The effect of cancer prevalence on the use of different cut-off levels and the performance of mathematical models to distinguish benign from malignant adnexal masses.