Pragmatic real-world evaluation of computer-aided detection in colonoscopy: A within-endoscopist comparative study

Computer-aided detection (CADe) improves adenoma detection rate (ADR) in randomised trials; however, translation into routine practice remains uncertain, with real-world studies showing inconsistent results.

To evaluate CADe in a pragmatic multicentre setting using a within-endoscopist design, with endoscopists serving as their own comparators under routine clinical conditions.

Pragmatic multicentre observational study analysing prospectively recorded data from 13 units. Six endoscopists performed procedures in a unit with routine CADe use and in units without CADe, enabling within-operator comparisons under naturalistic conditions. The primary outcome was ADR. Secondary outcomes included combined adenoma plus clinically significant serrated polyp detection rate (A+CSSPDR) and additional polyp-related metrics.

Of 3161 colonoscopies, 2973 were analysed (947 CADe; 2026 non-CADe). ADR was higher with CADe (34.4%vs 30.7%; p = 0.042). A+CSSPDR was also higher (40.3%vs 35.8%; p = 0.018), with consistent improvements across other metrics. In multivariable analysis, CADe remained independently associated with higher ADR (aOR 1.31, 95%CI 1.11-1.55; p = 0.002) and A+CSSPDR (aOR 1.33, 95%CI 1.13-1.56; p = 0.001), with attenuation after adjustment for endoscopist.

In a pragmatic real-world setting, CADe improved detection of colorectal neoplasia. This within-endoscopist design provides a balanced estimate of effectiveness by minimising behavioural bias affecting both trial and real-world studies.