Mayo Clinic College of Medicine and Science Rochester, MN
Edward V. Loftus, MD, FACG1, David T. Rubin, MD2, Nayantara Coelho-Prabhu, MD, FACG3, Walter Reinisch, MD, PhD4, Chakib Battioui, PhD5, Abbey Brinkey, 6, Daniel Colucci, BS6, William J. Eastman, MD5, Hannah Thompson, 6, Yeli Wang, PhD6, Shrujal Baxi, MD7 1Mayo Clinic College of Medicine and Science, Rochester, MN; 2The University of Chicago Medicine, Chicago, IL; 3Mayo Clinic, Rochester, MN; 4Medical University of Vienna, Vienna, Wien, Austria; 5Eli Lilly and Company, Indianapolis, IN; 6Iterative Health Inc., Cambridge, MA; 7Iterative Health, Cambridge, MA
Introduction: The endoscopic Mayo Score (eMS) provides an objective measure of therapeutic efficacy in Ulcerative Colitis (UC) trials; however, disagreement among central readers in assessing the eMS limits its reliability. Central readers have observed variability in the quality of endoscopic videos presented for scoring, suggesting one possible explanation of the discord. We identified the quality factors associated with reproducible eMS assessments in UC trials.
Methods: We used endoscopic video recordings from participants in a clinical trial for moderate-to-severe UC that went through the 2+1 central reading eMS assessment workflow. We stratified videos in a 1:2 ratio into an agreement cohort (n=50) with videos with an equivalent score provided by the first two readers, and a disagreement cohort (n=95) with videos with a unique score provided by all three readers. Candidate factors associated with video quality were defined at procedure, endoscopist, and technical levels, and assessed by trained annotators independently during insertion and withdrawal. We additionally applied a machine learning (ML) model to assess the proportion of interpretable frames in each video. Descriptive statistics assessed the relationship between candidate factors and videos in both cohorts.
Results: Quality of bowel preparation was worse among videos in the disagreement cohort during insertion (24% vs 42.1% inadequate/poor, p=0.03), though similar during withdrawal (22% vs 29.5% inadequate/poor, p=0.54). Withdrawal time was greater in the disagreement cohort (11:08±05:00 vs 14:19±08:47, p=0.006). Other procedure and endoscopist factors were similar among videos in both cohorts during insertion and withdrawal (Table 1). ML model results identified no difference in the proportion of interpretable frames between cohorts during withdrawal (43.2%±15.7% vs 43.9%±15.0%, p=0.80) (Figure 1). Incomplete procedure recordings occurred at equal rates (26% vs 22.1%, p=0.62) and video resolution was standard definition in both cohorts.
Discussion: Videos with disagreement in eMS assessments in trials are those with a worse quality of bowel preparation. The improvement in quality of bowel preparation in the disagreement cohort during withdrawal is likely attributed to the increased time spent on the examination. The absence of differences in other factors during scope withdrawal may indicate that human factors also contribute to reader disagreement. Standardized solutions to address these differences in video interpretation are needed
Figure: Figure 1: Interpretability model scores for videos in the agreement and disagreement cohorts.
Note: The table for this abstract can be viewed in the ePoster Gallery section of the ACG 2024 ePoster Site or in The American Journal of Gastroenterology's abstract supplement issue, both of which will be available starting October 27, 2024.
Chakib Battioui: Eli Lilly & Company – Employee, Stock-publicly held company(excluding mutual/index funds).
Abbey Brinkey: Iterative Health, Inc. – Employee, Stock Options.
Daniel Colucci: Iterative Health, Inc. – Employee, Stock Options.
William Eastman: Eli Lilly & Company – Employee, Stock-publicly held company(excluding mutual/index funds).
Hannah Thompson: Iterative Health, Inc. – Employee, Stock Options.
Yeli Wang: Iterative Health, Inc. – Employee, Stock Options.
Shrujal Baxi: Iterative Health, Inc. – Employee, Stock Options.
Edward V. Loftus, MD, FACG1, David T. Rubin, MD2, Nayantara Coelho-Prabhu, MD, FACG3, Walter Reinisch, MD, PhD4, Chakib Battioui, PhD5, Abbey Brinkey, 6, Daniel Colucci, BS6, William J. Eastman, MD5, Hannah Thompson, 6, Yeli Wang, PhD6, Shrujal Baxi, MD7. P2395 - The Role of Video Quality in Central Reader Disagreement in Assessment of the Endoscopic Mayo Score in Ulcerative Colitis Trials, ACG 2024 Annual Scientific Meeting Abstracts. Philadelphia, PA: American College of Gastroenterology.