[Estimation of predictive prostate cancer probability with logistic regression equation]

Low specificity of PSA for early diagnosis of prostate cancer (PC) is the cause of search for new tests. The aim of our study was to develop the logistic regression model and estimate the value of the regression equation as a diagnostic tool for prostate cancer detection. A total of 518 male patients aged 47-83 years (mean 65.5 +/- 6.5 years) who had undergone TRUS-guided 12-core systematic transrectal prostate biopsy were included in the study. PC detection rate in our study was 43.8%. The logistic regression model with PC detection as a response and age, prostate volume, PSA, induration on DRE and hypoechoic lesion on TRUS as effects was designed. With regression equation PC probability for any patient was calculated. The regression equation was tested as a PC diagnostic tool. As the combination of model effects (chi-square 87.9; p < 0.0001; R2 = 0.124) any of the effects independently may predict prostate cancer detection. The obtained regression equation is: P(Pca) = 1/{1 + 2.718(-[-4.029 + (0.068 x AGE) + (0.022 x PSA) + (-0013 x PROSTATE VOLUME) + (0.375 x DRE) + (0.254 x TRUS)])} Accuracy (area under ROC-curve) of our regression equation as a PC detection diagnostic tool was 73%. Probability cutoff of 0.26 leads to sensitivity of 90% and specificity of 30% and eliminates 12% of unnecessary biopsies in patients with benign prostate diseases (chi-square 10.91; p < 0.0001). Thus, the obtained logistic regression equation may be used as a PC diagnostic tool in the suspects. Multicenter trial may improve regression equation diagnostic performance.