Agreement in assessing endometrial pathology: The postmenopausal estrogen/progestin interventions (PEPI) trial

We report on agreement in interpreting endometrial biopsy specimens between the local and central pathologists of the Postmenopausal Estrogen/Progestin Interventions (PEPI) trial. This trial was a 3-year, multicenter, randomized, double-masked, placebo-controlled trial of four groups taking estrogen or estrogen/progestin combinations. A total of 1804 follow-up biopsies were performed in 596 subjects. Relative sensitivity and relative specificity using the diagnosis from the central pathologist as the gold standard and overall agreement are presented. Almost 90% of the diagnoses were reported normal by both readers. There were significant differences in agreement among clinics and treatment arms (p < 0.0001). The visit at which the biopsy specimen was obtained, age at baseline, prior postmenopausal estrogen use, parity, and drug adherence were not associated with agreement between the two readers. Higher proportions of disagreement were seen in two clinics (13% and 11%) compared with the other five clinics (2%-5%). Biopsy specimens from participants who were taking conjugated equine estrogens (CEE) only were more likely to be diagnosed differently by both readers (11%) than biopsy specimens from women taking a placebo (2%) or CEE combined with progestins (5%). Relative specificity varied from 86.4% to 98.9% among the clinics (p < 0.0001). Relative sensitivity was based on a small number of diagnoses, as few biopsy specimens were classified abnormal by the central pathologist. In patients assigned to CEE combined with progestin, 5 of the 7 biopsy specimens that were recorded abnormal by the central pathologist received a normal diagnosis locally. Our findings show that sample size requirements for study designs in which a central reader is used can be at least threefold lower than the requirements for designs relying on local diagnoses. Centralized protocols for endometrial histopathology reading and staff training are highly desirable in multicenter trials.