Pathologist interobserver variability of histologic features in childhood brain tumors: results from the CCG-945 study.

In the Children's Cancer Group-945 trial, study design allowed estimation of overall interpathologist observational agreement for 6 histologic features frequently used in brain tumor diagnoses. We evaluated agreement between pairs of 5 experienced neuropathologists, who had knowledge of the general diagnoses prior to slide readings. We performed this study in an attempt to further improve pathologist interinstitutional agreement. The features mitosis, necrosis, and giant cells had "fair" overall kappa estimates of reproducibility of around 0.5, while endothelial proliferation had only a "poor" overall kappa of 0.35. The Rogot reproducibility index averaged 0.5 for pleomorphism and hyperchromia. The upper bounds for the 10 pair summary agreement estimates were at best 0.65 ("good") for all 6 features. These relatively low-reproducibility estimates for the very small number of histologic features being assessed in tumors institutionally diagnosed as high-grade gliomas indicate that neuropathologists either used different operational definitions or interpreted them differently. We found that we could rank the histologic features from best to worst agreement among study pathologists as necrosis, giant cells, mitosis, endothelial proliferation, hyperchromic nuclei, and pleomorphic cells. We suggest that neuropathologists involved in multi-institutional studies of putative therapies not discard these traditional histologic features, but rather develop standardized operational definitions and measure their variability before beginning the studies. Only after such histologic feature variability studies are conducted will we have the data to identify specific histologic features of value to clinicians and researchers. Agreement and strict adherence to improved nonsubjective diagnostic criteria would improve histologic feature reliability and, consequently, their usefulness in studies.