Ultrasound diagnosis of endometrial cancer by subjective pattern recognition in women with postmenopausal bleeding: a prospective inter‐rater agreement and reliability study

OBJECTIVES The objective of our study was to assess the inter-rater agreement and reliability of using subjective pattern recognition to diagnose endometrial cancer on ultrasound in women with postmenopausal bleeding. METHODS This was a prospective cross-sectional study at a gynaecological rapid access clinic, between Oct 2016 - Dec 2017, where consecutive women with postmenopausal bleeding and endometrial thickness of ≥4.5mm on transvaginal ultrasound scan were included. We excluded women on hormone replacement therapy, tamoxifen or with a history of primary gynaecological malignancy. Two raters independently performed ultrasound examinations, blinded to each other's findings, and classified women into having uniformly thickened endometrium, benign polyp or endometrial cancer, by subjective pattern recognition. Inter-rater reliability of the ultrasound diagnoses was assessed by using Cohen's kappa statistics. All women subsequently underwent either outpatient endometrial biopsy, hysteroscopy or hysterectomy. RESULTS Forty women were included in the study with a median age of 61 (IQR 57-69) and a median endometrial thickness of 11.0mm (IQR 6.2-20.3). Final histological analysis confirmed 16 (40%) women with endometrial cancer, 16 (40%) endometrial polyps, 4 (10%) atrophic endometrium, 3 (7%) proliferative endometrium and 1 (3%) endometrial hyperplasia. Inter-rater agreement for the ultrasound diagnoses of uniformly thickened endometrium, polyp and cancer, were 14/16 (87.5%), 22/30 (73.3%) and 28/34 (82.4%), respectively; the inter-rater reliability was good (κ = 0.69, 95% C.I. 0.49-0.88). When the ultrasound diagnoses were combined as either cancer or no cancer, the inter-rater agreement was 85% and the inter-rater reliability was also good (κ = 0.78, 95% C.I. 0.61-0.95). Rater A correctly identified 14/16 cancers and Rater B identified 15/16. Endometrial cancers were misdiagnosed as benign polyps on ultrasound in two women by Rater A, and in one woman by Rater B. The overall accuracies of Rater A and Rater B in differentiating between benign endometrial pathologies and malignancy were 90% and 85%, respectively. CONCLUSIONS Our results showed good inter-rater reliability of subjective pattern recognition in diagnosing uniformly thickened endometrium, polyp and cancer on ultrasound in women with postmenopausal bleeding. Our findings should facilitate wider use of subjective pattern recognition in routine clinical practice. This article is protected by copyright. All rights reserved.