PREDICTORS OF PROSTATE CANCER EVALUATED BY RECEIVER OPERATING CHARACTERISTICS PARTIAL AREA INDEX: A PROSPECTIVE INSTITUTIONAL STUDY

Publisher Website
Google Scholar
Abstract
Purpose: To our knowledge we introduce the ROC partial area under the curve (AUC) index as a method of evaluating the discriminative power of different prostate cancer predictors. Peripheral zone volume and peripheral zone prostate specific antigen (PSA) density are introduced as potential predictors and compared with other known predictors of prostate cancer. Materials and Methods: During 1999, 220 consecutive patients with suspected early prostate cancer were examined using total PSA, free PSA, total prostate volume, transition zone volume and transrectal ultrasonography guided sextant biopsy of the prostate. The free-to-total PSA ratio, PSA density, transition zone PSA density, peripheral zone volume and peripheral zone PSA density were calculated. Usually total AUC is used to evaluate the discriminative power of different parameters. In this study parameters were evaluated by the ROC partial area index, which includes only the AUC in highly sensitivity parts of the ROC curve. Explorative analysis using logistic regression analysis was performed to investigate the ability of combinations of parameters to predict cancer. Results: Of the 220 patients 75 were diagnosed with cancer. In the subgroup of 160 patients with PSA less than 10 μg/l 44 had cancer. Transition zone PSA density and PSA density had significant discriminative power in the total group, while none of the parameters were discriminative in the subgroup of patients. Conclusions: When high sensitivity is demanded, the ROC partial area index seems to be meaningful for evaluating the discriminative power of potential predictors. In our study none of the evaluated parameters had discriminative power in patients with PSA less than 10 μg/l, while transition zone PSA density and PSA density showed discriminative power in the total group of patients.

This publication has 15 references indexed in Scilit: