P1780 - Comparing Adenoma Detection Rate (ADR) in Single vs Dual Observer vs Artificial Intelligence-Assisted Colonoscopy: A Network Analysis of Randomized Controlled Trials

Manesh Kumar Gangwani, MD¹, Hossein Haghbin, MD², Rizwan Ishtiaq, MD³, Dushyant Singh. Dahiya, MD⁴, Umar Hayat, MD⁵, Wade M.. Lee-Smith, MLS¹, Anooja Rani, ⁶, Yusuf Nawras, ¹, Muhammad Aziz, MD¹, Toseef Javaid, MD¹, Benjamin Hart, MD, PhD^7
1University of Toledo, Toledo, OH; ²Ascension Providence Hospital, Southfield, MI; ³Academic Hospitalist, St. Francis Hospital, Hartford, CT; ⁴University of Kansas School of Medicine, Kansas City, KS; ⁵Geisinger Wyoming Valley Medical Center, Wilkes-Barre, PA; ⁶Dow University of Health Sciences, Karachi, Sindh, Pakistan; ⁷University of Michigan, Ann Arbor, MI

Introduction: The adenoma detection rate (ADR) is a crucial quality measure for screening colonoscopy. Recent randomized controlled trials have shown the effectiveness of Artificial Intelligence (AI) and second observers in improving ADR. However, there is a lack of comparative analysis considering all three modalities to identify the optimal approach. Therefore, we conducted a network meta-analysis of randomized controlled trials (RCTs) to compare the ADR detection rates among single operators, second observers, and AI.

Methods: MEDLINE, Embase, Web of Science Core Collection, Cochrane Central Register of Controlled Trials, SciELO, the Korean Citation Index, and Global Index Medicus were systematically searched from inception through April 2023. Network (direct + indirect) meta-analysis was performed using a random effects model and a frequentist approach. The risk ratio (RR) with a 95% confidence interval (CI) was calculated.

Results: A total of 1,119 articles were screened, and eventually, 11 RCTs with 9,333 participants were included in the analysis.

In the network meta-analysis, no statistical difference was found between the second observer and AI-assisted colonoscopy, with RR 0.88 (95% CI: 0.75-1.04, p = 0.01, I2 = 0). However, a significant statistical difference was observed between the single observer and AI-assisted colonoscopy, favoring the AI-assisted group, with an RR of 0.70 (95% CI: 0.62-0.80, p = 0.01, I2 = 47%). Similarly, a significant statistical difference was noted between the second observer group and the single observer group, with an RR of 1.26 (95% CI: 1.10-1.44, I2 = 27%, p = 0.01). The model exhibited a heterogeneity of 58.7%. The direct and indirect models demonstrated consistency, indicating high precision and low study bias.

On net ranking, the AI group had the highest P-value, followed by the second observer and single observer, indicating a higher probability of detection with a higher P value. Although the analysis appeared to favor AI over the second observer, no statistical significance was reached.

Discussion: The use of artificial intelligence (AI) and a second observer demonstrates superiority in ADR compared to a single observer. However, to establish a more conclusive comparison between AI and the second observer, further randomized controlled trials (RCTs) with direct head-to-head comparisons are required.



Disclosures:


Manesh Kumar Gangwani, MD¹, Hossein Haghbin, MD², Rizwan Ishtiaq, MD³, Dushyant Singh. Dahiya, MD⁴, Umar Hayat, MD⁵, Wade M.. Lee-Smith, MLS¹, Anooja Rani, ⁶, Yusuf Nawras, ¹, Muhammad Aziz, MD¹, Toseef Javaid, MD¹, Benjamin Hart, MD, PhD⁷. P1780 - Comparing Adenoma Detection Rate (ADR) in Single vs Dual Observer vs Artificial Intelligence-Assisted Colonoscopy: A Network Analysis of Randomized Controlled Trials, ACG 2023 Annual Scientific Meeting Abstracts. Vancouver, BC, Canada: American College of Gastroenterology.