Reproducibility of endometrial intraepithelial neoplasia diagnosis is good, but influenced by the diagnostic style of pathologists

Abstract

AbstractEndometrial intraepithelial neoplasia (EIN) applies specific diagnostic criteria to designate a monoclonal endometrial preinvasive glandular proliferation known from previous studies to confer a 45-fold increased risk for endometrial cancer. In this international study we estimate accuracy and precision of EIN diagnosis among 20 reviewing pathologists in different practice environments, and with differing levels of experience and training. Sixty-two endometrial biopsies diagnosed as benign, EIN, or adenocarcinoma by consensus of two expert subspecialty pathologists were used as a reference comparison to assess diagnostic accuracy of 20 reviewing pathologists. Interobserver reproducibility among the 20 reviewers provided a measure of diagnostic precision. Before evaluating cases, observers were self-trained by reviewing published textbook and/or online EIN diagnostic guidelines. Demographics of the reviewing pathologists, and their impressions regarding implementation of EIN terminology were recorded. Seventy-nine percent of the 20 reviewing pathologists' diagnoses were exactly concordant with the expert consensus (accuracy). The interobserver weighted κ values of 3-class EIN scheme (benign, EIN, carcinoma) diagnoses between expert consensus and each of reviewing pathologists averaged 0.72 (reproducibility, or precision). Reviewing pathologists demonstrated one of three diagnostic styles, which varied in the repertoire of diagnoses commonly used, and their nonrandom response to potentially confounding diagnostic features such as endometrial polyp, altered differentiation, background hormonal effects, and technically poor preparations. EIN diagnostic strategies can be learned and implemented from standard teaching materials with a high degree of reproducibility, but is impacted by the personal diagnostic style of each pathologist in responding to potential diagnostic confounders.