Interobserver agreement of proliferation index (Ki-67) outperforms mitotic count in pulmonary carcinoids

Abstract

Evaluation of proliferative activity is a cornerstone in the classification of endocrine tumors; in pulmonary carcinoids, the mitotic count delineates typical carcinoid (TC) from atypical carcinoid (AC). Data on the reproducibility of manual mitotic counting and other methods of proliferation index evaluation in this tumor entity are sparse. Nine experienced pulmonary pathologists evaluated 20 carcinoid tumors for mitotic count (hematoxylin and eosin) and Ki-67 index. In addition, Ki-67 index was automatically evaluated with a software-based algorithm. Results were compared with respect to correlation coefficients (CC) and kappa values for clinically relevant grouping algorithms. Evaluation of mitotic activity resulted in a low interobserver agreement with a median CC of 0.196 and a median kappa of 0.213 for the delineation of TC from AC. The median CC for hotspot (0.658) and overall (0.746) Ki-67 evaluation was considerably higher. However, kappa values for grouped comparisons of overall Ki-67 were only fair (median 0.323). The agreement of manual and automated Ki-67 evaluation was good (median CC 0.851, median kappa 0.805) and was further increased when more than one participant evaluated a given case. Ki-67 staining clearly outperforms mitotic count with respect to interobserver agreement in pulmonary carcinoids, with the latter having an unacceptable low performance status. Manual evaluation of Ki-67 is reliable, and consistency further increases with more than one evaluator per case. Although the prognostic value needs further validation, Ki-67 might perspectively be considered a helpful diagnostic parameter to optimize the separation of TC from AC.