CADe Fails To Improve Polyp Detection in Community-Based Randomized Controlled Trial
Douglas K. Rex, MD, MASGE, reviewing Wei MT, et al. Am J Gastroenterol 2023 Mar 9.
The overwhelming majority of randomized trials testing the impact of computer-aided detection (CADe) programs have shown substantial benefits. In a randomized controlled trial in 4 community-based endoscopy centers in the U.S., involving 7 endoscopists with adenoma detection rates ranging from 25% to 37%, patients were randomized in blocks to colonoscopy with or without CADe. The CADe program was EndoVigilant (Millersville, Md).
There were 769 enrolled patients, of whom 72% had screening as the indication. The remainder were low-risk surveillance patients. The endoscopist had the choice of viewing the artificial intelligence input on the same versus a second monitor, and 4 chose the second monitor.
There was no difference in adenomas per colonoscopy with CADe versus non-CADe (0.73 vs 0.67) or sessile serrated lesions (SSLs) per colonoscopy (0.08 vs 0.08). The number of polyps detected per colonoscopy that were neither adenomas nor SSLs went up with CADe (0.90 vs 0.51), as did withdrawal time (11.7 vs 10.7 minutes).
The study had some strange results, including that two monitors were associated with a 3-fold higher find rate for polyps that were neither adenomas nor SSLs, a 4-fold increase in adenomas ≥10 mm in size, and more flat polyps.
COMMENTThis is the third recent U.S.-based study to show no benefit from CADe in polyp detection and the first to show an actual detriment, which was due to loss of specificity for precancerous polyps. This change from a series of positive randomized trials to negative results among community-based or nonspecialist academic endoscopists is challenging to understand – and concerning.
A potential black mark on the study was allowing the endoscopists to choose between one or two monitors. Also, the bizarre differences in detection results between the two groups of endoscopists suggest that important differences in colonoscopy behavior and bias between endoscopists contributed to some of the overall results.
Note to readers: At the time we reviewed this paper, its publisher noted that it was not in final form and that subsequent changes might be made.
CITATION(S)
Wei MT, Shankar U, Parvin R, et al. Evaluation of computer aided detection during colonoscopy in the community (AI-SEE): A multicenter randomized clinical trial. Am J Gastroenterol 2023 Mar 9. (Epub ahead of print) (https://doi.org/10.14309/ajg.0000000000002239)
