Article Text

Download PDFPDF

EP238/#102  Support for standardization: ultrasound risk stratification models accurately discriminate benign from malignant adnexal lesions in the hands of novice operator
  1. Luigi De Vitis1,
  2. Gabriella Schivardi1,
  3. Leah Grcevich1,
  4. Ilaria Capasso1,
  5. Diletta Fumagalli1,
  6. Daniel Breitkopf1,
  7. Shannon Laughlin-Tommaso1,
  8. Angela Fought2,
  9. Melanie Caserta3,
  10. Mary Clingan3,
  11. Andrea Mariani1 and
  12. Carrie Langstraat1
  1. 1Mayo Clinic, Department of Obstetrics and Gynecology, Rochester, USA
  2. 2Mayo Clinic, Division of Clinical Trials and Biostatistics, Rochester, USA
  3. 3Mayo Clinic, Department of Radiology, Jacksonville, USA


Introduction It is unclear whether ultrasound risk stratification models for adnexal lesions perform well when used by novice providers. We aim to compare the performance of four commonly used models to detect ovarian cancer, when the operator has only basic experience.

Methods Women with adnexal masses treated in 2019 were identified retrospectively. Patients were included if they underwent surgery within 3 months of diagnosis or had at least 12±2 months of follow-up. A non-expert operator (European Federation of Societies for Ultrasound in Medicine and Biology level I) classified each lesion using ADNEX, two-step strategy (benign descriptors followed by ADNEX), O-RADS 2019, and O-RADS 2022. The primary outcome measure was AUC [95% confidence interval], compared across the four models.

Results A total of 556 women were included in the analyses: 452 benign and 104 malignant. The AUCs of ADNEX, the two-step strategy, O-RADS 2019, and O-RADS 2022 were 0.90 [0.87–0.94], 0.91 [0.88–0.94], 0.88 [0.85–0.91], and 0.88 [0.84–0.91], respectively (figure 1). The two-step strategy performed significantly better than the O-RADS algorithms (both p=0.01). With all the algorithms, the observed malignancy rate was 1.91–2.17% among lesions categorized as ‘almost certainly benign’, two-fold higher than the expected <1% (table 1). Out of the four methods, lesions wrongly classified as ‘almost certainly benign’ were borderline tumors (n=4) and metastases (n=3).

Conclusion/Implications In the hands of a novice provider, all algorithms performed well, and were able to distinguish benign from malignant lesions. ADNEX misclassified only one malignant patient as ‘almost certainly benign’, compared to 5–6 patients by the other models.

Abstract EP238/#102 Figure 1

Diagnostic performance of ADNEX, two-step strategy, O-RADS 2019 and O-RADS 2022. Above, ROC curves for the four models, Below, AUC, sensitivity, specificity, accuracy, positive and negative predictive values with 95% confidence intervals. Abbreviations: AUC, area under the curve; ROC, receiver operating curve; PPV, positive predictive value; NPV, negative predictive value

Abstract EP238/#102 Table 1

The calibration (i.e., the observed malignancy rate compared to the expected rate) is shown in Table 1A. Data are reported as number of malignant cases per cell/total number of patients in the cell (cell%). Table 1B describes the clinical and radiological characteristics of malignant cases misclassified as ‘almost certainly benign’ by at least one of the models. Abbreviations: BC, benign descriptor; EC, endometrial cancer; Gl, gastro-intestinal; na, not available; N, number; US, ultrasound

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.