Diagnostic Performances of Ultrasound-Based Models for Predicting Malignancy in Patients with Adnexal Masses

Abstract

This study compared the diagnostic performance of different ultrasound-based models in discriminating between benign and malignant ovarian masses in a Filipino population. This was a prospective cohort study in women with findings of an ovarian mass on ultrasound. All included patients underwent a physical examination before level III specialist ultrasonographic and Doppler evaluation using the different International Ovarian Tumor Analysis (IOTA) Group’s risk models. Serum CA-125 and a second-generation multivariate index assay (MIA2G) were also determined for all patients. The ovarian imaging and biomarker results were correlated with the histological findings. A total of 260 patients with completed ultrasound, CA-125, MIA2G, and histopathologic results was included in the study. The presence of papillae with blood flow and irregular cyst walls during the ultrasound were significantly associated with a 20-fold (OR: 20.13, CI: 8.69–46.67, p < 0.01) and 10-fold (OR: 10.11, CI: 5.30–19.28, p < 0.01) increase in the likelihood of a malignant lesion, respectively. All individual sonologic procedures performed well in discerning malignant and benign ovarian lesions. IOTA-LR1 showed the highest accuracy (82.6%, 95% CI: 77.5–87%) for identifying ovarian cancer. IOTA-ADNEX showed the highest sensitivity (93.3%, 95% CI: 87.2–97.1%) while IOTA-LR2 exhibited the highest specificity (84.4%, 95% CI: 77.3–90%). Among the different serial test combinations, IOTA-LR1 with MIA2G and IOTA-LR2 with MIA2G showed the highest diagnostic accuracy (AUROC = 0.82). This study showed that all individual ultrasound-based models performed well in discerning malignant and benign ovarian lesions, with IOTA-LR1 exhibiting the highest accuracy.