, 2025-04-16 13:37:00
An American Gastroenterological Association (AGA) multidisciplinary panel has reached the conclusion that no recommendation can be made for or against the use of computer-aided detection (CADe)–assisted colonoscopy for colorectal cancer (CRC), the third most common cause of cancer mortality in the United States.
The systematic data review is a collaboration between the AGA and The BMJ’s MAGIC Rapid Recommendations. The BMJ issued a separate recommendation against CADe shortly after the AGA guideline was published.
Led by Shahnaz S. Sultan, MD, MHSc, of the Division of Gastroenterology, Hepatology, and Nutrition at University of Minnesota, Minneapolis, and recently published in Gastroenterology, found only very low certainty of GRADE-based evidence for several critical long-term outcomes, both desirable and undesirable. These included the following: 11 fewer CRCs per 10,000 individuals and two fewer CRC deaths per 10,000 individuals, an increased burden of more intensive surveillance colonoscopies (635 more per 10,000 individuals), and cost and resource implications.
This technology did, however, yield an 8% (95% CI, 6-10) absolute increase in the adenoma detection rate (ADR) and a 2% (95% CI, 0-4) increase in the detection rate of advanced adenomas and/or sessile serrated lesions. “How this translates into a reduction in CRC incidence or death is where we were uncertain,” Sultan said. “Our best effort at trying to translate the ADR and other endoscopy outcomes to CRC incidence and CRC death relied on the modeling study, which included a lot of assumptions, which also contributed to our overall lower certainty.”
The systematic and meta-analysis included 41 randomized controlled trials with more than 32,108 participants who underwent CADe-assisted colonoscopy. This technology was associated with a higher polyp detection rate than standard colonoscopy: 56.1% vs 47.9% (relative risk [RR], 1.22, 95% CI, 1.15-1.28). It also had a higher ADR: 44.8% vs 37.4% (RR, 1.22; 95% CI, 1.16-1.29).
But although CADe-assisted colonoscopy may increase ADR, it carries a risk for overdiagnosis, as most polyps detected during colonoscopy are diminutive (< 5 mm) and of low malignant potential, the panel noted. Approximately 25% of lesions are missed at colonoscopy. More than 15 million colonoscopies are performed annually in the United States, but studies have demonstrated variable quality of colonoscopies across key quality indicators.
“Artificial intelligence [AI] is revolutionizing medicine and healthcare in the field of GI [gastroenterology], and CADe in colonoscopy has been brought to commercialization,” Sultan told Medscape Medical News. “Unlike many areas of endoscopic research where we often have a finite number of clinical trial data, CADe-assisted colonoscopy intervention has been studied in over 44 randomized controlled trials and numerous nonrandomized, real-world studies. The question of whether or not to adopt this intervention at a health system or practice level is an important question that was prioritized to be addressed as guidance was needed.”
Commenting on the guideline but not involved in its formulation, Larry S. Kim, MD, MBA, a gastroenterologist in Denver, said his practice group has used the GI Genius AI system in its affiliated hospitals but has so far chosen not to implement the technology at its endoscopy centers. “At the hospital, our physicians have the ability to utilize the system for select patients or not at all,” he told Medscape Medical News.
The fact that The BMJ reached a different conclusion based on the same data, evidence-grading system, and microsimulation, Kim added, “highlights the point that when evidence for benefit is uncertain, underlying values are critical.” In declining to make a recommendation, the AGA panel balanced the benefit of improved detection of potentially precancerous adenomas vs increased resource utilization in the face of unclear benefit. “With different priorities, other bodies could reasonably decide to recommend either for or against CADe.”
The Future
According to Sultan, gastroenterologists need a better understanding of patient values and preferences and the value placed on increased adenoma detection, which may also lead to more lifetime colonoscopies without reducing the risk for CRC. “We need better intermediate- and long-term data on the impact of adenoma detection on interval cancers and CRC incidence,” she said. “We need data on detection of polyps that are more clinically significant such as those 6-10 mm in size, as well as serrated sessile lesions. We also need to understand at the population or health system level what the impact is on resources, cost, and access.”
Ultimately, the living guideline underscores the trade-off between desirable and undesirable effects and the limitations of current evidence to support a recommendation, but CADe has to improve as an iterative AI application with further validation and better training.
With the anticipated improvement in software accuracy as AI machine learning reads increasing numbers of images, Sultan added, “the next version of the software may perform better, especially for polyps that are more clinically significant or for flat sessile serrated polyps, which are harder to detect. We plan to revisit the question in the next year or two and potentially revise the guideline.”
These guidelines were fully funded by the AGA Institute with no funding from any outside agency or industry.
Sultan is supported by the US Food and Drug Administration. Co-authors Shazia Mehmood Siddique, Dennis L. Shung, and Benjamin Lebwohl are supported by grants from the National Institute of Diabetes and Digestive and Kidney Diseases. Theodore R. Levin is supported by the Permanente Medical Group Delivery Science and Applied Research Program. Cesare Hassan is a consultant for Fujifilm and Olympus. Peter S. Liang reported doing research work for Freenome and advisory board work for Guardant Health and Natera.
Kim is president-elect of the AGA. He disclosed no competing interests relevant to his comments.