- Design
- systematic review and frequentist network meta-analysis of randomised trials, GRADE-assessed, searched to 25 July 2026
- Population
- 13 randomised trials, 4,156 adults undergoing colonoscopy, 8 computer-aided detection systems versus standard colonoscopy
- Primary outcome
- mean polyp detection counts stratified by size (5 mm or less, 6-9 mm, 10 mm or more)
- Effect
- SMD 0.21 (95% CI 0.07-0.35, 95% PI -1.12 to 1.54) for 5 mm or less; 0.02 (-0.02 to 0.06) for 6-9 mm; 0.01 (0.00-0.02) for 10 mm or more
This network meta-analysis pooled 13 randomised trials and 4,156 participants comparing eight computer-aided detection systems against standard colonoscopy, with polyp counts stratified by size. No trial compared one system against another, so every cross-platform ranking is indirect.
For diminutive polyps of 5 mm or less, artificial intelligence gave a modest advantage: standardised mean difference 0.21 (95% CI 0.07 to 0.35). For polyps of 6 to 9 mm the difference was 0.02 (95% CI -0.02 to 0.06), and for those of 10 mm or more it was 0.01 (95% CI 0.00 to 0.02). Heterogeneity for the diminutive category was high at I-squared 86.6%, and the 95% prediction interval ran from -1.12 to 1.54, crossing the null comfortably. GRADE certainty was very low for diminutive polyps and low for the others.
This matters because the case for buying these systems has rested on aggregate adenoma detection rates that do not separate by size. If the entire gain is in lesions of 5 mm or less, the clinical benefit is much harder to argue: diminutive polyps rarely harbour cancer, and finding more of them shortens surveillance intervals, adds polypectomies and adds histopathology cost without a demonstrated effect on colorectal cancer incidence.
Nothing here says the technology does not work. It says the benefit is concentrated where it matters least and that the evidence is weak, and it is honest enough to label its own conclusions hypothesis-generating. Before committing capital, ask what the system is expected to do for the detection of lesions of 10 mm or more, and note that on this evidence the answer is almost nothing.
- Ask any vendor for size-stratified detection data, not an aggregate adenoma detection rate
- Expect more diminutive polyps, and plan for the extra polypectomy and histopathology workload that follows
- Do not expect computer-aided detection to improve the finding of lesions 10 mm or larger on this evidence
- Withdrawal technique, adequate bowel preparation and withdrawal time remain the interventions with the strongest evidence behind them
- Read the platform rankings as indirect: no trial in this network compared one system against another
The statistics, in plain English
A standardised mean difference of 0.21 is a small effect by convention, and the prediction interval running from -1.12 to 1.54 means that in a new colonoscopy unit the effect could plausibly go either way, which is a far more useful figure than the confidence interval here. An I-squared of 86.6% says the trials disagree with each other substantially, so the pooled estimate is an average across genuinely different results rather than a precise summary. For the larger polyp sizes the intervals are so tight around zero that the finding is not uncertainty but a confident absence of meaningful effect.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for gastroenterology & hepatology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free