Receiver-operating characteristics curves for semen analysis variables: methods for evaluating diagnostic tests of...

This study draws on decision analysis to estimate the utility of seminal variables in predicting pregnancy. The infertile couples selected (n = 709) excluded those who had donor insemination or known female factors (history of pelvic infection, ovulatory disorders, endometriosis, and tubal disease). Diagnostic test properties were calculated with respect to pregnancy for two variables derived from the semen analysis report: sperm density, and the total number of motile sperm in the ejaculate (TMS). Receiver-operating characteristic (ROC) curves based on sensitivity and specificity revealed that none of the proposed thresholds for sperm density or TMS is clearly a best choice. The highest positive predictive values for sperm density (75%) and TMS (75%) were associated with cutoff points at five million sperm per milliliter and five million motile sperm per ejaculate, respectively. The false positive rates when these cutoff points were used were 4.3% (sperm density) and 5.6% (TMS). ROC analysis provides visual and numeric evidence of the quality of a diagnostic test and constitutes a set of procedures for incorporating information from new tests of male gamete function into the diagnostic assessment of infertility.