PURPOSE: We hypothesized that the outcome of repeat biopsy could be accurately predicted. We tested this hypothesis in a contemporary cohort from 3 centers. MATERIALS AND METHODS: The principal cohort of 1,082 men from Hamburg, Germany was used for nomogram development as well as for internal 200 bootstrap validation in 721 and external validation in 361. Two additional external validation cohorts, including 87 men from Milan, Italy and 142 from Seattle, Washington, were also used. Predictors of prostate cancer on repeat biopsy were patient age, digital rectal examination, prostate specific antigen, percent free prostate specific antigen, number of previous negative biopsy sessions and sampling density. Multivariate logistic regression models were used to develop the nomograms. RESULTS: The mean number of previous negative biopsies was 1.5 (range 1 to 6) and the mean number of cores at final repeat biopsy was 11.1 (range 10 to 24). Of the men 370 (30.2%) had prostate cancer. On multivariate analyses all predictors were statistically significant (p <or =0.028). After internal validation the nomogram was 76% accurate. External validation showed 74% (Hamburg), 78% (Milan) and 68% (Seattle) accuracy. CONCLUSIONS: Relative to the previous nomograms (10 predictors or 71% accuracy) our tool relies on fewer variables (6) and shows superior accuracy in European men. Accuracy in American men is substantially lower. Racial, clinical and biochemical differences may explain the observed discrepancy in predictive accuracy.
PURPOSE: We hypothesized that the outcome of repeat biopsy could be accurately predicted. We tested this hypothesis in a contemporary cohort from 3 centers. MATERIALS AND METHODS: The principal cohort of 1,082 men from Hamburg, Germany was used for nomogram development as well as for internal 200 bootstrap validation in 721 and external validation in 361. Two additional external validation cohorts, including 87 men from Milan, Italy and 142 from Seattle, Washington, were also used. Predictors of prostate cancer on repeat biopsy were patient age, digital rectal examination, prostate specific antigen, percent free prostate specific antigen, number of previous negative biopsy sessions and sampling density. Multivariate logistic regression models were used to develop the nomograms. RESULTS: The mean number of previous negative biopsies was 1.5 (range 1 to 6) and the mean number of cores at final repeat biopsy was 11.1 (range 10 to 24). Of the men 370 (30.2%) had prostate cancer. On multivariate analyses all predictors were statistically significant (p <or =0.028). After internal validation the nomogram was 76% accurate. External validation showed 74% (Hamburg), 78% (Milan) and 68% (Seattle) accuracy. CONCLUSIONS: Relative to the previous nomograms (10 predictors or 71% accuracy) our tool relies on fewer variables (6) and shows superior accuracy in European men. Accuracy in American men is substantially lower. Racial, clinical and biochemical differences may explain the observed discrepancy in predictive accuracy.