An Artificial Intelligence-Guided Strategy to Reduce Poor Bowel Preparation: A Multicenter Randomized Controlled Study

INTRODUCTION: Proper bowel preparation is crucial for increasing the adenoma detection rate. A novel application based on the use of a convolutional neural network can differentiate between adequate and inadequate bowel preparation on the basis of images of rectal effluents taken before the examination. The aim of this study was to evaluate whether a software-driven approach improves colon cleansing quality during colonoscopies compared with standard care. METHODS: A multicenter randomized controlled trial was conducted. Consecutive patients were assigned to a standard-care group or an intervention group. The latter group was trained to use a web application linked to a convolutional neural network, enabling them to send an image of their most recent rectal effluent and receive feedback on the adequacy of their bowel preparation. Patients were instructed to follow the recommendations provided by the platform. RESULTS: Overall, 774 patients were eligible and randomized . The intention-to-treat analysis revealed statistically significant differences in bowel cleansing quality in favor of the intervention group (91% vs 84.2%, OR 1.88, 95% CI [1.21-2.93], P = 0.005). The right and left colon exhibited better cleansing in the intervention group (90.4% vs 84.8%, P = 0.016 and 95.3% vs 91.5%, P = 0.03, respectively). In the per-protocol analysis, bowel cleansing quality was also significantly higher in the intervention group, both overall (93.3% vs 85.6%, OR 2.34 (1.36-4.02), P = 0.002) and by segment. When aiming for excellent bowel preparation (>7), cleansing was significantly better in the intervention group overall and by segment. DISCUSSION: A software application-driven colon cleansing process improves preparation quality in outpatients (NCT05871814).