Withdrawal time (WT) is a key quality indicator and technical determinant of colonoscopy performance. The integration of artificial intelligence (AI), particularly computer-aided detection (CADe) systems, has demonstrated potential to enhance adenoma detection rate (ADR). However, the interaction between the CADe and procedural dynamics, including its potential to inadvertently prolong WT, remains poorly understood. In this study, we sought to evaluate the independent and combined effects of CADe and WT on ADR in a real-world multicentre setting.

Data were analysed from the NAIAD trial, a prospective, stepped-wedge, multicentre study conducted across 25 hospitals involving 4,514 colonoscopies. The study comprised three phases: Phase 1 (baseline), Phase 2 (CADe active), and Phase 3 (post-intervention). Changes in WT were examined across phases. Mixed-effects logistic regression was employed to assess the independent contributions of CADe (Phase 2 vs Phases 1+3) and WT (per minute) to ADR, adjusting for patient age, endoscopist expertise and gender, bowel preparation quality, and hospital type, with random effects for site and endoscopist.

Implementation of CADe (Phase 2) was associated with increased procedural duration and improved detection. Mean WT increased from 10.2 ± 5.5 minutes at baseline to 10.9 ± 5.7 minutes during AI-assisted colonoscopy, accompanied by an ADR increase from 30.0% to 34.8% (p < 0.001). In multivariable mixed-effect regression analysis, both CADe use and WT were independent predictors of ADR. CADe use increased the odds of adenoma detection by 17% (aOR 1.17; 95% CI, 1.01–1.35; p = 0.040), while each additional minute of WT increased ADR by 11% (aOR 1.11; 95% CI, 1.09–1.13; p < 0.001). Following CADe withdrawal (Phase 3), mean WT and ADR declined to 8.9 minutes and 27.6%, respectively, below baseline levels.

CADe implementation and withdrawal time independently and synergistically improve adenoma detection. Although AI use incidentally prolongs WT, its impact on ADR extends beyond this effect. The decline in WT and ADR after CADe withdrawal suggests a transient behavioural adaptation. While CADe can augment endoscopic performance, sustained meticulous withdrawal technique remains fundamental to high-quality colonoscopy.