20th World Congress on Ultrasound in Obstetrics and Gynecology Oral communication abstracts simple rules were not applicable. 649 of these masses had CA-125 results available. The malignancy rate was 31% (203/649). The data were randomly divided into a development set (70% of the data, n = 457) and a test set (30% of the data, n = 192). Results: In the development set, the following variables were retained as predictors of malignancy: age, personal history of ovarian cancer, largest diameter of the lesion, largest diameter of the largest solid component, presence of blood flow in papillary projection and CA- 125. The area under the ROC curve (AUC) was 0.86 (95% CI, 0.82–0.90) in the development set and 0.83 (0.76–0.89) in the test set. Expert’s subjective assessment and RMI had an AUC in the test set of 0.88 (0.82–0.94) and 0.69 (0.60–0.77), respectively. To be able to compare the performance of the new model with LR1 and LR2, we excluded from the test set the 53 cases that were part of the original development set for LR1 and LR2 (IOTA phase 1, 1999–2004). On the reduced test set of 139 cases, the AUC of the new model, subjective assessment, RMI, LR1 and LR2 were 0.83 (0.76–0.91), 0.87 (0.80–0.94), 0.72 (0.63–0.81), 0.79 (0.71–0.87) and 0.72 (0.63–0.82), respectively. Conclusions: In adnexal masses where the simple rules are not applicable, our new mathematical model might be useful, but its performance must be prospectively validated. OC23.02 Which histologies limit the ability of CA-125 to characterize adnexal masses? An observational study of 3511 patients B. Van Calster 1 , C. Van Holsbeke 2,8 , J. Zhang 3 , D. Jurkovic 4 , A. A. Lissoni 5 , L. I. Valentin 6 , A. Testa 7 , A. Czekierdowski 9 , D. Fischerova 10 , G. Van de Putte 2 , I. Vergote 8 , S. Van Huffel 1 , T. Bourne 8,11 , D. Timmerman 8 1 Katholieke Universiteit Leuven, Leuven, Belgium; 2 Ziekenhuis Oost-Limburg, Genk, Belgium; 3 Chinese PLA General Hospital, Beijing, China; 4 University College London Hospital, London, United Kingdom; 5 Ospedale S. Gerardo, Universit ` a di Milano Bicocca, Milan, Italy; 6 Malm ¨ o University Hospital, Lund University, Malmo, Sweden; 7 Universit ` a Cattolica del Sacro Cuore, Rome, Italy; 8 University Hospitals K.U. Leuven, Leuven, Belgium; 9 Medical University of Lublin, Lublin, Poland; 10 General Teaching Hospital, 1 st Medical Faculty of the Charles University, Prague, Czech Republic; 11 Imperial College London, Hammersmith Campus, London, United Kingdom Objectives: The tumor marker CA-125 performs moderately well for the diagnosis of adnexal masses as benign or malignant. Using a large multi-center dataset collected by the International Ovarian Tumor Analysis group, we aimed to gain more insight in the performance of CA-125. Methods: We used 3511 patients in the analysis, recruited from 21 centers worldwide. Missing values for CA-125 (n = 853, 24%) were accounted for using multiple imputation. CA-125 distributions for several histological subgroups were obtained using Kernel Density Estimation. Based on these distributions and on subgroup prevalences, subgroup probabilities by CA-125 were derived. Results: We focused on six subgroups: 755 endometrioma and abscesses (as they often have elevated CA-125 levels), 1805 ‘other benign’ tumors, 186 borderline tumors, 136 stage I cancers, 509 higher stage cancers (i.e. stage II-IV and rare malignant tumor), and 120 metastatic tumors. The ‘other benign’ tumors and the higher stage cancers had least overlap in CA-125 distributions. Using a CA-125 level threshold of 30 U/ml to indicate malignancy, 82% of the ‘other benign’ and 90% of the higher stage cancers were correctly classified. The CA-125 distributions of the other groups overlapped substantially and therefore decreased the performance of CA-125 on all tumors (68% specificity and 78% sensitivity). For CA-125 levels below 95 U/ml, the probability of an ‘other benign’ tumor exceeded the probability of a higher stage invasive cancer. Furthermore, most tumors with CA-125 levels between 40 and 180 U/ml were endometriomas or abscesses. Similar results were obtained in pre- and postmenopausal patients separately. Conclusions: Endometriomas, abscesses, borderline tumors, stage I cancers and metastatic tumors constitute a large group of masses (n = 1197, 34%) that negatively influence the overall performance of CA-125 as a predictor of malignancy. OC23.03 Accuracy of diagnostic parameters used for contrasted sonography of ovarian masses A. C. Fleischer 1 , A. Lyshchik 1 , D. Fishman 2 1 Radiology, Vanderbilt University Medical Center, Nashville, TN, USA; 2 Obstetrics and Gynecology, Mt. Sinai Medical Center, New York, NY, USA Objectives: Assess the accuracy of enhancement kinetic parameters in distinguishing benign vs. malignant ovarian masses with contrasted transvaginal sonography. Methods: Transvaginal color Doppler sonography was performed with an iU22 scanner (Philips) prior to intravenous injection of 0.3 ml Definity (Lantheus) microbubble contrast. Q Lab (Philips) software was used to quantitate maximum contrast enhancement (dB), time to peak (sec), half wash-out time (sec), and area under curve (sec -1 ). Results: 39 women with 48 tumors, 11 malignant/borderline and 37 benign were studied. Based on receiver operator curve characteristics, the following sensitivities and specificities and 95% confidence intervals were obtained. (Please see Table 1). Conclusions: Wash-out times and area under the curve (vascular volume) provided the most sensitive and specific parameters for distinguishing benign from malignant ovarian masses. Further studies, possibly multicenter, are warranted. Supported by grant 1R21 CA 125227-05 NIH. OC23.03: Table 1 Criterion Sensitivity 95% CI Specificity 95% CI Max Enhancement > 17.2 dB 90.9 58.7–98.5 97.3 85.8–99.5 Time to peak > 14.1 sec 100 71.3–100.0 59.5 42.1–75.2 1/2 wash-out > 41 sec 100 71.3–100.0 91.9 78.1–98.2 AUC > 786.5 sec-1 100 71.3–100.0 91.9 78.1–98.2 Wash-out time > 121.2 sec 100 71.3–100.0 75.7 58.8–88.2 OC23.04 Which adnexal masses are difficult to classify as benign or malignant with prediction models? B. Van Calster 1 , A. Czekierdowski 2 , R. Fruscio 3 , G. B. Melis 4 , S. Guerriero 4 , L. Savelli 5 , C. Van Holsbeke 6,7 , W. Ombelet 7 , S. Van Huffel 1 , I. Vergote 6 , L. I. Valentin 8 , T. Bourne 6,9 , D. Timmerman 6 1 Katholieke Universiteit Leuven, Leuven, Belgium; 2 Medical University of Lublin, Lublin, Poland; 3 Ospedale S. Gerardo, Universit ` a di Milano Bicocca, Milan, Italy; 4 Ospedale San Giovanni di Dio, University of Cagliari, Cagliari, Italy; 5 University of Bologna, Bologna, Italy; 6 University Hospitals K.U. Leuven, Leuven, Belgium; 7 Ziekenhuis Oost-Limburg, Genk, Belgium; 8 Malm ¨ o University Hospital, Lund University, Malmo, Sweden; 9 Imperial College London, Hammersmith Campus, London, United Kingdom Objectives: The multicenter International Ovarian Tumor Analysis (IOTA) group has developed prediction models to assist clinicians in the preoperative diagnosis of adnexal masses as benign or malignant. These models have been successfully validated on 2757 patients. The aim of this study was to investigate which histologies were difficult to classify using the prediction model LR1. 42 Ultrasound in Obstetrics & Gynecology 2010; 36 (Suppl. 1): 1–51