TrialsgrambetaRCTs are beautiful

Wallace CADe (tandem)

2022

Impact of Artificial Intelligence on Miss Rate of Colorectal Neoplasia

In 230 patients undergoing screening or surveillance colonoscopy, AI-assisted colonoscopy performed first reduced the adenoma miss rate at same-day tandem colonoscopy compared with standard colonoscopy performed first.

Result

Efficacy: Adenoma miss rate (AMR)adjusted OR 0.38 (95% CI 0.23-0.62)

Below 1 favours AI-assisted colonoscopy performed first, above 1 favours standard colonoscopy performed first.

Event rates
Absolute difference
16.9 points
NNT
6

Absolute difference and NNT are derived from the event rates.

Also reported: Effect largest for lesions ≤5 mm (15.9% vs 35.8%; OR 0.34) and nonpolypoid lesions (16.8% vs 45.8%; OR 0.24). False-negative subjects 6.8% vs 29.6% (OR 0.17, 95% CI 0.05–0.67); no difference in adverse events.

Original publication
Subtopic
Missed adenomas
Design
International multicentre randomised tandem-colonoscopy trial (Italy, UK, USA)
Relevance
Important: one of several pillars.
Source
Wallace MB, et al. Gastroenterology. 2022;163(1):295-304.e5.doi 10.1053/j.gastro.2022.03.007

Discussion & critique

The cleanest demonstration that CADe corrects perceptual error: a roughly two-fold fall in miss rate across three countries, concentrated in small and flat lesions. Tandem design measures misses, not clinical outcomes; 230 patients; endoscopists could not be masked. Read together with Glissen Brown it explains where the ADR gains of the parallel-group trials come from.