Effect of a Computer-Aided Device on Endoscopists' Biopsy Practice for Gastric Neoplasms: A Secondary Analysis of a Randomized Controlled Trial.
Hongliu Du, Shuzhe Tan, Ziyi Zeng et al.
✦ AI-curated · Sources linked
In 30 seconds
This secondary analysis of a randomized controlled trial compared AI-assisted with standard gastric screening, examining biopsy practice for documented lesions. Within the AI group, lesions marked as high risk with red boxes were biopsied more often than lesions without red boxes or lesions in the control group. Neoplasm biopsy efficiency was 8.19% in the red-box group versus 0.36% in the non-red-box group and 1.77% in the control group (both p < 0.001). Red-box classification was associated with more targeted sampling; false-positive alerts were not associated with indiscriminate biopsies.
Key findings
- Non-biopsy rates were 18.00% for red-box lesions, 43.26% for non-red-box lesions, and 45.60% in controls (both comparisons p < 0.001).
- Neoplasm biopsy efficiency was 8.19% in the red-box group versus 0.36% in the non-red-box group and 1.77% in controls (both p < 0.001).
- Red-box classification was associated with a lower non-biopsy rate (IRR 0.393, 95% CI 0.360-0.429, p < 0.001).
- False-positive alerts had a weak inverse correlation with biopsy counts (ρ = -0.182, p < 0.001).
Why it matters
Biopsy decisions affect whether gastric neoplasms are sampled and how efficiently endoscopy resources are used. These findings suggest that AI risk prompts may help direct biopsies toward lesions classified as high risk, without evidence in this analysis that false-positive prompts led to indiscriminate sampling. The results inform evaluation of AI-assisted endoscopic workflows, while not establishing that the prompts improve patient outcomes.
What to keep in mind
This was a secondary analysis of a randomized trial, and comparisons between red-box and non-red-box lesions within the AI group were not themselves randomized.