ABSTRACT Objectives The primary aims of this study were to validate externally the diagnostic performance of the Assessment of Different NEoplasias in the adneXa (ADNEX) model for distinguishing between malignant and benign adnexal masses, when used by non‐expert ultrasound examiners in general gynecological settings with a low prevalence of ovarian cancer, and to compare the performance of ADNEX with that of the Risk of Malignancy Index (RMI) and serum CA 125 alone. The secondary aim was to assess model performance in detecting early‐stage ovarian malignancy. Methods This was a prospective multicenter external validation study conducted in all gynecological departments (five hospitals) in the Central Denmark Region between May 2021 and December 2022, including all consecutive patients diagnosed with a non‐physiological adnexal mass on transvaginal and usually transabdominal ultrasound. Standardized ultrasound terminology developed by the International Ovarian Tumor Analysis (IOTA) group was implemented uniformly across study sites, all of which received training in IOTA terminology before and during the study period. IOTA certification was offered to clinicians performing the ultrasound examinations. Sonographic features according to IOTA terminology were recorded systematically using a structured electronic form integrated into the medical record, which automatically calculated the RMI score and ADNEX risk estimates. The reference standard was histopathological diagnosis within 12 months after the inclusion scan. For patients managed expectantly, the reference standard was determined by follow‐up during a 1‐year observation period. This included serial ultrasound examinations and review of medical records and national histopathology records. Ovarian cancer, borderline tumors and metastases to the ovary were considered malignant. Areas under the receiver‐operating‐characteristic curve (AUCs) for ADNEX, RMI and CA 125 were calculated and compared using meta‐analysis of center‐specific results and DeLong's test. Diagnostic metrics were estimated at different cut‐offs for any‐stage and early‐stage malignancy. Clinical utility for guiding referral was assessed by decision‐curve analysis, with net benefit calculated over a range of risk thresholds (1–50%). Results A total of 1815 patients diagnosed with an adnexal mass on ultrasound examination were included, of whom 1656 were examined by non‐experts in general gynecological settings (59.2% postmenopausal, 8.3% prevalence of malignancy). The AUC of the ADNEX model (0.93 (95% CI, 0.90–0.95) with CA 125 and 0.92 (95% CI, 0.89–0.94) without CA 125) was significantly greater than that of the RMI (0.88 (95% CI, 0.85–0.91)) and CA 125 alone (0.82 (95% CI, 0.77–0.87)). Including CA 125 did not improve the discriminative ability of ADNEX for identifying malignancy. For diagnosing early‐stage malignancy, ADNEX without CA 125 had an AUC of 0.90 (95% CI, 0.85–0.94), which was significantly greater than that of the RMI (0.82 (95% CI, 0.76–0.88)) and CA 125 alone (0.73 (95% CI, 0.65–0.80)). At a cut‐off of 5%, the ADNEX model without CA 125 detected 84.4% (95% CI, 69.0–93.0%) of early‐stage ovarian malignancies, with a specificity of 81.6% (95% CI, 76.9–85.6%). Clinical‐decision curves showed that ADNEX had clinical utility superior to that of the RMI at a cut‐off score of 200 over the entire range of risk thresholds. Conclusions The ADNEX model demonstrated excellent diagnostic performance for identifying adnexal malignancy when used prospectively in routine clinical practice in a general gynecological setting by non‐expert ultrasound examiners trained in IOTA terminology. ADNEX significantly outperformed the RMI and serum CA 125 alone, including in the detection of early‐stage ovarian malignancy. Including CA 125 in the ADNEX model did not provide additional diagnostic benefit for estimating the likelihood of malignancy. © 2026 The Author(s). Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of International Society of Ultrasound in Obstetrics and Gynecology.
Hjorth et al. (Thu,) studied this question.