Effect of cancer prevalence on the use of risk-assessment cut-off levels and the performance of mathematical models to distinguish malignant from benign adnexal masses.