AGA stuck in neutral on AI-assisted colonoscopy guidance; ‘close tradeoff’ of cost/benefit
Key takeaways:
- The AGA made no recommendation for or against the use of AI-driven polyp detection in colonoscopy.
- Insufficient evidence to determine if polyp detection via AI prevented colorectal cancer.
The AGA reported that it could not issue a recommendation for or against the use of computer-aided detection in colonoscopy due to “very low certainty of evidence” and “the close tradeoff between the desirable and undesirable effects.”
The association cautioned that, although AI is credited with increased detection of low-risk colorectal polyps during colonoscopy, there remains considerable uncertainty on how many of the identified polyps would have developed into colorectal cancer.
“In recent years, there has been an exponential rise in studies assessing the impact of artificial intelligence-based technology with real-time computer-aided detection (CADe) of polyps,” Shahnaz Sultan, MD, MHSc, AGAF, from the division of gastroenterology, hepatology and nutrition at the University of Minnesota, and colleagues wrote in Gastroenterology. “Although CADe-assisted colonoscopy may increase [adenoma detection rate (ADR)], there is a risk of overdiagnosis, as most of the polyps detected during colonoscopy are diminutive in size and of low malignant potential.”
The authors warned that: “As detection rates increase, the likelihood of both screening effectiveness and overdiagnosis increases. Thus, improving colonoscopy quality by means of increasing ADR will also increase the proportion of colonoscopy recipients who are put into a more intensive surveillance algorithm, leading to increased costs, risks, resource use and burden associated with surveillance colonoscopy.”
To determine whether sufficient evidence exists to back a recommendation for CADe-assisted colonoscopy, a multidisciplinary panel of experts used the Grading of Recommendations, Assessment, Development and Evaluation methodology to evaluate potential risk-benefit of this approach, longer-term patient-relevant outcomes, evidence for patient preferences/values and gastroenterology providers’ level of trust in AI technology.
Although the panel initially proposed a conditional recommendation in favor of CADe-assisted colonoscopy, following considerable feedback during the public comment period, they opted instead to make no recommendation at this time.
“The panel agreed that the increase in ADR is an important benefit because it is an established quality indicator,” Sultan and colleagues wrote. “The ADR increase, however, was primarily attributable to detection of diminutive to small polyps with low potential for progression to adenocarcinoma. This results in surveillance colonoscopies at shorter intervals and a greater lifetime number of colonoscopies, most of which are not individually likely to prevent CRC.”
Instead, the panel prioritized the long-term CRC outcomes, considering the “close tradeoff between the uncertain and trivial-to-small benefits on reducing CRC incidence and CRC-related mortality” weighed against likelihood of possibly unnecessary surveillance colonoscopies, the risk for overdiagnosis and greater burden on health care resources.
In their systematic review and meta-analysis of 44 randomized controlled trials with more than 30,000 participants who received CADe-assisted colonoscopy, the panel found that CADe could be responsible for reducing CRC incidence by 11 fewer cancers per 10,000 individuals over the span of a decade, with “a number needed to treat of 909 (but we are very unsure).”
The panel noted that the evidence for CADe reducing CRC-related mortality was even less apparent, with data suggesting “there may be a trivial or no benefit” with only two fewer CRC-related deaths per 10,000 individuals over 10 years. In contrast, CADe was found to have “little to no impact on adverse events from colonoscopy,” including perforation and bleeding, and no meaningful improvement in withdrawal time.
Based on the data, the panel estimated that CADe would likely boost the rate of non-neoplastic polyp detection and increase the number of colonoscopies performed per patient over a 10-year period. They projected that CADe implementation would enlist an additional 635 individuals per 10,000 to surveillance colonoscopy for follow-up on detected adenomas, “especially diminutive or small adenomas, which are of low malignant potential, but contribute to a higher number of polyps, resulting in more frequent colonoscopy.”
Although the panel recognized the significant potential for CADe in early CRC detection, it urged that additional studies on long-term patient outcomes are needed before it could recommend widespread adoption. Additionally, there are ongoing concerns for resource requirements, equitable access to colonoscopy and cost-effectiveness of CADe implementation that need to be addressed before a recommendation can be made.
The AGA noted that it will revisit this topic within 1 to 2 years once additional data become available.
Collapse