EUROPEAN UROLOGY 58 (2010)

EUROPEAN UROLOGY 58 (2010) 551 558 available at www.sciencedirect.com journal homepage: www.europeanurology.com Prostate Cancer Prostate Cancer Prevention Trial and European Randomized Study of Screening for Prostate Cancer Risk Calculators: A Performance Comparison in a Contemporary Screened Cohort Vítor Cavadas *, Luís Osório, Francisco Sabell, Frederico Teves, Frederico Branco, Miguel Silva-Ramos Department of Urology, Centro Hospitalar do Porto, Oporto, Portugal Article info Article history: Accepted June 14, 2010 Published online ahead of print on June 22, 2010 Keywords: Diagnosis Prostate biopsy Prostate cancer Risk calculators Abstract Background: Several models can predict the risk of prostate cancer (PCa) on biopsy. Objective: To evaluate the performance of the Prostate Cancer Prevention Trial (PCPT) and European Randomized Study of Screening for Prostate Cancer (ERSPC) risk calculators in detecting PCa in a contemporary screened cohort. Design, setting, and participants: We analyzed prebiopsy characteristics of 525 consecutive screened patients submitted to biopsy, as required by the risk calculators, in one European center between 2006 and 2007. Measurements: Comparisons were done using tests of accuracy (area under the receiver operating characteristic curve [AUC-ROC]), calibration plots, and decision curve analysis. Biopsy predictors were identified by univariate and multivariate logistic regression. Results and limitations: PCa was detected in 35.2% of the subjects. Among predictors included in the calculators, the logarithmic transformations of prostate volume and prostate-specific antigen (PSA), digital rectal examination, previous biopsy status, and age were significantly associated with PCa; transrectal ultrasound abnormalities and family history were not. AUC-ROC for the ERSPC calculator was significantly higher than the PCPT calculator and PSA alone (80.1%, 74.4%, and 64.3%, respectively). Calibration plots showed better performance for the ERSPC calculator; nevertheless, ERSPC may underestimate risk, while PCPT tends to overestimate predictions. Decision curve analysis displayed higher net benefit for the ERSPC calculator; 9% and 23% unnecessary biopsies can be avoided if a threshold probability of 20% and 30%, respectively, is adopted. In contrast, the PCPT model displayed very limited benefit. Our findings apply to a screened European cohort submitted to extended biopsy schemes; consequently, caution should be exerted when considering different populations. Conclusions: The ERSPC risk calculator, by incorporating several risks factors, can aid in the estimation of individual PCa risk and in the decision to perform biopsy. The ERSPC calculator outperformed the PCPT model, which is of very limited value, in a contemporary cohort of screened patients. # 2010 European Association of Urology. Published by Elsevier B.V. All rights reserved. * Corresponding author. Serviço de Urologia, Hospital Geral de Santo António Centro Hospitalar do Porto, Largo Prof. Abel Salazar, 4099-001 Porto, Portugal. Tel. +351 964 423 838; Fax: +351 222 077 507. E-mail address: vcavadas@gmail.com (V. Cavadas). 0302-2838/$ see back matter # 2010 European Association of Urology. Published by Elsevier B.V. All rights reserved. doi:10.1016/j.eururo.2010.06.023

552 EUROPEAN UROLOGY 58 (2010) 551 558 1. Introduction Nomograms and risk calculators have shown better accuracy than prostate-specific antigen (PSA) in predicting prostate cancer (PCa) on biopsy. Nevertheless, constructed models may not apply well externally since widely varying risk levels are generated for similar patients by different models [1]. Two online risk calculators to predict individual risk of a positive biopsy have recently become available [2,3]. The risk of positive biopsy for the Prostate Cancer Prevention Trial (PCPT) calculator depends on PSA, family history, outcome of digital rectal examination (DRE), and prior biopsy [4]. The European Randomized Study of Screening for Prostate Cancer (ERSPC) risk calculator estimates the chance of positive biopsy in previously unscreened (risk indicator 3), previously screened but not biopsied (risk indicator 4), and previously screened and biopsied (risk indicator 5) men, according to PSA, ultrasound-assessed prostate volume (PV), outcome of DRE, outcome of transrectal ultrasound (TRUS), and prior biopsy status [5]. Currently, four studies have externally validated the PCPT risk calculator [6 9] and only one study has externally validated the risk indicator 3 (previously unscreened) of the ERSPC calculator [10]. Risk indicators 4 and 5 have not been validated. A comparison of the PCPT and ERSPC risk calculators using virtual standard index cases pointed out differences in risk estimates between them [11], but no studies have compared their performance in a clinical setting. In this study we analyze and compare their performance in a contemporary cohort of screened patients. 2. Materials and methods We reviewed the records of 593 consecutive patients who underwent TRUS-guided prostate biopsy between January 2006 and December 2007 at our institution. Men with previous diagnosis of PCa or atypical small acinar proliferation of the prostate, PSA <0.5 ng/ml or >50 ng/ml, or PV <10 ml or >150 ml, were excluded (PSA and PV boundaries defined by the admitted input). Previously screened patients (PSA judged as normal in the previous 2 yr, with or without DRE) were referred for prostate biopsy by our department urologists based on new clinical findings (abnormal PSA or DRE); no specific cut-off for PSA was used at that time for biopsy referral. Systematic 10-core and 12-core TRUS-guided prostate biopsies were performed on initial biopsy (I-biopsy) and repeat biopsy (R-biopsy), respectively. 2.1. Statistical analysis PCPT risk calculation was performed using the available formula [4]; for the ERSPC, since no formula is available, risk estimation was obtained by individual data input to the online calculators 4 ( no prior biopsy ) and 5 ( prior negative biopsy ). Continuous variables are reported as the median and range; categorical variables are reported as the number of occurrences and frequency. Mann-Whitney or the Wilcoxon signed-rank test and the Pearson x 2 test were used for statistical comparisons of continuous and categorical variables, respectively. Univariate and multivariate logistic regressions were performed to identify independent predictors of PCa on biopsy. The area under the receiver operating characteristic curve (AUC- ROC) was calculated for both risk calculators and PSA for the entire cohort as well as for the subsets of I-biopsy and R-biopsy. Differences in predictive accuracy estimates were tested for statistical significance with the Mantel-Haenszel test. Performance characteristics of the risk calculators were examined by calibration plots, where the x-axis represents the predicted probability and the y-axis represents the actual observed proportion of positive biopsy. Calibration was assessed by grouping patients into 12 groups (each comprising 43 or 44 patients) with respect to their predicted probabilities and then comparing the mean of each group with the observed proportion of men with cancer. The sum of squares of the residuals (SSR) was used to assess the deviation from perfect prediction (the 458 line). For the subsets of I-biopsy and R-biopsy, respectively eight and four groups were used (to also incorporate 43 or 44 patients). Finally, decision curve analysis [12] was used to explore the clinical effects of the calculators. This method estimates a net benefit for prediction models by summing the benefits (true positives) and subtracting the harms (false positives). As the value of a true positive (early detection of cancer) may be different from the disadvantages resulting from a false positive (avoidable biopsy), the net benefit weighs true and false positives differently by using the threshold probability at which a patient would opt for biopsy. The best model displays the higher net benefits throughout a range of threshold probabilities. As suggested by Steyerberg and Vickers [13], the large majority of patients (and we suspect referring doctors likewise) would have a threshold probability to undergo prostate biopsy between 10% and 40%, so this range was chosen for analysis. All tests were two-sided with a significance level set at 0.05. Statistical analysis was performed using SPSS v.16.0 (SPSS Inc, Chicago, IL, USA), and MedCalc v.11.1.1.0 (MedCalc Software bvba, Mariakerke, Belgium). 3. Results Among the 593 records surveyed, we identified 545 patients who met our criteria for analysis; 20 patients were excluded for missing information about family history (n = 17), and/ or DRE results (n = 5). Therefore, 525 different patients were included in this analysis; their characteristics are shown in Table 1. PCa was diagnosed in 185 patients (35.2%), 94 being high grade (Gleason 7). The detection rate was significantly lower on R-biopsy compared to I-biopsy (21.6% and 42.1%, respectively, p < 0.001). Age and PSA were significantly higher for positive-biopsy patients ( p < 0.001, for both). In contrast, PV was significantly lower in PCa-diagnosed patients ( p < 0.001). On univariate logistic regression, neither TRUS findings nor family history were significant predictors of a positive biopsy. On univariate and multivariate logistic regression, the logarithmic transformations of PV and PSA, DRE results, previous biopsy status, and age were significant predictors of PCa (Table 2). Overall, PCPT- and ERSPC-calculated prebiopsy risks of harboring cancer were significantly different: median risk 51% (first quartile [Q1]: 41%, third quartile [Q3]: 65%, range: 18 92%) for the PCPT risk calculator, compared with median risk 22% (Q1: 13%, Q3: 35%, range: 3 97%) for the ERSPC risk calculator ( p < 0.001, Wilcoxon signed-rank test). PCPTcalculated risk was significantly higher in the positive-biopsy

EUROPEAN UROLOGY 58 (2010) 551 558 553 Table 1 Clinical characteristics of the study cohort Cohort Negative biopsy Positive biopsy p value Race Caucasian, No. (%) 525 (100) 340 (64.8) 185 (35.2) N/A Age, median (range), yr 67 (42 89) 65 (42 89) 69 (42 88) <0.001 * Age, No. (%) <55 yr 35 (6.7) 27 (7.9) 8 (4.3) <0.001 ** 55 59 yr 80 (15.2) 62 (18.2) 18 (9.7) 60 64 yr 96 (18.3) 75 (22.1) 21 (11.4) 65 69 yr 124 (23.6) 75 (22.1) 49 (26.5) 70 74 yr 103 (19.6) 59 (17.4) 44 (23.8) 75 yr 87 (16.6) 42 (12.4) 45 (24.3) PSA level, median (range), ng/ml 8.14 (0.50 50) 7.50 (0.50 38.65) 9.60 (1.37 50) <0.001 * PSA level, No. (%) <2.5 ng/ml 15 (2.9) 13 (3.8) 2 (1.1) <0.001 ** 2.5 3.99 ng/ml 20 (3.8) 16 (4.7) 4 (2.2) 4 9.99 ng/ml 297 (56.6) 208 (61.2) 89 (48.1) 10 19.99 ng/ml 127 (24.2) 87 (25.6) 40 (21.6) 20 ng/ml 66 (12.6) 16 (4.7) 50 (27.0) DRE, No. (%) Normal 350 (66.7) 266 (78.2) 84 (45.4) <0.001 ** Abnormal 175 (33.3) 74 (21.8) 101 (54.6) TRUS, No. (%) Normal 413 (78.7) 272 (80.0) 141 (76.2) 0.312 ** Abnormal 112 (21.3) 68 (20.0) 44 (23.8) Prostate volume, median (range), ml 55 (10 150) 61 (22 150) 40 (10 150) <0.001 * Prostate volume, No. (%) <30 ml 66 (12.6) 20 (5.9) 46 (24.9) <0.001 ** 30 59 ml 217 (41.3) 127 (37.4) 90 (48.6) 60 89 ml 164 (31.2) 129 (37.9) 35 (18.9) 90 119 ml 48 (9.1) 40 (11.8) 8 (4.3) 120 ml 30 (5.7) 24 (7.1) 6 (3.2) Family history, No. (%) Negative 504 (96) 326 (95.9) 178 (96.0) 0.852 ** Positive 21 (4) 14 (4.1) 7 (4.0) Previous biopsy, No. (%) No 349 (66.5) 202 (59.4) 147 (79.5) <0.001 ** Yes 176 (33.5) 138 (40.6) 38 (20.5) PSA = prostate-specific antigen; DRE = digital rectal examination; TRUS = transrectal ultrasound; N/A = not applicable. * Mann-Whitney test. ** x 2 test. group of patients (median 65%, Q1: 48%, Q3: 81%, range: 26 92%;) compared with those with no cancer (median 47%, Q1: 39%, Q3: 59%, range: 18 90% ( p < 0.001, Mann-Whitney test)). Likewise, ERSPC-calculated risk was higher for those with a positive biopsy (median 39%, Q1: 22.5%, Q3: 62.5%, range: 6 97%) compared with those with no cancer (median 17%, Q1: 11%, Q3: 25%, range: 3 72% ( p < 0.001, Mann- Whitney test)). Table 2 Univariate and multivariate logistic regression for biopsy outcome predictors Variable Univariate logistic regression Multivariate logistic regression OR 95% CI p value Predictive accuracy, % OR 95% CI p value Predictive accuracy, % Ln, volume 0.205 0.136 0.309 <0.001 69.9 0.144 0.086 0.242 <0.001 78.3 Ln, PSA 2.652 1.977 3.559 <0.001 70.9 3.254 2.219 4.771 <0.001 DRE 4.322 2.934 6.367 <0.001 69.9 2.848 1.789 4.535 <0.001 Prior biopsy 0.378 0.249 0.574 <0.001 64.8 0.416 0.255 0.677 <0.001 Age, yr 1.067 1.042 1.093 <0.001 67.6 1.038 1.007 1.071 0.015 TRUS 1.248 0.812 1.919 0.312 64.8 1.000 0.579 1.727 0.999 N/A Family history 0.916 0.363 2.310 0.852 64.8 1.124 0.401 3.148 0.824 CI = confidence interval; DRE = digital rectal examination; PSA = prostate-specific antigen; Ln = natural logarithm; N/A = not applicable; OR = odds ratio; TRUS = transrectal ultrasound.

554 EUROPEAN UROLOGY 58 (2010) 551 558 Fig. 1 Receiver operating characteristic curves for the European Randomized Study of Screening for Prostate Cancer (ERSPC) and Prostate Cancer Prevention Trial (PCPT) risk calculators and prostate-specific antigen (PSA) alone for (A) the entire cohort, (B) initial biopsy, and (C) repeat biopsy. Table 3 Comparison of areas under the receiver operating characteristic curve for European Randomized Study of Screening for Prostate Cancer (ERSPC) and Prostate Cancer Prevention Trial (PCPT) risk calculators and prostate-specific antigen (PSA) alone AUC 95% CI p value ERSPC PCPT PSA Cohort ERSPC 0.801 0.764 0.834 0.012 <0.001 PCPT 0.744 0.705 0.781 0.012 <0.001 PSA 0.643 0.600 0.684 <0.001 <0.001 Initial biopsy ERSPC 0.812 0.767 0.852 0.011 <0.001 PCPT 0.752 0.704 0.797 0.011 0.004 PSA 0.696 0.645 0.744 <0.001 0.004 Repeat biopsy ERSPC 0.742 0.671 0.805 0.216 0.036 PCPT 0.657 0.582 0.727 0.216 0.031 PSA 0.562 0.486 0.637 0.036 0.031 AUC = area under the curve; CI = confidence interval. A significantly higher AUC was observed for the PCPT and ERSPC risk calculators compared with PSA alone, with the ERSPC achieving the highest predictive accuracy (Fig. 1 and Table 3). Analyzing the calibration plots for the entire cohort, ERSPC tends to underestimate risk while PCPT overestimates risk of positive biopsy, with ERSPC showing overall better calibration (SSR of 0.14 for the ERSPC vs 0.56 for PCPT calculator). The ERSPC risk calculator shows very good calibration for predicted risks <35%; similar results are seen for the subcohort of I-biopsy. For the R-biopsy subset of patients, only the ERSPC model retains good calibration and only for predicted risks <19% (Fig. 2). The ERSPC calculator has the highest net benefit in the defined range of interest (10 40% probability), consistently outperforming the PCPT model and the Treat all strategy for probabilities above 14% (Fig. 3). Table 4 shows the number of cancers missed and the reduction in biopsies according to threshold probability for both calculators, along with the missed cancers if a PSAbased decision was undertaken for the same amount of reduction in biopsies. Fig. 2 Calibration plots depicting the agreement between predicted and observed probabilities of positive biopsy for the European Randomized Study of Screening for Prostate Cancer (ERSPC) and Prostate Cancer Prevention Trial (PCPT) risk calculators for (A) the entire cohort, (B) initial biopsy, and (C) repeat biopsy.

Fig. 3 Decision curve analysis for positive biopsy prediction by European Randomized Study of Screening for Prostate Cancer (ERSPC) and Prostate Cancer Prevention Trial (PCPT) risk calculators for (A) the entire cohort, (B) initial biopsy, and (C) repeat biopsy, in the range of 10 40% threshold probability. The bottom table displays net benefit and reduction in avoidable biopsies for each of the two risk calculators compared with the Treat all strategy of performing biopsy on every patient in the cohort for different threshold probabilities in the same range. EUROPEAN UROLOGY 58 (2010) 551 558 555

556 EUROPEAN UROLOGY 58 (2010) 551 558 Table 4 Number of total and high-grade (defined as Gleason I7) cancers missed and reduction in biopsies, according to threshold probability in the range of 10 40% for the risk calculators, and the missed cancers if a prostate-specific antigen (PSA) based decision was made for the same proportion of reduction in biopsies Calculator-based decision Biopsies spared, No. (%) PSA-based decision Threshold probability, % Cancers missed, No. (%) High grade missed, No. (%) Cancers missed, No. (%) High grade missed, No. (%) 10 ERSPC 7 (3.8) 3 (3.2) 81 (15.4) 17 (9.2) 5 (5.3) PCPT 0 (0) 0 (0) 0 (0) N/A N/A 15 ERSPC 21 (11.4) 7 (7.4) 160 (30.5) 43 (23.2) 14 (14.9) PCPT 0 (0) 0 (0) 0 (0) N/A N/A 20 ERSPC 36 (19.5) 11 (11.7) 229 (43.6) 63 (34.1) 21 (22.3) PCPT 0 (0) 0 (0) 4 (0.8) 0 (0) 0 (0) 25 ERSPC 52 (28.1) 17 (18.1) 303 (57.7) 85 (45.9) 33 (35.1) PCPT 0 (0) 0 (0) 4 (0.8) 0 (0) 0 (0) 30 ERSPC 66 (35.7) 23 (24.5) 343 (65.3) 96 (51.9) 39 (41.5) PCPT 1 (0.5) 0 (0) 14 (2.7) 2 (1.1) 1 (1.1) 35 ERSPC 84 (45.4) 29 (30.9) 390 (74.3) 110 (59.5) 46 (48.9) PCPT 5 (2.7) 2 (2.1) 50 (9.5) 7 (3.8) 2 (2.1) 40 ERSPC 93 (50.2) 35 (37.2) 411 (78.3) 117 (63.2) 48 (51.1) PCPT 19 (10.3) 6 (6.4) 112 (21.3) 26 (14.1) 10 (10.6) ERSPC = European Randomized Study of Screening for Prostate Cancer; PCPT = Prostate Cancer Prevention Trial; N/A = not applicable. 4. Discussion Statistical and computational models have been developed to predict more accurately an individual s risk of harboring PCa at biopsy, mostly because PSA and PSA-related measurements have proved to be limited in this task. In a recent review including 23 studies examining 36 predictive models, 14 direct comparisons between model and PSA accuracies (AUC-ROC) showed a benefit from nomograms or artificial neural networks over PSA alone varying between 2% and 26% [1]. The original study from which the PCPT calculator was derived reported an AUC-ROC of 70.2% for the calculator, slightly higher than the 67.8% reported for PSA alone [4]. Four studies externally validating this calculator have been published since [6 9], reporting AUCs between 56.9% and 69.1% for the PCPT model, increasing predictive accuracy between 1.5% and 4.8% compared with PSA alone (Table 5). No external validations have been published for the ERSPC model predicting risk in screened patients (our cohort being studied); results of an external validation of the ERSPC risk calculator for previously unscreened men showed good discrimination with an AUC of 77% (comparable to the 79% for the original cohort) [10]. Our study shows that, in a screened population, both ERSPC and PCPT risk calculators outperform PSA alone in predicting PCa on biopsy. The AUC-ROC of 64.3% for PSA alone is significantly increased to 74.4% using the PCPT model and 80.1% by the ERSPC calculator. Furthermore, the ERSPC risk calculator was found to have significantly higher predictive accuracy than the PCPT model, except for the R-biopsy subcohort where differences failed to achieve statistical significance. Another important issue when evaluating prediction tools concerns calibration. Risk underestimation would be expected from both calculators since a significantly higher number of cores was taken in our study (10 of 12 cores vs 6 in ERSPC and >80% of PCPT biopsies) [11]. Nevertheless, we have shown good calibration for the ERSPC calculator, especially for predicted risks <35%; above this threshold there is, indeed, a tendency to underestimate risk. Calibration is worse for the PCPT calculator, which surprisingly overestimates risk. Neither of these calculators displays good calibration for the subcohort of R-biopsy patients. Decision curve analysis has shown the superiority of the ERSPC model in the range of interest (10 40% threshold probability, as suggested by Steyerberg and Vickers [13]) and a very limited value for the use of the PCPT calculator. Table 5 Studies evaluating Prostate Cancer Prevention Trial (PCPT) risk calculator performance versus prostate-specific antigen (PSA) alone Cohort, No. Cancer incidence, % AUC for PCPT, % AUC for PSA, % PCPT vs PSA Increase in AUC, % p value Thompson et al. [4] 5519 21.9 70.2 67.8 2.4 N/A Parekh et al. [6] 446 33.2 65.5 64.0 1.5 >0.05 Hernandez et al. [7] 1108 35.6 66.7 61.9 4.8 <0.001 Eyre et al. [8] 645 43.4 69.1 65.5 3.6 0.009 Nguyen et al. [9] 4515 41.2 56.9 52.5 4.4 N/A Current study 525 35.2 74.4 64.3 10.1 <0.001 AUC = area under the curve.

EUROPEAN UROLOGY 58 (2010) 551 558 557 For example, at a threshold probability of 20%, basing biopsy decision on the ERSPC calculator is equivalent to a strategy that reduced the number of biopsies by 9% but which missed no cancers; increasing the threshold to 30%, then, the reduction in avoidable biopsies would be 23%. Of course, when a threshold of 20% or 30% is set, we are assuming by definition that we could miss as much as 20% or 30% of cancers, respectively. But Table 4 shows that basing decisions on PSA alone to spare the same number of biopsies of a specific threshold probability for the ERSPC model would roughly double the total number of missed tumors, as well of missed high-grade cancers. Taking all of the above into consideration, the ERSPC emerges as the best predictive model. Many reasons could account for its superiority in our study cohort. First, our study population being European in origin may resemble better the one used to develop the ERSPC calculator. Second, only the ERSPC model takes into account PV, which was found to be highly predictive of PCa in our cohort. Third, family history did not reach predictive value in our cohort and was also discharged in the ERSPC but considered for the PCPT calculator. Fourth, cancer detection rates according to PSA level in our study cohort were more similar to the ones reported by the ERSPC: 96.7%, 74.8%, and 24.7% of the total number of cancers were detected in the PSA range >4 ng/ml in our study, ERSPC, and PCPT, respectively [11]. Level of complexity is also a parameter analyzed in predictive tools. Both online risk calculators are user friendly and require few input variables. Nevertheless, some might point out as a major drawback the need for prebiopsy TRUS (to evaluate PV and the existence of visible lesions) in order to estimate risk and aid decision when using the ERSPC model. This could be overcome by estimating PV when performing DRE; then discuss the possible risks and agree on a threshold to biopsy; and finally, decide, when performing TRUS, whether or not to biopsy immediately. Some limitations are acknowledged in our study. First, this is a retrospective study. Second, our cohort comprised only Caucasian patients; nevertheless, race is not a predictive variable in both calculators. Third, family history of PCa was reported in only 4% of our patients, which could account for its lack of predictive value. Fourth, the risk calculators are mainly based on the results of sextant biopsies, while we adopted 10- and 12-core schemes for I-biopsy and R-biopsy, respectively. From another standpoint, our study provides new and invaluable information on the clinical applicability of these online risk calculators. It constitutes the first study comparing their performance outside a clinical trial setting and provides data on discrimination, calibration, and decision analysis to support their external validity. Predictive models do not replace clinical judgment or patient preference, but may be useful in providing a starting point when deciding to perform a prostate biopsy. As a consequence of the improved predictive accuracy, unnecessary biopsies can be effectively avoided without compromising early detection of PCa. 5. Conclusions The ERSPC risk calculator, by incorporating several risks factors, can aid in estimating individual risk of PCa and deciding the need for prostate biopsy in our daily practice, especially on initial biopsy, although risk could be underestimated mainly due to higher sampling schemes in current practice. In contrast, the PCPT calculator has very limited value when applied to biopsy decision-making in a screened population. Author contributions: Vítor Cavadas had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Cavadas, Silva-Ramos. Acquisition of data: Cavadas, Sabell, Teves. Analysis and interpretation of data: Cavadas, Osório, Silva-Ramos. Drafting of the manuscript: Cavadas, Teves, Branco. Critical revision of the manuscript for important intellectual content: Osório, Silva-Ramos. Statistical analysis: Cavadas, Sabell, Branco. Obtaining funding: None. Administrative, technical, or material support: None. Supervision: Silva-Ramos. Other (specify): None. Financial disclosures: I certify that all conflicts of interest, including specific financial interests and relationships and affiliations relevant to the subject matter or materials discussed in the manuscript (eg, employment/affiliation, grants or funding, consultancies, honoraria, stock ownership or options, expert testimony, royalties, or patents filed, received, or pending), are the following: None. Funding/Support and role of the sponsor: None. References [1] Schröder F, Kattan MW. The comparability of models for predicting the risk of a positive prostate biopsy with prostatespecific antigen alone: a systematic review. Eur Urol 2008;54: 274 90. [2] Risk of biopsy-detectable prostate cancer. University of Texas Health Science Center San Antonio Web site. http://deb.uthsca.edu/urorisk Calc/Pages/uroriskcalc.jsp. Accessed November 2009. [3] Background information. SWOP Prostate Cancer Research Foundation Web site. http://www.prostatecancer-riskcalculator.com/via.html. Accessed November 2009. [4] Thompson IM, Ankerst DP, Chi C, et al. Assessing prostate cancer risk: results from the Prostate Cancer Prevention Trial. J Natl Cancer Inst 2006;98:529 34. [5] Roobol MJ, Steyerberg EW, Kranse R, et al. A risk-based strategy improves prostate-specific antigen driven detection of prostate cancer. Eur Urol 2010;57:79 85. [6] Parekh DJ, Ankerst DP, Higgins BA, et al. External validation of the Prostate Cancer Prevention Trial risk calculator in a screened population. Urology 2006;68:1152 5. [7] Hernandez DJ, Han M, Humphreys EB, et al. Predicting the outcome of prostate biopsy: comparison of a novel regression-based model, the prostate cancer risk calculator, and prostate-specific antigen level alone. BJU Int 2009;103:609 14. [8] Eyre SJ, Ankerst DP, Wei JT, et al. Validation in a multiple urology practice cohort of the Prostate Cancer Prevention Trial

558 EUROPEAN UROLOGY 58 (2010) 551 558 calculator for predicting prostate cancer detection. J Urol 2009; 182:2653 8. [9] Nguyen CT, Yu C, Moussa A, Kattan MW, Jones JS. Performance of Prostate Cancer Prevention Trial risk calculator in a contemporary cohort screened for prostate cancer and diagnosed by extended prostate biopsy. J Urol 2010;183:529 33. [10] Roobol MJ, Kranse R, Maattanen L, Schröder FH. External validation of the Riskindicator 1. ERSPC Rotterdam, Helsinki and Tampere. Eur Urol Suppl 2009;8:192. [11] Van den Bergh RCN, Roobol MJ, Wolters T, van Leeuwen PJ, Schröder FH. The Prostate Cancer Prevention Trial and European Randomized Study of Screening for Prostate Cancer risk calculators indicating a positive prostate biopsy: a comparison. BJU Int 2008;102:1068 73. [12] Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making 2006;26:565 74. [13] Steyerberg EW, Vickers EJ. Decision curve analysis: a discussion. Med Decis Making 2008;28:146 9.