Ovarian cancer prediction in adnexal masses using ultrasound based logistic regression models: a temporal and external validation study by the IOTA group

OBJECTIVE: To temporally and externally validate the diagnostic performance of two logistic regression models containing clinical and ultrasound variables to estimate the risk of malignancy in adnexal masses, and to compare them with the subjective interpretation of ultrasound findings by an experienced ultrasound examiner ("subjective assessment"). METHODS: Patients with adnexal masses from 19 centers underwent a standardized transvaginal ultrasound examination by a gynecologist or radiologist specialized in ultrasonography. The examiner prospectively collected information on clinical and ultrasound variables, and classified each mass as benign or malignant on the basis of subjective evaluation of ultrasound findings. The gold standard was the histology of the mass with local clinicians deciding whether to operate on the basis of ultrasound results and clinical picture. The models' ability to discriminate between malignant and benign masses was assessed, as well as the accuracy of the risk estimates. RESULTS: 1938 patients were included; 1396 benign, 373 primary invasive, 111 borderline malignant, and 58 metastatic tumors. On external validation (997 patients from 12 centers), the area under the receiver operating characteristic curve (AUC) for a model containing 12 predictors (LR1) was 0.956, for a reduced model with six predictors (LR2) 0.949, and for subjective assessment 0.949. Subjective assessment gave a positive likelihood ratio of 11.0 and a negative likelihood ratio of 0.14. The corresponding likelihood ratios for a previously derived probability threshold (0.1) were 6.84 and 0.09 for LR1, and 6.36, and 0.10 for LR2. On temporal validation (941 patients from seven centers) the AUCs were 0.945 (LR1), 0.918 (LR2), and 0.959 (subjective assessment). CONCLUSIONS: Both models provide excellent discrimination between benign and malignant masses. Because the models provide an objective and fairly accurate risk estimation, they may improve the management of women with suspected ovarian pathology. Copyright (c) 2010 ISUOG. Published by John Wiley & Sons, Ltd.