Reproducibility of Endometrial Pathologic Findings Obtained on Hysteroscopy, Transvaginal Sonography, and Gel Infusion Sonography in Women With Postmenopausal Bleeding

STUDY OBJECTIVE: To evaluate and compare interobserver variation in endometrial pattern recognition with hysteroscopy (HY) and transvaginal sonography (TVS) and gel infusion sonography (GIS) with regard to the diagnosis of endometrial pathology. DESIGN: Prospective study (Canadian Task Force II-1). SETTING: University clinic. PATIENTS: One hundred twenty-two consecutive women with postmenopausal bleeding and an endometrium thickness ≥ 5 mm. INTERVENTION: Two observers using HY and 2 others using TVS and GIS evaluated the endometrial pattern in recorded video clips. Interobserver agreement regarding findings obtained with TVS, GIS, and HY for a diagnosis of cancer, hyperplasia, polyps, and no endometrial pathology was expressed by κ coefficients and compared. MEASUREMENT AND MAIN RESULTS: Interobserver agreement (κ) was as follows: identification of normal endometrium: HY (.74), TVS (.68), and GIS (.48); diagnosis of cancer: HY (.56), TVS (.59), and GIS (.34); classification in all categories of endometrial pathology: HY (.70), TVS (.47), and GIS (.41) (p < .05 HY vs GIS). The presence of additional endometrial polyps decreased agreement on HY in patients with hyperplasia or cancer. Observer agreement was poor regarding the diagnosis of hyperplasia by all techniques. CONCLUSION: Observer agreement regarding both HY and TVS was reliable for the diagnosis of a normal endometrium but poor with HY, TVS, and especially GIS for a diagnosis of cancer. In patients with hyperplasia or cancer, agreement between observers was especially low in the presence of additional polyps when HY was used. These findings call attention to the need for systematic methods to improve reliability in endometrial pattern recognition.