Colorectal Lesion Detection According to Daily Procedure Order with and Without Artificial Intelligence: An Exploratory Secondary Analysis of a Randomized Trial

Whether artificial intelligence (AI) modifies the association between procedure order and colorectal lesion detection remains uncertain. We performed an exploratory secondary analysis of a single-center randomized trial including 1009 patients undergoing AI-assisted (n = 481) or conventional colonoscopy (n = 528). Procedure order was grouped as positions 1–3, 4–6, and ≥7. The primary outcome was detection of at least one histologically confirmed adenoma or adenocarcinoma. Logistic regression evaluated the AI-by-order interaction before and after adjustment for patient characteristics and randomization strata. Primary outcome rates were 45.7%, 47.9%, and 49.5% with AI and 44.4%, 45.3%, and 50.5% without AI across the three respective categories. In the adjusted analysis (n = 1008), the odds ratio for positions ≥7 versus 1–3 was 1.00 (95% confidence interval 0.61–1.64) with AI and 1.06 (0.64–1.74) without AI (interaction p = 0.965). Restricting the outcome to conventional adenomas produced similar findings (interaction p = 0.959). No differential association between procedure order and detection was demonstrated. Wide confidence intervals, nonrandomized procedure order, and unavailable endoscopist and session identifiers limit inference.