowel Preparation: Development and Validation.

Area of Science:

• Gastroenterology
• Medical Informatics
• Predictive Analytics

Background:

• Colonoscopy is crucial for diagnosing colorectal diseases.
• Effective bowel preparation is essential for colonoscopy quality.
• Predicting inadequate preparation can optimize patient outcomes.

Purpose of the Study:

• Develop and validate a machine learning model to predict inadequate bowel preparation.
• Utilize non-pharmacological parameters for risk prediction.
• Create a clinical risk assessment tool for inadequate bowel preparation.

Main Methods:

• Prospective data collection from colonoscopy patients.
• Bowel preparation quality assessed using the Boston Bowel Preparation Scale.
• Machine learning algorithms and feature selection methods applied for model development.

Main Results:

• Identified six significant risk factors for inadequate bowel preparation.
• Firth regression model achieved AUC of 0.718 (training) and 0.715 (validation).
• Clinical prediction tool demonstrated good discrimination and calibration in validation.

Conclusions:

• Body mass index, waist-to-hip ratio, GI symptoms, diabetes, and smoking/alcohol intake are key risk factors.
• Hematochezia was found to have a protective effect.
• The Firth regression model and risk tool effectively identify patients at risk for inadequate bowel preparation.