Reproducibility and validity of pathologic classifications of benign breast disease and implications for clinical applications.

Background: Research studies on the relationship between benign breast diseases and cancer risk typically identify certain conditions as risk factors, and others as carrying no prognostic significance. This study addresses several issues concerning the relevance of such research results for advising individual patients in a clinical setting.

Methods: Data were obtained as part of a "blinded" retrospective pathology review of benign breast biopsies. A random sample of cases was reviewed twice, providing information about reliability. Comparisons with diagnoses that used information from the operative reports and gross pathology descriptions as well as microscopic histology were used to assess validity.

Results: Among cases that were reviewed twice, excellent agreement was achieved for diagnosing carcinoma and lobular neoplasia, good agreement for adenosis and intraductal papilloma, and relatively poor agreement about levels of hyperplasia and atypia, and whether ducts or lobules were involved. Distinctions among levels of hyperplasia also apparently were influenced by the number of slides available for review. The "blinded" review diagnoses frequently differed from the diagnoses that used all information available at the time of surgery in detecting the presence or absence of gross cystic disease, and in distinguishing solitary from multiple papillomas.

Conclusions: Problems with reliability of precise distinctions among levels and sites of hyperplasia and atypia seem to limit the usefulness of such classifications as guidelines for individual patient care. For conditions with some clinical manifestations, diagnoses based exclusively on histologic review of biopsy specimens often are not accurate.