OC149: The ability of four ultrasound methods to discriminate between benign and malignant tumors considered difficult to classify by ultrasound examiners

Abstract

Objectives: To prospectively evaluate the results of the CA-125 analyses performed on the IOTA (International Ovarian Tumor Analysis group) phase 1 data. Methods: Using the IOTA phase 1 data, logistic regression models with and without CA-125 as a predictor were developed to predict malignancy of ovarian tumors in preand postmenopausal women. These models were applied to the prospectively collected data from phase 2 (n = 1940). Their performance was evaluated using receiver operating characteristic (ROC) curves, area under the curve (AUC), and the sensitivity at 75% specificity (Sens75). Next, the performance of the expert operator’s subjective impression (pattern recognition) was compared to CA-125 alone, and the median CA125 level was computed for each histological endpoint. To this end, the data from IOTA phases 1, 1b, and 2 were used (n = 3513). Results: In premenopausal women, the logistic regression models with and without CA-125 had AUCs of 0.940 and 0.939, respectively. Sens75 levels were 94.2 and 95.9%. For postmenopausal women, the models with and without CA-125 had AUC values of 0.937 and 0.920, respectively, with Sens75 levels of 91.3 and 92.0%. The ROC curves for models with and without CA-125 were nearly identical in their top left part where sensitivity and specificity are high. Pattern recognition (sensitivity 92%, specificity 90%) clearly performed better than CA-125 alone. CA-125 levels were high in endometriomas (median = 20 U/ml) and abscesses (40 U/ml). Borderline stage I tumors had a median CA-125 level of 29 U/ml, rare primary invasive tumors of 52 U/ml. The highest CA-125 levels were obtained for primary invasive tumors of stage II–IV. Conclusions: The results confirmed our previous conclusions regarding the role of CA-125 in ovarian tumor diagnosis. This marker need not be used in mathematical models for preor postmenopausal patients. Characterizing ovarian pathology using CA-125 alone is inferior to pattern recognition by an experienced operator.