S541 Is Artificial Intelligence Cost-Effective in Colonoscopy? A Systematic Review and Meta-Analysis of Economic Evaluations and Modeling Studies

Introduction: Artificial intelligence (AI) is increasingly being used to enhance colonoscopy and improve adenoma detection rates (ADRs), potentially lowering colorectal cancer (CRC) incidence. However, its cost-effectiveness remains unclear. We conducted a systematic review and meta-analysis to assess whether AI-assisted colonoscopy is a cost-effective or cost-saving strategy across different healthcare systems. Methods: A systematic literature search was conducted across 6 databases: PubMed (MEDLINE), Google Scholar, ScienceDirect, PLOS ONE, Cochrane Library, and ClinicalTrials.gov. Studies comparing AI-assisted colonoscopy to standard colonoscopy were included. Outcomes of interest included incremental cost-effectiveness ratios (ICERs), net monetary benefit (NMB), and other modeled cost and effectiveness measures. Data were standardized to USD where applicable and pooled using a random-effects meta-analysis. Risk of bias was assessed using the Philips checklist for modeling studies. A forest plot and cost-effectiveness plane were constructed to visualize study results in relation to a $50,000/QALY willingness-to-pay threshold using RStudio (Version 2025.05.0+496). Results: Nine studies were included, 7 of which reported ICER or NMB values and were synthesized quantitatively. The pooled net monetary benefit was $1,300 (95% CI: $786–$1,814; P < 0.001), supporting the cost-effectiveness of AI-assisted colonoscopy. Heterogeneity was high (I² = 87.7%) due to variations in model design and healthcare settings. All studies fell within the cost-effective or dominant quadrants of the cost-effectiveness plane. Bustamante-Balén et al. and Barkun et al. showed dominant strategies, while Yu et al. and Hassan et al. reported favorable ICERs below $50,000/QALY. Two studies were assessed qualitatively: Mori et al. reported cost savings using a “diagnose-and-leave” strategy; Halvorsen et al. noted increased surveillance burden despite marginal cancer risk reduction. Conclusion: AI-assisted colonoscopy was consistently cost-effective, with several studies demonstrating dominant strategies. Despite substantial heterogeneity, all studies fell within the cost-effective or dominant regions of the cost-effectiveness plane. Integration into clinical practice should balance potential cost savings with the implications of increased surveillance burden. As current evidence is primarily based on modeling studies, cautious implementation and further validation in diverse real-world settings are warranted.