From the Journals

Can AI conquer the late-shift dip in colonoscopy quality?



New research confirms that colonoscopies conducted later in an endoscopist’s shift are associated with a decline in adenoma detection and demonstrates that artificial intelligence (AI) can help eliminate the problem.

AI systems “may be a potential tool for minimizing time-related degradation of colonoscopy quality and further maintaining high quality and homogeneity of colonoscopies in high-workload centers,” Honggang Yu, MD, with the department of gastroenterology, Renmin Hospital of Wuhan (China) University, said in an interview.

The study was published online in JAMA Network Open.

Fatigue a factor?

Adenoma detection rate (ADR) is a critical quality measure of screening colonoscopy. Time of day is a well-known factor related to suboptimal ADR – with morning colonoscopies associated with improved ADR and afternoon colonoscopies with reduced ADR, Dr. Yu and colleagues write.

“However, an objective approach to solve this problem is still lacking,” Dr. Yu said. AI systems have been shown to improve the ADR, but the performance of AI during different times of the day remains unknown.

This cohort study is a secondary analysis of two prospective randomized controlled trials, in which a total of 1,780 consecutive patients were randomly allocated to either conventional colonoscopy or AI-assisted colonoscopy. The ADR for early and late colonoscopy sessions per half day were then compared.

Colonoscopy procedures were divided into two groups according to the end time of the procedure. The early group included procedures started in the early session per half day (8:00 a.m.–10:59 a.m. or 1:00 p.m.–2:59 p.m.). The late group included procedures started in the later session per half day (11:00 a.m.–12:59 p.m. or 3:00 p.m.–4:59 p.m.).

A total of 1,041 procedures were performed in the early sessions (357 conventional and 684 AI assisted). A total of 739 procedures were performed in the late sessions (263 conventional and 476 AI assisted).

In the unassisted colonoscopy group, later sessions per half day were associated with a decline in ADR (early vs. late, 13.73% vs. 5.7%; P = .005; odds ratio, 2.42; 95% confidence interval, 1.31-4.47).

With AI assistance, however, no such association was found in the ADR (early vs. late, 22.95% vs. 22.06%; P = .78; OR, 0.96; 95% CI, 0.71-1.29). AI provided the highest assistance capability in the last hour per half day.

The decline in ADR in late sessions (vs. early sessions) was evident in different colonoscopy settings. The investigators say accrual of endoscopist fatigue may be an independent factor of time-related degradation of colonoscopy quality.

More exploration required

“We’re excited about the great potential of using the power of AI to assist endoscopists in quality control or disease diagnosis in colonoscopy practice, but it’s too early to see AI as the standard,” Dr. Yu told this news organization.

“Despite recent achievements in the design and validation of AI systems, much more exploration is required in the clinical application of AI,” Dr. Yu said.

Dr. Yu further explained that, in addition to regulatory approval, the results of AI output must be trusted by the endoscopist, which remains a challenge for current AI systems that lack interpretability.

“Therefore, at the current stage of AI development, AI models can only serve as an extra reminder to assist endoscopists in colonoscopy,” Dr. Yu said.

This study was supported by the Innovation Team Project of Health Commission of Hubei Province. The authors have indicated no relevant financial relationships.

A version of this article originally appeared on

This article was updated 2/1/23.

Next Article:

First Humira biosimilar launches in U.S.