Use of computerized morphometric analyses of endometrial hyperplasias in the prediction of coexistent cancer

OBJECTIVE: Our purpose was to determine whether computerized morphometric analysis is predictive of coexistent cancer in uteri that show endometrial hyperplasia in curettings or biopsy specimens. STUDY DESIGN: Forty-five patients with endometrial hyperplasia and 10 patients with well-differentiated cancers diagnosed from curettings or biopsy specimens and treated by hysterectomy at Thomas Jefferson University Hospital between 1989 and 1993 were identified from the pathology department archives. Curettings were analyzed by computerized morphometric analysis at the Free University Hospital in Amsterdam. Pathologists performing the morphometric analyses were blinded to the pathologic diagnoses obtained by examining the hysterectomy specimens. The histopathologic classification of the hysterectomy specimens were used as the end point. RESULTS: Twelve of 45 patients with endometrial hyperplasia (26.7%) by preoperative histopathologic classification showed coexistent carcinoma at hysterectomy. All instances of carcinoma occurred in patients with atypical hyperplasia. Sensitivity of morphometric analysis to predict carcinoma was 100%, with a specificity of 88.5%. The positive predictive value was 83.3%, and the negative predictive value was 100%. A blinded reanalysis of the quantitative analysis in 16 patients showed good reproducibility of this technique (r = 0.93). CONCLUSIONS: Morphometric analysis is useful for predicting which patients with endometrial hyperplasia have coexistent carcinomas. Computerized morphometric analysis may be useful in therapeutic decision making for complex atypical hyperplasia.