AI-assisted colonoscopy increases adenoma detection rate

Carol A. Burke, , MD, AGAF, FACG, FASGE, FACP

Quality matters and is particularly important in the performance of colonoscopy, whereupon quality and post-colonoscopy colorectal cancer (PCCRC) occurrence are inversely linked. Over the past 15 years, quality metrics have been established in order to enhance the protective benefit of colonoscopy. The most well-established metric is the adenoma detection rate (ADR), although a recent publication determined the detection rate of clinically significant serrated polyps is also associated with PCCRC.

Unfortunately, there is wide variability in colorectal ADR among endoscopists, and PCCRC is more likely to occur with endoscopists who have lower ADR. Emerging data have shown that the use of artificial intelligence to aid in the detection of polyps during colonoscopy is associated with a two-fold increase in ADR over standard colonoscopy. In the multicenter study from Wallace and colleagues, 230 patients scheduled for screening or surveillance colonoscopies were randomized into two groups to receive back-to-back procedures performed by the same endoscopist: the first done with the use of a commercially available, computer-aided polyp detection system (CAD) or standard colonoscopy, followed by a second colonoscopy with the other technology first. Investigators involved in the trial had performed a minimum of 1,000 colonoscopies and had historic moderate ADRs between 20% and 40%. No trial endoscopist singly contributed to performing examinations in more than 90 of the randomized patients. Withdrawal time was at least 6 minutes, exclusive of polyp resection, and less than 6 mm hyperplastic-appearing polyps in the rectum and sigmoid colon to 25 cm were not included in the analysis. Most patients in the study were white men undergoing surveillance colonoscopy.

This well-designed study confirmed the consistency of results from other trials showing a benefit for computer-aided polyp detection system (CAD) compared with standard colonoscopy. In this trial, the authors demonstrated a 50% lower ADR and a four-fold lower false negative rate in the CAD-first group than the standard colonoscopy-first group. The major driver of their statistically significant results was a lower miss rate of diminutive (less than 6 mm) adenomas but not 6-9 mm or greater than 10 mm adenomas. In addition, larger lesions are the most impactful lesions to detect. Diminutive adenomas comprised about 80% of polyps detected in each group.

How do we position these results in terms of quality of colonoscopy and benefit for our patients? I suggest one approach is by considering the features of adenomas (number, size and pathology) which portend risk for future advanced neoplasia and affect post-polypectomy surveillance intervals. Strong evidence suggests patients with one to two tubular adenomas less than 10-mm in size — and particularly patients with diminutive adenomas — are a low-risk group for metachronous advanced neoplasia and enjoy a protective benefit from CRC for at least 10 years after a high-quality colonoscopy. 

The U.S. Multi-Society Task Force on CRC guidelines now suggest patients with one to two small, tubular adenomas undergo their next colonoscopy in 7 to 10 years. The guidelines also stratify patients into a post-polypectomy interval of 3 years vs. 3 to 5 years for those with 5 to 10 vs. 3 to 4 less than 10-mm tubular adenomas, respectively. In this study, investigators found a statistically significant reduction in the mean number of adenomas detected on the second colonoscopy in the CAD vs standard colonoscopy first group (0.33 vs. 0.70). Based upon their data, no difference in change of surveillance intervals was afforded by the CAD system. As more data is obtained from future trials of CAD on polyp detection, the true benefit of this innovation on metachronous advanced neoplasia and its cost effectiveness will likely be ascertained.