Reproducibility of the diagnosis of endometrial hyperplasia, atypical hyperplasia, and well-differentiated carcinoma

Many studies have attempted to identify histologic features that aid in the distinction of atypical hyperplasia (AH) from hyperplasia without atypia and well-differentiated endometrioid carcinoma, but few have evaluated the reproducibility of these diagnoses. Five pathologists independently reviewed 100 endometrial biopsy and curettage specimens chosen to represent the entire spectrum of proliferative lesions of the endometrium, including proliferative endometrium (PEM), hyperplasia without atypia, AH, and well-differentiated endometrioid carcinoma. Slides were reviewed twice for diagnosis, with an intervening evaluation of a checklist of histologic features. Intraobserver and interobserver agreement were assessed using the kappa statistic. Intraobserver kappa values ranged from 0.67 to 0.89 (76% to 89% agreement). Interobserver kappa values by diagnostic category were: proliferative endometrium: 0.86; hyperplasia without atypia: 0.60; AH: 0.47; well-differentiated endometrioid carcinoma: 0.83; with a kappa value of 0.69 for all cases combined. Associations between the selected histologic features and the given diagnoses for each pathologist were analyzed using multiple logistic regressions to identify features that were useful for distinguishing among diagnostic categories. Histologic features determined by univariable and multivariable analyses that were found to be most associated with distinguishing diagnostic categories were: proliferative endometrium versus hyperplasia without atypia: gland crowding (univariable, multivariable), and gland branching (univariable); hyperplasia without atypia versus AH: presence of nucleoli (univariable, multivariable), nuclear enlargement (univariable), vesicular chromatin change (univariable), nuclear pleomorphism (univariable), chromatin irregularities (univariable), and loss of polarity (univariable); hyperplasia without atypia versus carcinoma: glandular confluence/complex cribriform pattern (univariable, multivariable), stromal alteration (univariable, multivariable), and necrosis (univariable). In summary, interobserver agreement was good but was lowest for AH. Only the presence of nucleoli was strongly associated with distinction of AH from hyperplasia without atypia. Individual pathologists use additional features to diagnose atypia, but these features are not consistently associated with that diagnosis. Cribriform architectural pattern and stromal alteration were associated with the distinction of well-differentiated endometrioid carcinoma from AH.