Single-blinded randomized controlled trials (RCTs) have shown that computer-aided detection (CADe) increases adenoma detection rate (ADR), whereas real-world implementation studies have reported no gains. Single-blinded design and awareness of being studied may inflate effect estimates while implementation studies might suffer from selection bias. We evaluated CADe in routine practice within a concealed trial in which patients, endoscopists, and clinical staff were unaware of study participation and randomization.

A concealed two-arm multicenter randomized controlled trial was conducted between July 2023 and August 2025 across three centers, enrolling patients aged 45–89 years undergoing diagnostic, screening, or surveillance colonoscopy. Patients were randomized 1:1 to an endoscopy room equipped with a CADe system (intervention) or a room without CADe (control). Endoscopists, nursing staff and patients were blinded to the existence of the trial. A research assistant did activate CADe every morning before arrival of the endoscopy team, however, endoscopists could choose to deactivate CADe. The primary outcome was ADR; secondary outcomes included CADe utilization rate, adenomas per colonoscopy (APC), sessile serrated lesion detection rate (SSLDR), advanced lesion detection rate (ALDR), proximal APC (pAPC), and the proportions of neoplastic and non-neoplastic polyps. The primary analysis was intention-to-treat (ITT), with per-protocol (PP) analyses based on actual CADe use. Generalized linear models estimated group differences.

Of 1,744 randomized patients, 1,673 were included in the analysis (control n=837; intervention n=836). Cross-over occurred in 89 (10.6%) control and 71 (8.5%) intervention patients. In ITT analysis, ADR did not differ significantly between groups groups (control: 31.5%, 95% confidence interval, CI: 28.4–34.8 vs intervention: 34.4%, 31.2–37.8, p=0.225). In PP analysis, ADR was higher when CADe was used (40.6%, 36.3–45.1 vs 34.7%, 30.1–38.6; p=0.045). When assigned to a CADe-equipped room, endoscopists used the system in 57.5% of eligible colonoscopies (481/836; ITT) and 92.9% (471/507) in the PP population. Among PP patients, ALDR, APC, and pAPC were significantly higher with CADe (e.g., ALDR 8.3%, 6.1–11.1 vs 3.9%, 2.5–5.8; p=0.002), whereas SSLDR and the proportion of non-neoplastic polyps did not differ.

In this concealed trial, randomization to a CADe-equipped room did not significantly increase ADR in ITT analysis, but colonoscopies in which CADe was used showed higher ADR and advanced lesion detection. These findings suggest that incomplete and selective adoption in routine care can attenuate the benefits observed in RCTs. Strategies to ensure consistent CADe use are likely required for CADe to meaningfully improve ADR outside of controlled study conditions.