A multivariate statistical study to obtain effective criteria to detect well‐differentiated adenocarcinoma in endometrial cytology

Abstract

Consistency in endometrial cytology is relatively poor. This can be partly attributed to generally accepted criteria based on cellular features. The cytological distinction between grade-1 adenocarcinoma and endometrial hyperplasia is more reliant on architectural features than cellular features. We examined statistical criteria based on cytoarchitecture for detecting grade-1 adenocarcinoma in endometrial cytology. Histologically, the study population consisted of 11 cases of grade-1 adenocarcinoma, 6 of atypical endometrial hyperplasia, 16 of endometrial hyperplasia without atypia, and 74 of a normal proliferative endometrium. In each case, all cellclumps were divided into five patterns (tubular; sheet; dilated and/or branched tubular; regular overlapping; atypical). The frequencies of each pattern were submitted to five-variate cluster analysis. The validity and reproducibility of cluster analysis were tested by canonical discriminant analysis and multigroup linear discriminant analysis, respectively. All 107 cases were classified into three groups, A (11), B (36), and C (60), by five-variate cluster analysis. In comparison with this classification and histopathologic diagnosis, group A corresponded to adenocarcinoma, and groups B and C correlated with non-carcinoma. Most cases of atypical endometrial hyperplasia were included in group B. These data suggest that statistical groupings based on cytoarchitecture are useful in the discrimination of grade-1 adenocarcinoma from endometrial hyperplasia and normal tissue.