The establishment and evaluation of a new model for the prediction of prostate cancer

Abstract

To develop a new prostate cancer predictor (PCP) model using the combination of total prostate-specific antigen (tPSA), free PSA (fPSA), and complexed PSA (cPSA). The diagnoses of all the included patients were confirmed pathologically in Daping Hospital between December 1, 2011 and December 1, 2014. There were 54 PCa cases and 579 benign prostatic hyperplasia (BPH) cases with tPSA levels of 2 to 10ng/mL, and 48 PCa cases and 147 BPH cases with tPSA levels of 10 to 20ng/mL. Logistic regression and receiver operating characteristic curve (ROC) analyses were employed to compare the value of PCP (PCP = tPSA / fPSA×√cPSA) with tPSA, fPSA, the ratio of fPSA to tPSA (%fPSA), and cPSA for the differential diagnosis of PCa and BPH. Meanwhile, bootstrapping analysis was used to calculate the distribution and confidence intervals (CIs) for the area under the curve (AUC), and Hosmer-Lemeshow tests were used to calculate P values. When tPSA levels were 2 to 10ng/mL, the AUC of PCP (0.680) was significantly higher than that of tPSA (0.588), fPSA (0.571), %fPSA (0.675), and cPSA (0.613). When the sensitivity for the diagnosis of PCa was 90.7%, the specificity of PCP (22.8%) was higher than that of tPSA (11.1%), fPSA (11.2%), %fPSA (17.4%), and cPSA (15.5%). When tPSA levels were 10 to 20ng/mL, the AUC of PCP (0.686) was significantly higher than that of tPSA (0.603), fPSA (0.643), %fPSA (0.679), and cPSA (0.647). When the sensitivity for the diagnosis of PCa was 91.7%, the specificity of PCP (29.3%) was higher than that of tPSA (10.9%), fPSA (10.2%), %fPSA (23.1%), and cPSA (18.4%). PCP is a novel model for the prediction of PCa; it has more predictive value than tPSA, fPSA, %fPSA, and cPSA when tPSA levels are 2 to 20ng/mL.