Abstract

Background: Adenoma Detection Rate (ADR) is the principal quality indicator for colonoscopy and is inversely associated with interval colorectal cancer. Computer-Aided Detection (CADe) systems raise ADR in randomised trials and are being adopted rapidly. Whether that gain falls on the lesions that drive cancer risk has attracted less scrutiny than the headline figure. Methods: Systematic reviews with meta-analysis of randomised trials comparing CADeassisted with standard colonoscopy and reporting ADR were eligible. Reviewlevel risk ratios were pooled on the log scale in a DerSimonian-Laird randomeffects model, with Cochran’s Q, I², τ² and 95% prediction intervals. Secondary outcomes—adenoma miss rate, sessile serrated lesion detection, polyp detection and advanced adenoma detection—were extracted and compared. The pooled risk ratio was translated to absolute detection rates across reported baselines. Analyses were performed in Python 3. Results: Three reviews contributing ADR estimates and representing 61 randomised trials in 53,433 patients were pooled. CADe increased adenoma detection: pooled risk ratio 1.23 (95% CI 1.19-1.27; z = 11.22, p < 0.001), with no detectable heterogeneity (Q = 1.34, df = 2, p = 0.512; I² = 0.0%) and a 95% prediction interval of 1.14-1.33 that excluded the null. At the reported baselines of 33.0% and 35.9%, this corresponds to absolute detection rates of 40.5% and 44.1%, gains of 7.5 and 8.2% points, and 12 to 13 colonoscopies needed to yield one additional patient with a detected adenoma. Adenoma miss rate was approximately halved (pooled risk ratio 0.45, 95% CI 0.39-0.52; I² = 0.0%) and sessile serrated lesion detection increased (1.27, 1.11-1.47). Advanced adenoma detection was unchanged (1.01, 0.90-1.13). Conclusions: CADe reliably increases adenoma detection and roughly halves the adenoma miss rate, and the effect is homogeneous across reviews. The benefit is nonetheless concentrated in lesions of limited clinical consequence: advanced adenoma detection, the outcome most closely tied to cancer prevention, is unchanged. Whether CADe reduces interval cancer, rather than raising a surrogate, is not established by this evidence and requires outcome trials.

Keywords: Colonoscopy, Computer-aided detection, Artificial intelligence, Adenoma detection rate, Advanced adenoma, Colorectal cancer screening, Second-order meta-analysis