鶹ý

AI-Aided Colonoscopy Flops for Advanced Neoplasia Detection in Trial

— Meta-analysis turns up similar findings, though increased ADR overall

MedpageToday
A computer rendering of polyps in the colon.

Incorporating artificial intelligence (AI) computer-aided detection systems (CADe) into colonoscopy does not improve rates of advanced neoplasia detection, according to two studies in Annals of Internal Medicine.

In the randomized trial of patients with a positive fecal immunochemical test, detection of advanced colorectal neoplasia was virtually identical with or without the use of CADe during colonoscopy, at 34.8% versus 34.6%, respectively (rate ratio 1.01, 95% CI 0.92-1.10), reported Rodrigo Jover, MD, PhD, of the Hospital General Universitario de Alicante in Spain, and colleagues.

Meanwhile, a from Marco Spadaccini, MD, of Humanitas University in Rozzano, Italy, and co-authors showed that use of CADe increased the overall adenoma detection rate (ADR), but not when it came to finding advanced adenomas. And those that received CADe-assisted colonoscopy had more unnecessary removals of non-neoplastic polyps:

  • ADR: 44% with CADe vs 35.9% without (relative risk [RR] 1.24, 95% CI 1.16-1.33)
  • Advanced adenomas detected per colonoscopy: 0.111 vs 0.107 (P=0.54)
  • Non-neoplastic polyps removed per colonoscopy: mean 0.52 vs 0.34 (P<0.001)

"This tempers enthusiasm for CADe but does not negate the clear performance benefit for detecting adenomas of all sizes," said Dennis Shung, MD, PhD, of Yale University in New Haven, Connecticut, writing in an accompanying . He suggested that an overall improved ADR with widespread use of CADe could result in lengthened recommended surveillance intervals.

"However, before this and other computer algorithms are integrated into routine clinical care, we must focus on value and embrace the difficulty of defining the value and best use," wrote Shung.

CADILLAC

The randomized CADILLAC trial was conducted in six Spanish centers participating in population-based colorectal cancer screening programs. It included individuals (mean age 61 years, 53.4% male) who presented for colonoscopy after a positive fecal immunochemical test from April 2021 to March 2022.

Secondary outcomes, including the detection rates of advanced adenomas, advanced serrated lesions, colorectal cancers, and adenomas did not significantly differ between those who received or did not receive CADe:

  • Advanced adenomas: 30.5% vs 31.3%, respectively
  • Advanced serrated lesions: 6.5% vs 5.3%
  • Colorectal cancers: 3.7% vs 3.2%
  • Adenomas: 64.2% vs 62%

CADe did increase overall detection rates of polyps (73.4% vs 70.1%) and serrated lesions (21.3% vs 17.1%), as well as the mean number of nonpolypoid lesions, proximal adenomas, and lesions of 5 mm or smaller detected per colonoscopy.

"Artificial intelligence applications are in a dynamic phase," observed Jover's group. "The findings emphasize the need to train new versions of deep-learning models with larger data sets of advanced nonpolypoid lesions to locate these frequently difficult-to-detect lesions that potentially are the primary source of colonoscopy miss rates."

The researchers pointed out that previous studies evaluating colonoscopy with CAD have been inadequately powered to show significant differences in the detection of advanced lesions, whereas the current trial enrolled adequate numbers to detect potential differences, and also relied on a population with the highest prevalence of advanced neoplasias.

"This setting offered the best context for investigating the ability of computer-aided detection to support the diagnosis of advanced colorectal neoplasias," they said.

A study limitation, according to the team, was that the "high adenoma detection rate in the control group may limit the generalizability of the findings to endoscopists with low detection rates."

Meta-Analysis

For their meta-analysis and systemic review, Spadaccini and colleagues evaluated 21 randomized trials involving a total of 18,232 patients published from 2019 to 2023; the analysis for ADR was based on all 21 studies (low-certainty evidence), advanced adenomas on the basis of 15 studies (low-certainty evidence), and the analysis of non-neoplastic polyp removal was based on 12 studies (low-certainty evidence).

They said their findings "showed a consistent relative effect of CADe in improving the ADR by almost 25%, corresponding to a number needed to scope of around 13.5 to detect one additional adenoma... Such an ADR increase seemed to be driven by a 55% decrease in the miss rate of adenomas at per polyp analysis."

Spadaccini's group noted that "the clinical relevance of our analysis depends on the assumption that ADR is an important outcome of screening colonoscopy," adding that it may be considered a surrogate for the prevention of colorectal cancer.

But they added that the possibility of harm associated with unnecessary resections "deserves further consideration."

A study limitation was that there were differences across trials in how advanced adenomas were defined.

  • author['full_name']

    Mike Bassett is a staff writer focusing on oncology and hematology. He is based in Massachusetts.

Disclosures

The CADILLAC study was by supported by Medtronic, the Instituto de Salud Carlos III, and the Asociación para la Investigación en Gastroenterología de la Provincia de Alicante (AIGPA) in Spain.

Jover disclosed an institutional research grant from Medtronic. A co-author disclosed multiple relationships with industry.

Spadaccini disclosed no relationships with industry. Co-authors disclosed multiple relationships with industry.

Shung disclosed support from the NIH and the American Gastroenterological Association.

Primary Source

Annals of Internal Medicine

Mangas-Sanjuan C, et al "Role of artificial intelligence in colonoscopy detection of advanced neoplasias" Ann Intern Med 2023; DOI: 10.7326/M22-3678.

Secondary Source

Annals of Internal Medicine

Hassan C, et al "Real-time computer-aided detection of colorectal neoplasia during colonoscopy: A systematic review and meta-analysis" Ann Intern Med 2023; DOI: 10.7326/M22-3678.

Additional Source

Annals of Internal Medicine

Shung D "From tool to team member: A second set of eyes for polyp detection" Ann Intern Med 2023; DOI: 10.7326/M23-2022.