OC147: Confirming the limited role of CA-125 for adnexal mass diagnosis

Objectives: To prospectively evaluate previously developed logistic regression models to distinguish between malignant and benign adnexal masses. Methods: The logistic regression models, which were developed using the data set (n = 1066) from phase 1 of the International Ovarian Tumor Analysis study group (IOTA), are applied to the prospectively collected phase 2 data set (n = 1940). Model performance was evaluated using receiver operating characteristic (ROC) curves and the area under them (AUC). A distinction is made between patients from old centers (phase 1 centers) and patients from new centers. Results: The logistic regression models under consideration are the previously published models LR1 (twelve inputs) and LR2 (six inputs), and an unpublished model (ObjLR) in which the subjective predictors in LR1 (i.e. pelvic pain during examination and the color Doppler score of intratumoral blood flow) were replaced by a binary variable indicating whether the tumor was unilocular or not. On the IOTA phase 1 test data, these models had AUCs of 0.942, 0.922, and 0.937, respectively. On the phase 2 patients from the seven old centers, the AUCs were 0.943, 0.915, and 0.936, respectively. On the patients from the 12 new centers, the AUCs were 0.949, 0.944, and 0.946, respectively. All but one center had an AUC above 0.90 for LR1 and ObjLR, eleven centers had at least one AUC of at least 0.95 (five old, six new centers). Conclusions: The logistic regression models from IOTA phase 1 have excellent performance when evaluated prospectively on both old and new centers.