The Machine Learning Model for Predicting Inadequate Bowel Preparation Before Colonoscopy: A Multicenter Prospective...

Area of Science:

• Gastroenterology
• Medical Informatics
• Predictive Analytics

Background:

• Adequate bowel preparation (BP) is crucial for colonoscopy effectiveness in diagnosing colorectal diseases.
• Inadequate BP remains a significant challenge, potentially impacting diagnostic accuracy and patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning model for predicting inadequate BP in Chinese adults undergoing colonoscopy.
• To identify key risk factors associated with inadequate BP.

Main Methods:

• A multicenter prospective study involving 3,217 adult outpatients.
• Utilized logistic regression and four machine learning models: support vector machines, decision trees, extreme gradient boosting, and bidirectional projection network.
• Collected data on patient characteristics, comorbidities, medication use, and BP quality.

Main Results:

• 21.14% of patients experienced inadequate BP.
• The decision trees model achieved the highest predictive performance (AUC=0.80 in the validation cohort).
• Identified risk factors include body mass index, education, simethicone use, diabetes, age, prior inadequate BP, and longer intervals.

Conclusions:

• The developed decision trees model effectively predicts patients at high risk for inadequate BP.
• Identified risk factors can guide personalized interventions, such as increased polyethylene glycol or auxiliary medication, to optimize BP.