Artificial intelligence (AI) effectively reduced the decline in adenoma detection rates (ADRs) observed with colonoscopy procedures performed in late sessions per half day. These findings appear in JAMA Network Open.
Colonoscopy performed in the morning is associated with improved ADR. Research has shown that the rate of polyp detection rate (PDR) declines as the day progresses. This could have serious ramifications, as ADR/PDR is a key quality measure of screening colonoscopy, and effective screening colonoscopies significantly reduce the incidence of colorectal cancer.
Researchers sought to determine if an AI system could assist in overcoming this time-related decline in ADR during colonoscopy.
The team performed a secondary analysis of 2 randomized clinical trials (RCTs) from a university hospital in China. In the first RCT, consecutive patients who were undergoing colonoscopy were randomly assigned to an AI-assisted group (355 patients) or an unassisted (control) group (349 patients) between June 18, 2019 and September 6, 2019.
In the second RCT, during the period from July 1, 2020, to October 15, 2020, 1075 patients were randomly assigned into 4 groups: control group; CADe (a deep-learning based computer-aided detection system used to augment detection of polyps inside the visual field) group; CAQ (a computer-aided quality improvement system used to improve withdrawal visualization quality) group; and CADe plus CAQ group (271, 268, 269, and 268 patients, respectively).
For this study, the researchers combined all the intervention groups into an AI group, and combined the 2 control groups into a single group (control-c).
All colonoscopy procedures were divided into early and late sessions per half day, according to the end time of the procedure (early, 8:00 am to 10:59 am or 1:00 pm to 2:59 pm; late, 11:00 am to 12:59 pm or 3:00 pm to 4:59 pm).
Overall, 1041 of 1780 procedures (58.48%) were performed in the early sessions, with 357 (34.29%) in the control-c group and 684 (65.71%) in the AI group. The remaining 739 procedures (41.52%) were performed in late sessions, with 263 (35.59%) in the control-c group and 476 (64.41%) in the AI group.
A decline in adenoma detection was associated with the late colonoscopy sessions per half day (early vs late, 13.73% vs 5.70%; P =.005). However, this association was not observed with the assistance of AI systems (early vs late, 22.95% vs 22.06%; P =.78), suggesting that their use could overcome time-related declines in ADR. This could be particularly useful in large-workload clinical practices where reducing the number of colonoscopies or shortening shifts to eliminate the decline in ADR is not feasible.
“In conclusion, our results suggest that later sessions per half day were associated with a decline adenoma detection. Furthermore, AI systems could eliminate the time-related degradation of colonoscopy quality,” the researchers wrote. AI systems have the potential to maintain high quality and homogeneity of colonoscopy procedures, as well as further improve endoscopist performance in screening programs and centers with high caseloads, they concluded.
The researchers acknowledged some limitations. The result was only validated at 1 center, and the generalizability of the conclusion is uncertain in clinical centers that have different work schedules. They also noted that the ADR in the control-c group was lower than in western populations, and suggested that future studies include multiple regional studies to account for differences in diet, lifestyle, and environment.
Lu Z, Zhang L, Yao L, et al. Assessment of the role of artificial intelligence in the association between time of day and colonoscopy quality. JAMA Netw Open. 2023;6(1):e2253840. doi:10.1001/jamanetworkopen.2022.53840