P2751 - Comparing Artificial Intelligence-Assisted and Clinician-Performed Endoscopy in the Diagnosis of Chronic Atrophic Gastritis: A Diagnostic Accuracy Meta-Analysis

Award: Presidential Poster Award

Muhammed Elfaituri, MBBCh, Taha Khaled, , Hazem Faraj, MBBCh, Ahmed BenGhatnsh, MBBCh, Ahmed Msherghi, MBBCh
University of Tripoli, Tripoli, Tripoli, Libya

Introduction: Chronic Atrophic Gastritis (CAG) can progress to serious conditions like gastric cancer if not promptly identified. Traditional clinician-performed endoscopy, while standard, has variable accuracy due to individual interpretation. Artificial intelligence (AI) technologies present a potential avenue for improving diagnostic precision. This study aimed to evaluate the diagnostic performance of AI-assisted endoscopy compared to clinician-performed endoscopy for CAG.

Methods: Databases, including PubMed, Embase, and Cochrane Central, were searched to find relevant studies published up until January 2023. These studies compared the diagnostic performance of AI-assisted endoscopy and traditional clinician-performed endoscopy for CAG, using pathology-specific outcomes as the reference standard. A random effects univariate model was employed to pool diagnostic parameters, with outcomes presented with a 95% confidence interval. Statistical analyses were conducted with R software (version 4.0.3) and the mada package, pooling diagnostic parameters using a random effect univariate model. All outcomes are reported with a 95% confidence interval.

Results: Four studies involving 1922 participants (1374 confirmed cases by pathology and 548 negative specimens) met the inclusion criteria. For clinician-performed endoscopy, the pooled sensitivity was 37.7% (95% CI: 25.4% to 51.9%, I²=96.9%), specificity 65.1% (95% CI: 61.1% to 69%, I²=17.3%), false-positive rate 34.9% (95% CI: 31% to 38.9%), diagnostic odds ratio 1.13 (95% CI: 0.62 to 2.07), and positive and negative likelihood ratios (LR) were 1.08 (95% CI: 0.74 to 1.58) and 0.96 (95% CI: 0.76 to 1.19) respectively. In contrast, AI-assisted endoscopy demonstrated a higher sensitivity of 43.7% (95% CI: 35.9% to 51.7%, I²=91.5%), but a lower specificity of 35.3% (95% CI: 7.9% to 77.7%, I²=87.6%), a higher false-positive rate of 64.7% (95% CI: 22.3% to 92.1%), a lower diagnostic odds ratio of 0.42 (95% CI: 0.06 to 2.78), and positive and negative LR of 0.67 (95% CI: 0.34 to 1.33) and 1.59 (95% CI: 0.48 to 5.34) respectively.

Discussion: AI-assisted endoscopy, while demonstrating higher sensitivity, had lower specificity and a higher false-positive rate for CAG diagnosis compared to clinician-performed endoscopy. Despite its potential, it requires further refinement and standardization of AI algorithms and endoscopic techniques to increase diagnostic precision.


Disclosures:


Muhammed Elfaituri, MBBCh, Taha Khaled, , Hazem Faraj, MBBCh, Ahmed BenGhatnsh, MBBCh, Ahmed Msherghi, MBBCh. P2751 - Comparing Artificial Intelligence-Assisted and Clinician-Performed Endoscopy in the Diagnosis of Chronic Atrophic Gastritis: A Diagnostic Accuracy Meta-Analysis, ACG 2023 Annual Scientific Meeting Abstracts. Vancouver, BC, Canada: American College of Gastroenterology.