P2342 - Adenoma and Polyp Detection Rates by Implementing Artificial Intelligence Systems: A Meta-analysis

Hassam Ali, MD, Rahul Pamarthy, MD, Karissa Lambert, MD, Kara Regan, MD
East Carolina University, Greenville, NC

Introduction: Artificial intelligence (AI) using specially designed computer aided detection systems have been introduced for to improve adenoma and polyp detection and colonoscopy effectiveness in recent years.

Methods: We performed a meta-analysis of recent trials to determine the benefits of AI-based adenoma and polyp detection systems to detect colorectal cancer. Our databases for the search included PubMed, EMBASE, Cochrane, and Clinicaltrials.gov. Studies were screened for eligibility, the certainty of the evidence, and the risk of bias. We compared polyp and adenoma detection rates based on colonoscopy with and without AI based computer aided detection systems. We also report any changes associated with excellent bowel preparation based on the Boston bowel preparation scale (BBPS ≥7) and polyp/adenoma detection rates.

Results: Our study included five randomized trials irrespective of Boston bowel preparation scores (N = 4311). Additionally, a sub-analysis was done of two trials based on BBPS ≥7 (N = 1153). Colonoscopy with AI systems increased polyp detection rates (PDRs) compared to standard colonoscopy. PDR was 45.76 % with use of AI versus 30.72 % without AI; Odds ratio (OR) 1.91, confidence interval (CI) 1.68-2.16, p < 0.01. Adenoma detection rates (ADRs) were 30.71 % using artificial intelligence systems compared to 20.46% without AI; OR 1.77, CI 1.47-2.14, p < 0.01.
In a sub-analysis of two trials, with BBPS ≥7, polyp detection rates were 40.73 % using AI versus 28.91% without AI; OR 1.69, CI 1.32-2.16, p < 0.001. The individual ADR in one of the studies was not statistically significant based on BBPS ≥7. However, on meta-analysis, adenoma detection rates were 26.75% in artificial intelligence use versus 21.17% without AI; OR 1.36, CI 1.04-1.78, p < 0.03, showing statistical significance.

Discussion: Artificial intelligence-based adenoma and polyp detection practices during colonoscopy increases detection of adenomas and polyps. Further comparative data is required to implement a definitive artificial intelligence system. The cost effectiveness is yet to be investigated.


Figure: Forest plots for polyp and adenoma detection rates using Artificial intelligence.
a: Polyp detection rates in regular bowel preparation, b: Adenoma detection rates in regular bowel preparation, c: Polyp detection rates in excellent bowel preparation, d: Adenoma detection rates in excellent bowel preparation

Trial	Weightage	Odd ratios for polyp detection rates	Confidence intervals	Weightage	Odds ratios for adenoma detection rates	Confidence intervals
Wang P, et al. 2019	24.3%	1.99	1.55-2.57	23.57%	1.61	1.21-2.14
Wang P, et al. 2020	23.6%	1.86	1.44-2.41	24.27%	1.36	1.03-1.78
Gong D, et al. 2020	15.5%	1.76	1.28-2.42	11.21%	2.25	1.37-3.69
Liu WN, et al. 2020	23.2%	2.01	1.55-2.61	24.79%	2.05	1.56-2.68
Su JR, et al. 2020	13.4%	1.82	1.30-2.57	16.16%	2.06	1.40-3.02

Table: Prominent trials and their individual statistics for polyp and adenoma detection rates using Artificial intelligence based computer aided detection systems


Disclosures:
Hassam Ali indicated no relevant financial relationships.
Rahul Pamarthy indicated no relevant financial relationships.
Karissa Lambert indicated no relevant financial relationships.
Kara Regan indicated no relevant financial relationships.


Hassam Ali, MD, Rahul Pamarthy, MD, Karissa Lambert, MD, Kara Regan, MD. P2342 - Adenoma and Polyp Detection Rates by Implementing Artificial Intelligence Systems: A Meta-analysis, ACG 2021 Annual Scientific Meeting Abstracts. Las Vegas, Nevada: American College of Gastroenterology.