A Machine Learning-Based Clinical Tool for Predicting Inadequate Bowel Preparation: Development and Validation (Clinical and Translational Gastroenterology)

Plain Language SummaryThis study prospectively collected data from patients undergoing colonoscopy with anesthesia at Beijing Tiantan Hospital (November 2024–June 2025) to build a machine learning model that predicts inadequate bowel preparation using only nonpharmacological parameters. Using three feature selection methods and 5 machine learning algorithms, a Firth regression-based model performed best, with AUC values of 0.718 (95% CI: 0.647–0.789) in training and 0.715 (95% CI: 0.611–0.818) in validation. The resulting clinical prediction tool showed good discrimination (AUC, 0.709; 95% CI: 0.605–0.813). Higher body mass index, waist-to-hip ratio, lower gastrointestinal symptom score, diabetes, and smoking/alcohol score increased risk, whereas hematochezia decreased risk of inadequate bowel preparation.