OC21.03: Diagnostic accuracy of direct and indirect signs of adenomyosis and interrater reliability: a prospective multicentre study

To assess the prevalence and diagnostic accuracy of direct and indirect features of adenomyosis and the interrater reliability between less experienced examiners and one expert when predicting the presence of adenomyosis with transvaginal ultrasound (TVS). This prospective, observational multicentre study involved a consecutive cohort of 464 premenopausal women aged 25–52, scheduled for hysterectomy from 02/2020 to 06/2022. Participants received systematic 2D and 3D TVS exams by one of four gynecologists. An expert gynecologist later reassessed stored images, videos, and 3D volumes. Both direct and indirect signs of adenomyosis were recorded, along with the raters' overall impressions of the presence of adenomyosis. The diagnostic gold standard was the detailed, standardised histopathology of the hysterectomy specimen. Interrater reliability (IRR) was evaluated using Cohen's kappa. 254/464 (54.7%) had adenomyosis on histology. For the less experienced examiners and the expert, respectively, the diagnostic accuracy for the overall prediction of adenomyosis yes/no was 63.8 and 68.1%, sensitivity 72.1 and 58.6%, specificity 53.8 and 79.6%, PPV 65.4 and 77.7% and NPV 61.4% for both. Amongst the less experienced examiners, the diagnostic accuracy varied from 57.1–67.3%. Myometrial cysts was the most specific sign for all (specificity 78.3 and 95.4%, accuracy 56.8 and 59.3%), and hyperechogenic islands was the most sensitive sign (sensitivity 75.7 and 61%, accuracy 59.2 and 63%). The IRR showed a poor to fair agreement (κ) for the direct signs (subendometrial buds 0.16, hyperechogenic islands 0.19, myometrial cysts 0.28), and a poor to moderate agreement for the indirect signs (fan-shaped shadowing 0.13, wall asymmetry 0.30, globular uterus 0.41). This large study revealed that less experienced examiners can over-diagnose adenomyosis, leading to a low specificity and moderate diagnostic accuracy. A poor IRR for adenomyosis signs indicates examiner uncertainty in identifying specific features by TVS.