S1881 External Validation of the AutoIBD-UC Score in Assessing Severity of Ulcerative Colitis: A Single Center Analysis

Introduction: Over the last several years, artificial intelligence (AI) models have been studied and utilized with increasing frequency in various medical applications including diagnosis and evaluation of inflammatory bowel disease (IBD). One proprietary AI model, the AutoIBD-Ulcerative Colitis (UC) algorithm developed by Virgo Surgical Video Solutions (Virgo) (San Diego, CA), was designed for moderate-severe UC patient identification for clinical trial recruitment but had not previously been validated for assessing disease severity in patients with UC. We performed a single-center analysis of colonoscopy videos utilizing the AutoIBD-UC algorithm to evaluate its external validity in assessing disease severity with real-world data. Methods: We performed a single-center retrospective analysis of patients who underwent colonoscopy between August 2023 and March 2025. A total of 60 patients with UC in various stages of severity (i.e., Mayo 0 to 3) undergoing diagnostic or surveillance colonoscopy and 60 patients without IBD (control group) undergoing screening colonoscopy were identified. Software from Virgo was utilized for cloud video capture of all colonoscopies, and videos were retrospectively segregated and analyzed using the AutoIBD-UC algorithm. Independent-sample T testing, one-way ANOVA, and receiver operating characteristic (ROC) curves were utilized for statistical analysis. Results: AutoIBD scores were significantly different between all control patients and all UC patients (mean 0.168 vs 0.318 respectively, P < 0.001), UC patients with inactive (Mayo 0) vs active (Mayo ≥1) disease (mean 0.193 vs 0.356, P < 0.001), and UC patients with inactive-mild (Mayo 0 or 1) vs moderate-severe (Mayo 2 or 3) disease (mean 0.238 vs 0.465, P < 0.001). Using ANOVA, there was no significant difference in AutoIBD scores between UC patients with inactive vs mild disease (Mayo 0 vs 1; P = 0.213) or moderate vs severe disease (Mayo 2 vs 3; P = 0.309). ROC analysis indicated an area under the curve (AUC) of 0.849 for all control patients vs all UC patients. Conclusion: Our single-center results support the external validity of the Virgo AutoIBD-UC algorithm in assessing and stratifying disease severity in patients with UC using real-world colonoscopy data.