Using an AI device to aid colonoscopy significantly increases adenoma detection rates irrespective of endoscopist skill level, a large US study has shown.
The quality improvement study of over 334,000 colonoscopies published in Gastroenterology [link here] is one of the first real-world studies to corroborate evidence on the benefit of computer-aided detection demonstrated in clinical trials.
Researchers assessed colonoscopies from 269 endoscopists at 42 randomly selected US Veterans Health Administration facilities, which had been equipped with the AI detection devices, and 547 endoscopists at 97 control facilities.
The device used was the GI Genius Intelligent Endoscopy Module from Medtronic, which was approved by the US Food and Drug Administration in 2021 as the first AI system for computer-aided detection of colorectal neoplasia.
Over the study period, 71,594 and 35,399 colonoscopies were performed before and after deployment of the AI devices at device-equipped facilities, and 151,792 and 75,415 colonoscopies, respectively, were performed at control sites.
Findings showed availability of the devices was associated with a significant 22% increased odds of adenoma detection (odds ratio 1.22; 95% CI 1.15-1.28), but was not associated with detection of other polyps such as sessile serrated lesions.
Improvement in the the adenoma detection rate (ADR) was even greater for older patients (OR 1.02 per 5 years, 95% CI 1.00-1.03, p=0.02) and was not moderated by endoscopist specialty, sex, or years since medical degree (p >0.05 across all).
Comparing post- to pre-deployment, there was a slight increase of 4% in the absolute proportion of colonoscopies where adenomas were detected at device facilities (from 50.7% to 54.9%; 95% confidence interval (CI) 3.48-4.74), and a decrease of 0.7% at control sites (from 51.8% to 51.1%; 95% CI -1.16 to -0.28).
The benefit of the computer-aided detection devices was apparent for all endoscopists, including those with a pre-deployment ADR of <40%, with 55% of those who fell into this category at device-equipped facilities improving their ADR to ≥40% post deployment, compared with 20% of those at control facilities.
A survey of endoscopists in the AI-assisted group revealed their overall user experience was generally positive, with most respondents (77.4%) either moderately or strongly supportive of computer-aided detection for colonoscopy.
While 40% of respondents reported no perceived impact on ADR, 46% attributed the device to slight improvement and 2.4% thought there was a marked increase.
However, half of respondents reported the devices led to too many false positives and some expressed concern that AI adoption could lead to an over-reliance on technology and a lack of training for new endoscopists on polyp detection.
The concerns over deskilling and training were echoed in an accompanying editorial [link here], which suggested while the devices might improve colonoscopy quality, by serving as a “second pair of eyes” to reduce human error, it might also provide a crutch for trainees that could result in and hide skill deficiency until removed.
“This begs the question of whether or when CADe (computer-aided detection) should be made available during training as the balance of risk and benefit is uncertain,” the authors said.
Key questions also remained over the impact on long-term cancer outcomes, which was not investigated in the study, and whether the devices were better suited for selective or widespread use, considering their cost, they said.
They said a large cluster-randomised trial currently underway in Europe would provide insights on the value of the devices for CRC incidence and mortality.