Background and aim: The risk of cancer varies with the subtype of colorectal "laterally spreading tumors" (LSTs). However, visual interpretations vary among endoscopists. The aim of this study was to evaluate inter-observer agreement and accuracy in the endoscopic classification of LST subtypes among experts and trainees.
Methods: In total, 40 LST images were collected and reviewed independently by 14 gastroenterology experts and 10 trainees. All investigators recorded their findings as one of the following four categories: homogeneous, nodular mixed, flat-elevated, and pseudo-depressed. Agreement was assessed in terms of the kappa (κ) statistic and AC1 estimate. Accuracy is reported as percentage agreement with the gold standard, based on the gross morphology of the resected specimens.
Results: Of the possible 91 pair-wise κ estimates among experts, 41 (45.1%) were >0.75, indicating excellent agreement, while only 2 (4.44%) of the 45 pair-wise κ estimates among trainees were >0.75. Agreements for individual LST subtypes in the trainee group were significantly lower than those in the expert group. The κ and AC1 estimates showed similar values in individual subtypes of LSTs. The overall accuracy of LST was also significantly higher for the experts than the trainees (85.9 vs. 72.5%, P < 0.001). Notably, the flat-elevated subtype showed the lowest agreement and accuracy and was frequently misclassified as the pseudo-depressed subtype by both groups.
Conclusions: Inter-observer agreement and accuracy for LST subtype classification differ significantly between experts and trainees. Implementation of an adequate training system for beginners is necessary to better identify colorectal LSTs.