Introduction: Several randomized and real-world studies and meta-analyses have evaluated the benefits and harms of computer aided detection (CADe) in colonoscopy among experienced endoscopists. However, the same for trainee endoscopists remains largely unknown. We performed a meta-analysis to compare benefits and harms of CADe for endoscopists-in-training.

Methods: We searched MEDLINE, EMBASE, Cochrane and Google Scholar from inception to November 29, 2023. We included randomized and non-randomized studies that compared CADe-assisted and Standard colonoscopy (SC) among trainee endoscopists. Two investigators independently extracted study data and assessed quality. Benefits were assessed using adenoma detection rate (ADR), mean adenoma per colonoscopy (APC) and advanced adenoma detection rate (AADR). Harms were assessed using withdrawal time and mean non-neoplastic lesion per colonoscopy (NNLPC). Comparisons were performed between trainees with AI (CADe-T), trainees using standard colonoscopy without AI (SC-T) and expert endoscopists using standard colonoscopy without AI (SC-E). Pooled proportions were calculated using OpenMeta[analyst] software. Pairwise meta-analysis was performed utilizing risk ratio (RR) for dichotomous variables and risk difference (RD) or mean difference (MD) for continuous variables with a 95% confidence interval (95% CI) using RevMan 5.4.1.

Results: We identified a total of 5 studies including 2382 patients (CADe-T: 933; SC-T: 903; SC-E: 546) that were included in the meta-analysis. Trainees using AI had a trend towards higher ADR (47.1%) compared to trainees using SC (41.7%), RR: 1.15 [0.99 - 1.34] based on 4 studies, Image 1a). However, there was no differences in APC (RD: 0.23 adenoma [-0.09 - 0.56] based on 4 studies), or AADR (8% vs 9.3% patients; RR: 0.87 [0.63 – 0.121]. AI for polyp detection increased the PDR by 6.5%. Interestingly, there was no significant increase in harms: NNLPC (RD: 0.21 polypectomy [-0.18 – 0.61] based on 3 studies, Image 1b) and withdrawal time (MD 0.42 mins [-6.4 - 7.3], based on 3 studies). Additionally, as shown in Table 1, there was no difference in either of those outcomes when the following groups were compared: CADe-T vs SC-T; CADe- vs SC-E.

Conclusion: Based on this meta-analysis of nearly 2400 patients, we found a trend towards higher ADR by incorporating AI for polyp detection in colonoscopies performed by endoscopists-in-training, without an increase in burden of colonoscopy. Future studies with larger sample size are needed to support these findings.

Image 1: Trainee with AI vs Trainee without AI. 1a. Adenoma Detection Rate (ADR); 1b. Non-Neoplastic Lesion Per Colonoscopy (NNLPC)

Table 1. Benefits and Harms of Incorporating AI During Colonoscopy for Trainees

Sachin Srinivasan

Appearances