AI-Assisted Colorectal Lesion Detection Falls Short In Private Practice Trial

Key Takeaways

• The prespecified adenoma detection endpoint was similar with and without AI assistance in this randomized private-practice comparison.
• No significant difference was reported for polyp detection rate or for withdrawal time between the two study groups.
• The trial involved experienced examiners in a non-academic outpatient setting, and the authors noted factors that may have limited detection of a smaller effect.

In a multicenter randomized trial in German outpatient private-practice colonoscopy centers, real-time AI assistance with EndoMind did not significantly improve adenoma detection rate compared with traditional colonoscopy.

This prospective randomized controlled study ran from November 2021 through November 2022 in Germany. Six centers enrolled patients, but one was excluded, leaving five outpatient treating centers in the final analysis. Investigators randomized 933 participants and included 914 in the intention-to-treat population, with 452 assigned to computer-aided colonoscopy and 462 to traditional colonoscopy. More than 94% of examinations were screening or post-polypectomy surveillance procedures, and eligible adults also included those aged 50 years or older with positive fecal immunochemical testing. Ten examiners participated, each with more than 10 years of experience and more than 10,000 prior colonoscopies.

Adenoma detection rate was 34.5% with computer-aided colonoscopy and 32.9% with traditional colonoscopy (p=0.656). Secondary analyses found no significant difference in overall polyp detection rate or withdrawal time. Median withdrawal time was 7.5 minutes with computer-aided colonoscopy and 7.6 minutes with traditional colonoscopy, while corrected withdrawal time was 6.6 and 6.5 minutes, respectively. Use of the AI system was not associated with a measurable timing advantage.

The investigators interpreted the findings in the context of a non-academic private-practice environment staffed by experienced examiners. They noted that the control group achieved a higher-than-anticipated adenoma detection rate, which may have left the study underpowered to detect a smaller effect size.

A subgroup analysis suggested that lower-detector examiners may have been more affected by computer-aided detection, although that pattern was not significant. The findings reflect experienced outpatient private-practice colonoscopy rather than all colonoscopy settings.