Download PDF
CC BY-NC-ND 4.0 · Endosc Int Open 2024; 12(04): E598-E599
DOI: 10.1055/a-2295-2177
Letter to the editor

Computer-aided detection for colorectal neoplasia in randomized and non-randomized studies

Yuichi Mori
1   Clinical Effectiveness Research Group, University of Oslo, Oslo, Norway (Ringgold ID: RIN6305)
2   Digestive Disease Center, Showa University Northern Yokohama Hospital, Yokohama, Japan (Ringgold ID: RIN220878)
,
Harsh K Patel
3   Gastroenterology and Hepatology, University of Kansas Medical Center, Kansas City, United States
,
Alessandro Repici
4   Department of Biomedical Sciences, Humanitas University, Milan, Italy (Ringgold ID: RIN437807)
5   Endoscopy Unit, Humanitas Clinicial and Research Center - IRCCS, Milan, Italy
,
Douglas K. Rex
6   Division of Gastroenterology, Indiana University School of Medicine, Indianapolis, United States (Ringgold ID: RIN12250)
,
Prateek Sharma
3   Gastroenterology and Hepatology, University of Kansas Medical Center, Kansas City, United States
7   Gastroenterology and Hepatology, Kansas City VA Medical Center and University of Kansas School of Medicine, Kansas City, United States
,
Cesare Hassan 
4   Department of Biomedical Sciences, Humanitas University, Milan, Italy (Ringgold ID: RIN437807)
5   Endoscopy Unit, Humanitas Clinicial and Research Center - IRCCS, Milan, Italy
› Author Affiliations
› Further Information
Also available at
   Permissions and Reprints

The widespread adoption of artificial intelligence (AI) for polyp detection in colonoscopy necessitates a thorough understanding of its benefits and harms. While AI has shown promise in increasing the adenoma detection rate (ADR), potentially reducing colorectal cancer [1], it also raises concerns about increased removal of non-neoplastic polyps. However, conflicting results from recent meta-analyses [2] [3] about the effectiveness of AI in colonoscopy have caused confusion in clinical practice.

One meta-analysis of 21 randomized controlled trials (RCTs) involving 18,232 patients demonstrated that AI-assisted colonoscopy increased the ADR (risk ratio 1.24; 95% confidence interval 1.16–1.33) but also led to a higher rate of non-neoplastic polyp removal compared with standard colonoscopy (mean difference of 0.18 polyps per colonoscopy [0.11–0.26]) [2]. Conversely, another meta-analysis of eight non-randomized studies with 9,687 patients failed to find significant changes in these benefit and harm outcomes, respectively [3].

The debate over the reliability of study designs further complicates the issue. While non-blinded RCTs are generally considered the most trustworthy evidence, they may suffer from artificially controlled environments and unconscious bias favoring the intervention (e.g., the Hawthorne effect) [4]. On the other hand, non-randomized observational studies reflect real-world scenarios but are susceptible to selection bias and lack of adequately controlled groups.

Theoretically, pragmatic RCTs such as randomized health services studies could be the optimal way to measure the real-world effectiveness of medical interventions in which study subjects are less monitored but evenly controlled as compared with traditional randomized trials [5]. Given that we do not have robust results based on such study designs now, a comprehensive consideration of the benefit-harm balance of AI in colonoscopy is needed. Of particular importance is involving patients, physicians, academic societies, and policymakers in evaluating the use of AI to ensure patient-centered care.



Publication History

Received: 26 February 2024

Accepted: 20 March 2024

Article published online:
23 April 2024

© 2024. The Author(s). This is an open access article published by Thieme under the terms of the Creative Commons Attribution-NonDerivative-NonCommercial-License, permitting copying and reproduction so long as the original work is given appropriate credit. Contents may not be used for commercial purposes, or adapted, remixed, transformed or built upon. (https://creativecommons.org/licenses/by-nc-nd/4.0/).

Georg Thieme Verlag KG
Rüdigerstraße 14, 70469 Stuttgart, Germany

 

  • References

  • 1 Areia M, Mori Y, Correale L. et al. Cost-effectiveness of artificial intelligence for screening colonoscopy: a modelling study. Lancet Digital Health 2022; 4: e436-e444 DOI: 10.1016/S2589-7500(22)00042-5. (PMID: 35430151)
    CrossrefPubMedGoogle Scholar
  • 2 Hassan C, Spadaccini M, Mori Y. et al. Real-time computer-aided detection of colorectal neoplasia during colonoscopy: A systematic review and meta-analysis. Ann Intern Med 2023; 176: 1209-1220
    CrossrefPubMedGoogle Scholar
  • 3 Patel HK, Mori Y, Hassan C. et al. Lack of effectiveness of computer aided detection for colorectal neoplasia: A systematic review and meta-analysis of nonrandomized studies. Clin Gastroenterol Hepatol 2023; 23: 1542-3565
    PubMedGoogle Scholar
  • 4 Sedgwick P, Greenwood N. Understanding the Hawthorne effect. BMJ (clinical research ed) 2015; 351: h4672 DOI: 10.1136/bmj.h4672. (PMID: 26341898)
    CrossrefPubMedGoogle Scholar
  • 5 Gamerman V, Cai T, Elsäßer A. Pragmatic randomized clinical trials: best practices and statistical guidance. Health Serv Outcomes Res Method 2019; 19: 23-35
    CrossrefPubMedGoogle Scholar