A machine learning model to improve bowel preparation quality monitoring

A natural language processing (NLP) model successfully identified the quality of bowel preparation from free-text colonoscopy reports with high accuracy. This may enable large-scale, reproducible monitoring of colonoscopy quality and support the automated calculation of the adenoma detection rate (ADR) and data-driven quality benchmarking across centres.

“ADR is a cornerstone quality indicator in colonoscopies, and bowel preparation quality directly affects colonoscopy performance and ADR,” explained Dr Besim Agargun (University of Istanbul, Turkey) [1]. “NLP of free-text colonoscopy reports could enable automated extraction of colonoscopy quality indicators.” The retrospective study aimed to develop a rule-based NLP model to assess bowel preparation quality. In total, 11,374 colonoscopy reports were analysed, of which 1,018 were used for training purposes. The model incorporated parameters such as caecal intubation, polyp presence, size, and location.

The model achieved a cross-validated accuracy of 92%. The precision and recall were 0.98/0.99 for good, 0.85/0.82 for intermediate, and 0.85/0.87 for poor bowel preparation. The predicted distribution was 50.9% good, 19.1% intermediate, and 30.0% poor bowel preparations. “The model showed robust and reproducible performance in real-world free-text reports,” said Dr Agargun. Older age, male sex, successful caecal intubation, and better bowel cleanliness were all associated with an increased ADR (see Figure).

Figure: Factors associated with adenoma detection rate [1]



“This study demonstrates that NLP-based automated classification of bowel preparation quality is both feasible and valuable for generating colonoscopy datasets to support quality monitoring,” concluded Dr Agargun. “By assessing bowel cleanliness alongside other ADR-related factors, such as age, sex, and caecal intubation, benchmarks for ADR can be established and applied to quality control and improvement in endoscopy centres.”

1. Agargun BF, et al. NLP-based classification of bowel preparation quality and its association with ADR in colonoscopy reports. LB20, Latest News: From top to bottom, UEG Week, 4–7 October 2025, Berlin, Germany.

Posted on 1 December 20251 December 2025