PRACTICE (AI optical diagnosis)
2025
Safety of artificial intelligence-assisted optical diagnosis for leaving colorectal polyps in situ during colonoscopy (PRACTICE): a non-inferiority, randomised controlled trial
In 895 adults undergoing colonoscopy with computer-aided detection and diagnosis, leaving optically non-neoplastic rectosigmoid polyps ≤5 mm in situ was non-inferior to resecting all detected polyps for adenoma detection rate.
Result
- Absolute difference
- 1.8 points
absolute difference-1.895% CI -8.9 to 4.9 · P = 0.013
The absolute difference is derived from the event rates.
Also reported: 197 vs 211 adenomas or cancers detected; no perforation or bleeding in either group.
- Subtopic
- Optical diagnosis & deskilling
- Design
- Multicentre open-label non-inferiority RCT, Italy
- Relevance
- Supporting: context rather than must-know.
- Source
- Antonelli G, et al. Lancet Gastroenterol Hepatol. 2025;10(10):915-923.doi 10.1016/S2468-1253(25)00140-2
Discussion & critique
First randomised test of the guideline-endorsed leave-in-situ strategy, here with AI-supported optical diagnosis: oncological safety, measured as ADR, was preserved while unnecessary polypectomies were avoided. Open-label, four Italian centres, and ADR is a surrogate: the trial cannot show that polyps left behind never progress. Randomised, so it sits in the cluster as a safety trial rather than a detection trial.