Evaluation of the reproducibility of the World Health Organization classification of common ovarian cancers. With emphasis on methodology.

Seven pathologists independently classified 50 slides of ovarian tumors using category I of the World Health Organization classification (WHO I), each case being seen twice under different random code numbers. Intraobserver reproducibility and interobserver reproducibility, based on consistent interpretations, were both suboptimal. However, scrutiny suggested that no pathologist was a source of excessive variability, nor was suboptimal interobserver reproducibility simply due to intraobserver variability. Neither could excessive variability be attributed to skewing of results by a subgroup of unclassifiable cases. However, clearcut sources of variability were identified among the categories of WHO I, namely, mixed epithelial, unclassified epithelial, and undifferentiated carcinoma. There was also considerable variability in distinguishing serous and endometrioid neoplasms, and in identifying tumors of low malignant potential. These findings should not be misconstrued as implying that pathologists in routine practice cannot diagnose common ovarian cancers reproducibly for patient care purposes. Availability of clinical and macroscopic data, extensive sampling, histochemistry, and consultation combine, in an uncontrolled and highly individualistic fashion, to render routine service work very different from this highly controlled formal exercise. Furthermore, at the current state of the therapeutic art, many of the taxonomic problems identified in this study may have little clinical significance. Nonetheless, this study has strengthened the evidence that there may be important problems in classifying common ovarian cancers reproducibly using WHO I, and that WHO I may require greater clarity to enhance reproducibility. Current emphasis on quality assurance dictates reconsideration of the literature on reproducibility of histopathologic taxonomy, which has tended to inculpate pathologists as sources of variability. Virtually all of this literature is subject to some degree of skepticism due to deficiencies in methodology. Consideration of the question of how to measure reproducibility in anatomic pathology leads us to suggest that the community of pathologists should address the need to decrease ambiguity in classification systems as an important step toward optimizing reproducibility.