Low interobserver agreement among endoscopists in differentiating dysplastic from non-dysplastic lesions during inflammatory bowel disease colitis surveillance

Objectives. During endoscopic surveillance in patients with longstanding colitis, a variety of lesions can be encountered. Differentiation between dysplastic and non-dysplastic lesions can be challenging. The accuracy of visual endoscopic differentiation and interobserver agreement (IOA) has never been objectified. Material and methods. We assessed the accuracy of expert and nonexpert endoscopists in differentiating (low-grade) dysplastic from non-dysplastic lesions and the IOA among and between them. An online questionnaire was constructed containing 30 cases including a short medical history and an endoscopic image of a lesion found during surveillance employing chromoendoscopy. Results. A total of 17 endoscopists, 8 experts, and 9 nonexperts assessed all 30 cases. The overall sensitivity and specificity for correctly identifying dysplasia were 73.8% (95% confidence interval (CI) 62.1-85.4) and 53.8% (95% CI 42.6-64.7), respectively. Experts showed a sensitivity of 76.0% (95% CI 63.3-88.6) versus 71.8% (95% CI 58.5-85.1, p = 0.434) for nonexperts, the specificity 61.0% (95% CI 49.3-72.7) versus 47.1% (95% CI 34.6-59.5, p = 0.008). The overall IOA in differentiating between dysplastic and non-dysplastic lesions was fair 0.24 (95% CI 0.21-0.27); for experts 0.28 (95% CI 0.21-0.35) and for nonexperts 0.22 (95% CI 0.17-0.28). The overall IOA for differentiating between subtypes was fair 0.21 (95% CI 0.20-0.22); for experts 0.19 (95% CI 0.16-0.22) and nonexpert 0.23 (95% CI 0.20-0.26). Conclusion. In this image-based study, both expert and nonexpert endoscopists cannot reliably differentiate between dysplastic and non-dysplastic lesions. This emphasizes that all lesions encountered during colitis surveillance with a slight suspicion of containing dysplasia should be removed and sent for pathological assessment.