COMPARISON OF 3 ASSAYS FOR TOTAL SERUM PROSTATE-SPECIFIC ANTIGEN AND PERCENTAGE OF FREE PROSTATE-SPECIFIC ANTIGEN IN PREDICTING PROSTATE HISTOLOGY

Objectives. To determine the statistical performance of three differen t assays for prostate specific antigen (PSA) and the percentage of fre e PSA with respect to the differentiation of histologic benign prostat ic hyperplasia (BPH) and prostate cancer in men who underwent surgical removal of prostate tissue. Methods. Serum of 86 men scheduled for pr ostate surgery (transurethral resection of the prostate [TURP], simple open prostatectomy, radical prostatectomy, cystoprostatectomy) was fr ozen and subjected to measurement in batches using three different ass ays for total PSA (Hybritech Tandem-E, Abbott IMx, Tosoh AIA-600) and free PSA by the Hybritech method after a single freeze-thaw cycle. The histologic diagnosis of the removed tissue (35 BPH and 51 cancer) was used as a ''gold standard'' for classification of disease status. Sen sitivity, specificity, positive and negative predictive values, and di agnostic efficiency were calculated for the three total PSA assays and the free/total PSA ratios for the entire cohort and subsets. Receiver -operating characteristic (ROC) curve analysis was used to compare the performance of the assays and ratios. Results. Mean and median total PSA values differed slightly between the three assays for all patients , and for those with BPH and cancer, but this difference was not signi ficant. Because of a considerable overlap, the differences between the mean PSA values for men with BPH and prostate cancer were not signifi cant. At a cutpoint of 4.0 ng/mL, sensitivity with respect to the diff erentiation between BPH and prostate cancer was 68.6% for all three to tal PSA assays; the respective AUCs (0.613-0.625) were not significant ly different. While the performance of the free/total PSA ratios was s uperior, the differences were only significant when subsets of patient s were considered with a total PSA between 4 and 10 ng/mL or 4 and 15 ng/mL (AUCs 0.789-0.816). Likewise, sensitivity, specificity, and diag nostic efficiency was better in these subsets of patients. Conclusions . In this study in which a ''gold standard'' based on histologic analy sis of the entire (or large part of) the prostate gland was used to cl assify disease status, the three assays for total serum PSA (Hybritech Tandem-E, Abbott IMx, and Tosoh AIA-600) performed very similarly wit h identical sensitivities (at a cutpoint of 4.0 ng/mL) and comparable AUCs with respect to the differentiation of men with histologic BPH an d prostate cancer. The ratios of free/total PSA calculated as free PSA by the Hybritech manual immunoradiometric assay (IRMA) method over al l three total PSA assays, performed marginally better in the entire pa tient population. However, in the subsets of patients with a PSA of 4- 10 ng/mL and 4-15 ng/mL, all three ratios performed significantly bett er than the three total PSA assays. The proper choice of a cutpoint fo r the ratio (15%, 17%, 19%, or 21%) depends on the desirability of max imizing either sensitivity or specificity while optimizing diagnostic efficiency. Copyright 1996 by Elsevier Science Inc.