P3669 - Evaluating the Performance of a Computer-Aided Diagnosis System in Implementing Diagnose-and-Leave and Resect-and-Discard Strategies for Diminutive Colorectal Polyps: A Real-World Pragmatic Study

William Cheng, BSc¹, Yen-Lin Kuo, MD², I-Chia Su, MD², Bai-Ping Lee, RN², Yi-Ning Tsui, RN², Hui-Ting Lin, RN², Yun-Shien Lee, PhD³, Chi-Liang Cheng, MD^2
1Indiana University School of Medicine, Carmel, IN; ²Evergreen General Hospital, Taoyuan, Taoyuan, Taiwan; ³School of Health Technology, Ming Chuan University, Taoyuan, Taoyuan, Taiwan

Introduction: For acceptance of computer-aided diagnosis (CADx) of diminutive (≤ 5 mm) polyps, CADx should match the established Preservation and Incorporation of Valuable Endoscopic Innovations (PIVI) standards to implement the diagnose-and-leave and resect-and-discard strategies. However, available validation data in a real-world setting are limited. We therefore aimed to assess whether the CADx optical diagnosis of diminutive colorectal polyps met the criteria to be implemented in a clinical setting.

Methods: This retrospective study with an approved CADx system (CAD EYE, Fujifilm, Japan) was conducted in a community hospital in Taiwan. Consecutive patients aged ≥ 40 years were included. The CADx system differentiated adenomas from non-adenomas using blue-light imaging. For each polyp, CADx output was matched against the histology gold standard. The follow-up interval after colonoscopy and polypectomy was determined by combining CADx diagnosis for diminutive polyps and histological diagnosis for ≥ 6 mm polyps. The primary outcomes were CADx performance in the diagnose-and-leave strategy for rectosigmoid diminutive polyps and in the resect-and-discard strategy for all diminutive polyps.

Results: Between March and August 2023, a total of 1249 polyps were detected and resected in 471 patients. Table 1 shows demographic and procedural data. Of these, 311 (25%) diminutive rectosigmoid polyps were retrieved for histology. There were 247 of 311 polyps (79%) diagnosed by CADx as non-adenomatous and were amenable for the diagnose-and-leave strategy (Figure). The negative predictive value (NPV), sensitivity, specificity, and accuracy of CADx for rectosigmoid diminutive polyps were 83.8% (95% confidence interval [CI], 78.7%─87.9%), 54.5% (95% CI, 44.2%─64.5%), 92.8% (95% CI, 88.7%─95.5%), and 82.0% (95% CI, 77.3%─85.9%), respectively. Overall, 1084 diminutive polyps were retrieved throughout the colon. Based on the CADx output, 837 polyps were amenable to the resect-and-discard strategy, resulting in an 80.3% (95% CI, 76.4%─83.6%) and 94.7% (95% CI, 92.3%─96.4%) agreement between the CADx- and histology-based surveillance intervals according to American and European guidelines (Figure).

Discussion: In this study, CADx fell short of the PIVI thresholds for implementation in the diagnose-and-leave and resect-and-discard strategies according to US Multi-Society Task Force guidelines. Further research is warranted to improve the performance of CADx.



Disclosures:


William Cheng, BSc¹, Yen-Lin Kuo, MD², I-Chia Su, MD², Bai-Ping Lee, RN², Yi-Ning Tsui, RN², Hui-Ting Lin, RN², Yun-Shien Lee, PhD³, Chi-Liang Cheng, MD². P3669 - Evaluating the Performance of a Computer-Aided Diagnosis System in Implementing Diagnose-and-Leave and Resect-and-Discard Strategies for Diminutive Colorectal Polyps: A Real-World Pragmatic Study, ACG 2024 Annual Scientific Meeting Abstracts. Philadelphia, PA: American College of Gastroenterology.