Improving multivariable prostate cancer risk assessment using the Prostate Health Index

Transcription

1 Improving multivariable prostate cancer risk assessment using the Prostate Health Index Robert W. Foley*, Laura Gorman*, Neda Sharifi, Keefe Murphy, Helen Moore, lexandra V. Tuzova**, ntoinette S. Perry**, T. Brendan Murphy, Dara J. Lundon* and R. William G. Watson* *Conway Institute of Biomolecular and Biomedical Research, University College Dublin, UCD School of Medicine and Medical Science, University College Dublin, Department of Biochemistry, Beaumont Hospital, UCD School of Mathematical Sciences, University College Dublin, Insight Centre for Data nalytics, University College Dublin, **Prostate Molecular Oncology, Institute of Molecular Medicine, Trinity College Dublin, and Department of Urology, Mater Misericordiae University Hospital, Dublin, Ireland Objectives To analyse the clinical utility of a prediction model incorporating both clinical information and a novel biomarker, p2ps, in order to inform the decision for prostate biopsy in an Irish cohort of men referred for prostate cancer assessment. Patients and Methods Serum isolated from 250 men from three tertiary referral centres with pre-biopsy blood draws was analysed for total prostate-specific antigen (PS), free PS (fps) and p2ps. From this, the Prostate Health Index (PHI) score was calculated (PHI = (p2ps/fps)* tps). The men s clinical information was used to derive their risk according to the Prostate Cancer Prevention Trial (PCPT) risk model. Two clinical prediction models were created via multivariable regression consisting of age, family history, abnormality on digital rectal examination, previous negative biopsy and either PS or PHI score, respectively. Calibration plots, receiver-operating characteristic (ROC) curves and decision curves were generated to assess the performance of the three models. Results The PS model and PHI model were both well calibrated in this cohort, with the PHI model showing the best correlation between predicted probabilities and actual outcome. The areas under the ROC curve for the PHI model, PS model and PCPT model were 0.77, 0.71 and 0.69, respectively, for the prediction of prostate cancer (PCa) and 0.79, 0.72 and 0.72, respectively, for the prediction of high grade PCa. Decision-curve analysis showed a superior net benefit of the PHI model over both the PS model and the PCPT risk model in the diagnosis of PCa and high grade PCa over the entire range of risk probabilities. Conclusion logical and standardized approach to the use of clinical risk factors can allow more accurate risk stratification of men under investigation for PCa. The measurement of p2ps and the integration of this biomarker into a clinical prediction model can further increase the accuracy of risk stratification, helping to better inform the decision for prostate biopsy in a referral population. Keywords prostatic neoplasm, biopsy, predictive models, biomarkers, p2ps, Prostate Health Index Introduction Prostate cancer (PCa) is the most common solid-organ malignancy amongst men in Ireland and is second only to lung cancer as the single largest cause of cancer-specific mortality [1]. Ireland has one of the highest incidences of PCa in Europe and, according to published incidence rates from 2012, was >50% higher than the European Union average [1]. Death from cancer almost invariably results from disseminated disease; therefore, successful resection of a tumour before it spreads should prevent cancer-specific mortality. Hence, early detection of cancer remains the most promising strategy for cancer control. The gold standard for the diagnosis of PCa is a prostate biopsy. The decision on who to send for this procedure is a difficult one. The clinical judgment and recommendation to proceed with a prostate biopsy hinges on assessment of risk BJU International 2015 BJU International doi: /bju BJU Int 2016; 117: Published by John Wiley & Sons Ltd. wileyonlinelibrary.com

2 R. W. Foley et al. factors such as PS, DRE, a positive family history of PCa, age and previous prostate biopsy, as well as the patient s life expectancy, psychological status and the patient s wishes. Ireland uses a referral system for assessment of prostate biopsy based on PS level and DRE findings [2]; however, prostate biopsy carries with it the possibility of significant morbidity [3]. It is therefore important to identify those at greatest risk of significant disease, so that access to health resources may be allocated appropriately and all patients may be counselled more comprehensively before this procedure. Risk stratification tools exist and indeed their systematic use is supported by recent guidelines and consensus statements [4,5]. Through the systematic use of patient risk factors, in the form of PCa clinical prediction models, risk stratification can be standardized and used to guide clinical decisionmaking [6]. Methods to improve risk stratification include firstly, the improved utilisation of the clinical information available to the urologist and, secondly, the measurement of novel biomarkers of disease that are more specific for PCa or aggressive PCa. In the present study we combine both of these approaches and examine the potential impact of this combination. novel biomarker, p2ps, has been shown to be preferentially expressed in malignant prostate tissue and to correspond to aggressiveness [7,8]. Several papers have reported the potential benefit of p2ps measurement in men under investigation for PCa [9 14]. We hypothesized that the measurement of p2ps and its incorporation into a logistic regression model, along with other clinical variables, would improve risk stratification in an Irish cohort of men referred for PCa assessment. Patients and Methods Patient Cohort The study cohort consisted of 304 men who were identified from an ongoing multicentre risk stratification study of patients referred to an Irish Rapid ccess Clinic for the assessment of PCa between pril 2012 and ugust These men were selected based on the availability of a pre-biopsy biobanked serum specimen. Referral was performed according to the Irish National Cancer Control Programme guidelines [2]. The indications for referral according to these guidelines are an abnormal DRE or two abnormal PS levels at 6-week intervals. Table 1 ge-based assessment of PS as per the National Cancer Control Programme referral guidelines and the associated thresholds for an abnormal result [2]. ge PS* > > > >6.5 * Two bnormal levels >6 weeks apart. PS is an age-based assessment as outlined in Table 1. Men aged <40 years or those with a previous diagnosis of PCa were excluded. The final cohort consisted of 250 men after the exclusion of men with inadequate serum volume or without full clinical information. fter appropriate counselling and informed consent, prostate biopsy was performed. The patients all underwent a 12-core TRUS biopsy, with a standard template and extra sampling of echogenic lesions, if identified. Biopsy specimens were analysed histologically and graded according to the 2005 International Society of Urological Pathology Consensus Conference 2005 [15]. Each Gleason score was discussed among three consultant histopathologists before a final decision was made. Low grade PCa was defined as Gleason 6 and high grade PCa as Gleason 7. Laboratory Methods We collected 9 ml of blood directly into a vacutainer blood collection tube containing serum clot activator and this was centrifuged at 1500g for 10 min at room temperature. The serum was isolated into cryovials of 1 ml volume and placed into a polystyrene box filled with ice to maintain the samples at 4 C until storage at 80 C. The samples were stored at 80 C within 3 h of initial blood draw. The serum samples were aliquoted and anonymized before storage. The serum samples were measured at a freeze thaw cycle of 2. The samples were thawed, homogenized via rolling, and centrifuged at 1500g for 5 min before they were analysed consecutively. The authors were blinded as to the identity of each serum sample. Patient serum was analysed for total PS (tps), free PS (fps) and p2ps on the Beckman Coulter DxI800 Unicel Immunoassay system (Beckman Coulter, High Wycombe, UK). Statistical nalysis Basic statistical analyses of the study population s characteristics were performed using the unpaired Student s t-test and the Wilcoxon rank-sum test for continuous variables, while Pearson s chi-squared test was performed for categorical variables (Table 2). The three statistical tests, which together form the gold standard for the evaluation of a clinical prediction model, namely receiver-operating curve (ROC) analysis, calibration and decision-curve analysis, were used [16,17]. These methods allow the discriminative capacity as well as potential clinical utility to be examined. ROC analysis produces an area under the curve (UC) for each model by plotting the sensitivity and specificity of the model at each of its risk thresholds. Comparison of ROC curves took place via a method described by DeLong et al. [18]. Calibration plots were computed by comparing the predicted probabilities and observed probabilities of each model. Decision-curve analysis was undertaken to examine the potential net benefit of the application of each model in excess of the benefit offered by the strategies of performing 410 BJU International 2015 BJU International

3 Risk assessment incorporating PHI score Table 2 Study cohort characteristics. ll patients PCa (n) No PCa (n) P Patients, n Mean age, years * Previous biopsy, n Yes No DRE, n bnormal Normal Family history, n Gleason score category, n Gleason 6 35 Gleason 7 48 Gleason 8 15 Gleason Biomarkers Median (range) tps 6.4 ( ) 6.7 ( ) 6.01 ( ) <01 Median (range) fps ng/ml 085 ( ) 33 ( ) 03 ( ) 35 % ( ) 13.5 ( ) 17.6 ( ) <01 Median (range) p2ps pg/ml ( ) ( ) ( ) <01 % ( ) 20.5 ( ) 15.5 ( ) <01 Median (range) PHI ( ) ( ) ( ) <01 PCa, prostate cancer; fps, free PS; tps, total PS; PHI, Prostate Health Index. *Student s paired t-test. Chi-squared test. Wilcoxon rank-sum test. biopsy in all patients and performing biopsy in none. The biomarkers measured in the present study were analysed by univariable logistic regression, as were their various combinations by multivariable logistic regression. Two models were created by multivariable logistic regression, one of which incorporated tps (measured across the three institutions), along with age at biopsy, abnormality on DRE, family history of PCa and a previous negative biopsy (PS model). The second model incorporated the Prostate Health Index (PHI) score, along with age at biopsy, abnormality on DRE, family history of PCa and a previous negative biopsy (PHI model). Each model was then fitted to predict PCa and high grade PCa. The predictions of these models were then converted to percentage probabilities. Scatterplots were constructed to show the impact of the use of the PHI score to create a novel model by showing the change in probabilities compared with those of the PS model. To examine whether overfitting of the models was responsible for their disparity in performance, the models were subject to tenfold cross-validation, after which the difference between the PS model and PHI model was maintained. Finally, the patients clinical information and tps levels were used to compute each patient s individual probability according to the Prostate Cancer Prevention Trial (PCPT) risk calculator tool, which is available online [19]. This risk calculator is based on tps, age at biopsy, ethnicity, abnormality on DRE, family history of PCa and history of negative prostate biopsy. It has served as the benchmark for the evaluation of our novel multivariable logistic regression models. ll statistical analysis was carried out using the R studio software programme [20] and the relevant statistical packages [21,22]. Results Of the 250 men, 112 (45%) were diagnosed with PCa on biopsy and 77 (31%) were diagnosed with high grade disease. Table 2 shows the clinical characteristics of the patient cohort. nalysis of the available biomarkers showed that the PHI score was the most predictive of biopsy outcome in terms of discriminative ability (Table 3), with an UC for the Table 3 Discriminative ability of biomarkers and biomarker combinations. UC of biomarkers (95% CI) UC of biomarker combinations (95% CI) PCa High grade PCa High grade tps ( ) ( ) PHI ( ) ( ) fps ( ) ( ) tps + fps ( ) ( ) p2ps ( ) ( ) tps + %fps ( ) ( ) %fps ( ) ( ) fps + p2ps ( ) ( ) %p2ps ( ) ( ) %fps + %p2ps ( ) ( ) PHI ( ) ( ) PS + %fps + %p2ps ( ) ( ) UC, area under the curve. tps, total PS; fps, free PS; PCa, prostate cancer; PHI, Prostate Health Index. BJU International 2015 BJU International 411

4 R. W. Foley et al. PHI score of 0.71 in the prediction of PCa compared with an UC of 0.62 for the tps value that was available to the clinician (P < 1). This superiority was maintained for the diagnosis of high grade disease, with an UC for the PHI score of 0.78 compared with 0.70 for PS (P = 3). The UC values of the individual biomarkers fps, p2ps, %fps and %p2ps for both PCa and high grade PCa were all inferior to the combination of biomarkers that constitute the PHI score; however, of these, %p2ps had the highest UC value for PCa (0.68) and high grade disease (0.72). Figure 1 demonstrates the superior calibration of the PHI score when compared with the other biomarkers measured. This superior calibration was also evident in high grade PCa. Decision-curve analysis (Fig. 2) shows the superior net benefit of the PHI score over almost its entire range of threshold probabilities. To examine the PHI score calculation, an exhaustive comparison of all possible combinations of the measured biomarkers was undertaken (Table 3). It is clear that the best possible combination of tps, fps and p2ps (including their percentage values) is the PHI score, which achieved an Fig. 1 Calibration plots for the biomarkers and biomarker calculations for the diagnosis of PCa (Total PS [tps] as measured as part of the Prostate Health Index [PHI] score calculation). tps B fps C p2ps % tps E % p2ps F PHI D Fig. 2 Decision-curve analysis showing the net benefit of each of the biomarkers and biomarker calculations for the diagnosis of () prostate cancer (PCa) and (B) high grade PCa. The Prostate Health Index (PHI) score (top line) demonstrated separation from the other biomarkers across a wide range of threshold values Net benefit 0.1 None ll tps (Clinic) tps fps p2ps %fps %p2ps PHI B Net benefit None ll tps (Clinic) tps fps p2ps %fps %p2ps PHI Threshold probability Threshold probability BJU International 2015 BJU International

5 Risk assessment incorporating PHI score UC of 0.71 for the diagnosis of PCa and 0.78 for a diagnosis of high grade PCa. No other combination of these biomarkers was able to produce an UC superior to that of the PHI score for both PCa and high grade PCa diagnosis. To examine the clinical utility of the PHI score, the PHI model was compared with the PS model and both were compared with a well-validated PCa risk tool, the PCPT model. Figure 3 shows the probabilities of the model without the novel biomarker calculation (PS model) and the new probabilities given by the model incorporating the PHI score (PHI model). There is appreciable scatter from the 45 line, indicating that the PHI model is potentially useful in patient counselling, as its risk probabilities can be seen to vary markedly from those of the PS model [16]. If no appreciable scatter away was noted, even with a superior UC of a new model, it would provide only a small change in each patient s predicted risk and would be unlikely to influence clinical decision-making. The three clinical prediction models, the PHI, PS and PCPT models, were compared via ROC analysis (Fig. 4), calibration plots (Fig. 5) and decision-curve analysis (Fig. 6). The PHI model was found to have superior performance in all three statistical methods of comparison. The UC for the PHI model in the prediction of PCa was 0.77 and was significantly greater than that of the PS model (UC = 0.71; P < 1) and that of the PCPT model (UC = 0.69; P < 1). This was also true of high grade PCa, in which the UC for the PHI model, PS model and PCPT were 0.79, 0.72 and 0.72, respectively (all P < 1). The PHI model also outperformed the use of the PHI score in isolation, the discriminative ability of which is outlined in Table 3. The PHI model was the best calibrated clinical prediction tool, showing improved correlation between predicted probabilities and actual outcome. Decision-curve analysis also showed the superior clinical efficacy of the PHI model. There was a net benefit for the PHI model, i.e. true positives and true negatives in excess of false results, in comparison with the PS model and the PCPT model. Interestingly, when the cohort was separated into men who had undergone initial and those who had undergone repeat Fig. 3 Scatterplot demonstrating the relationship between the predicted probabilities of the Prostate Health Index (PHI) model and the PS model for both PCa () and high grade PCa (B). B probabilities of PHI Model probabilities of PHI Model probabilities of PS Model probabilities of PS Model Fig. 4 Receiver-operating characteristic curves for the Prostate Health Index (PHI) model, PS model and Prostate Cancer Prevention Trial (PCPT) model for the prediction of PCa () and high grade PCa (B). Sensitivity B Sensitivity PHI Model (UC: 0.77) PS Model (UC: 0.71) PCPT (UC: 0.69) PHI Model (UC: 0.79) PS Model (UC: 0.72) PCPT (UC: 0.72) Specificity Specificity BJU International 2015 BJU International 413

6 R. W. Foley et al. Fig. 5 Calibration plots of the clinical prediction models: () Prostate Cancer Prevention Trial (PCPT); (B) PS model and (C) Prostate Health Index (PHI) model for the diagnosis of PCa. The calibration of the models for high grade disease prediction is also shown: (D) PCPT model; (E) PS model; (F) PHI model. PCPT B PS Model C PHI Model High Grade Prostate Cancer D PCPT E PS Model F PHI Model Fig. 6 Decision-curve analysis of the three models for the diagnosis of () prostate cancer (PCa) and (B) high grade PCa. Net benefit 0.4 None ll PCPT PS Model PHI Model B Net benefit None ll PCPT PS Model PHI Model Threshold probability 5 Threshold probability biopsy (i.e. those with a previous negative biopsy), the potential value of the PHI test and its incorporation into a clinical prediction model became very pronounced. UC values for the prediction of PCa and Gleason 7 were 5 and 8, respectively, in our repeat biopsy patients (n = 60). This represents an improvement from respective UC values of 0.70 and 0.75 for the PS model (P < 1) and a striking improvement over that of the PCPT, with UC values of 0.56 and 0.55, respectively, for PCa and high grade PCa (P < 1). This performance was replicated on both calibration and decision-curve analysis. Discussion The accurate risk stratification of patients under investigation for PCa is of paramount importance. PCa is not alone in this regard, as Nguyen and Kattan have remarked, the ability to predict clinical outcomes accurately is critical to the proper management of any human disease [16]. Recent studies of PHI-based risk stratification have indeed created clinical prediction models to predict the outcome of prostate biopsy; however, too often these studies have failed to report clinically relevant results, which limit one s ability to examine accurately the potential clinical effect of these models. Concomitently, too few papers that deal with p2ps and the PHI score have reported the relationship with high grade disease (Gleason 7) [23 25], which is of utmost importance in PCa management as it is this association that dictates the need for aggressive treatment options. We have included this important metric throughout the study and can conclude that the PHI score, both independently and in a multivariable 414 BJU International 2015 BJU International

7 Risk assessment incorporating PHI score clinical prediction model, is clinically useful for the prediction of Gleason 7 PCa. Several papers examine PHI in isolation and do not form predictive models [26,27], or indeed form these models and neglect one or more important methods of statistical analysis [23]. Because any in-depth analysis of a novel biomarker would need to include decision analysis and calibration, it is immaterial to analyse a biomarker s clinical utility without the creation of a multivariable model and drawing a comparison with a model that does not include this novel biomarker [16]. In the present paper, we have performed a thorough analysis of the efficacy of a prediction model that incorporates the PHI score, which will allow urologists to make an informed decision as to whether the measurement of p2ps is of true clinical utility. To put these models into a clinical context, we have compared their performance in the present cohort to that of the PCPT model. The European Randomized study for Screening for Prostate Cancer (ERSPC) risk calculator was not included in this study as it relies on TRUS volume estimation and so does not fit into the current clinical practice within the Irish healthcare system. This measure is not commonly available to clinicians pre-biopsy, but if TRUS volume estimation was performed pre-biopsy, the ERSPC risk calculator could also incorporate a patient s PHI score along with other variables in the form of a smartphone app [28]. The PHI score is a continuous variable and, like PS, can be used to inform the decision for prostate biopsy. For example, a threshold, with an associated sensitivity and specificity, can be decided on and levels above this denoted as abnormal, much the same as has been done traditionally with PS. Risk stratification using this strategy, however, is a crude method to inform a biopsy decision. Instead, the PHI score should be incorporated into a logistic regression model along with all other clinical risk factors in order to best inform this decision. This multivariable approach, endorsed by the European ssociation of Urology in their 2013 guidelines [4] and the Melbourne consensus statement [5], represents the way forward in clinical decision-making. Clinical prediction models, such as the one shown in the present paper, represent a useful tool in clinical settings and these models will provide a probability of an outcome; PCa or high grade PCa in this case. This probability can then be used to guide the decision to proceed to prostate biopsy in consultation with the patient. We have shown that the measurement of p2ps and calculating the PHI score has the potential to improve decision-making in a clinical setting, by providing a clinical prediction model. Because we do not advocate the use of a universal threshold in all patients, we have not reported traditional true-positives and false-positives etc. Instead of these performance indicators, we have used statistical methods that analyse the clinical prediction models as a whole across every possible cut-off value. The PHI score is a more discriminant biomarker than PS, has superior calibration and has a superior net benefit; therefore, by using PHI-based risk stratification, urologists will be in a position to make a more informed decision in each patient. One potential confounding factor, which may help to explain, in part, the variation in the performance of the PS model and PHI model, is that the measurement of tps available to the clinician pre-biopsy was taken across the three institutional sites participating in the study. lthough this measurement has a standardized calibration throughout; the use of a single tps assessment technique on a single machine was used for the measurement of tps as part of the PHI score. There was, however, no significant difference in performance of the two measured tps values in the present cohort. We acknowledge several limitations to the present study. The sample size is relatively small and the study population was ethnically homogenous. Because recruitment was based on patient willingness to participate, it is impossible to exclude a selection bias. s with all predictive modelling, a model can only be as accurate as the outcome it attempts to predict. This model is based on the outcome of a TRUS biopsy, which carries with it an inherent false-negative rate [29]. It is likely that a model based on the outcome of transperineal biopsy, for example, which has a lower false-negative rate [30], would be more accurate. Several radiologists or urologists, depending on individual hospital practice, carried out prostate biopsy in this cohort. lthough a standardized approach was adopted there may be significant variability in biopsy technique; however, we believe this reflects a realistic pattern likely to be encountered in the majority of clinical settings. To risk stratify patients pre-biopsy and direct resources towards those patients most in need, thus reducing the risk of a biopsy to those who do not require the procedure, all possible tools in the urologist s armamentarium must be examined. This includes novel biomarkers for PCa and a multivariable approach to risk stratification. The PHI score improves significantly on PS and, when incorporated into a multivariable model, patient risk stratification is further improved. The PHI score represents a substantial step in PCa diagnosis. The incorporation of biomarkers such as the PHI score into clinical prediction models signifies the future of PCa management and is essential for the accurate risk stratification of patients and the accurate diagnosis of PCa. cknowledgements UniCel DxI 800 Immunoassay System analyser p2ps ([-2] props), fps and tps reagents were provided by Beckman BJU International 2015 BJU International 415

8 R. W. Foley et al. Coulter UK Ltd. and was performed using the UniCel Machine at Beaumont Hospital, Dublin. We would like to acknowledge the support of the Irish Cancer Society as part of the Prostate Cancer Research Consortium, and UCD School of Medicine through the Intercalated MSc Medical Science Scholarship Programme. Specimens used in this study were collected with the support of a Movember GP1 ward (to. S. P.) and the Dublin Centre for Clinical Research. We also acknowledge the clinical lead for each hospital participating in this study, namely; Prof. Tom Lynch, Mr. Richard Power and Mr. David Galvin. Conflict of Interest The authors have no conflicts of interest to declare. The study sponsor, Beckman Coulter UK Ltd., provided reagents for this study. Beckman Coulter UK Ltd played no role in the design of study, choice of enrolled patients, interpretation of data, preparation or approval of this manuscript. References 1 National Cancer Registry. Cancer in Ireland : nnual Report of the National Cancer Registry 2014, vailable at: report 2014.pdf. ccessed July Irish National Cancer Control Programme. nccp Prostate Cancer Referral Guideline, vailable at: NCCP Prostate Cancer Referral Guideline Version 1.3 January 2011.pdf. ccessed July Loeb S, Vellekoop, hmed HU et al. Systematic review of complications of prostate biopsy. Eur Urol 2013; 64: Heidenreich, Bastian PJ, Bellmunt J et al. EU guidelines on prostate cancer. part 1: screening, diagnosis, and local treatment with curative intent-update Eur Urol 2014; 65: Murphy DG, hlering T, Catalona WJ et al. The Melbourne Consensus Statement on the early detection of prostate cancer. BJU Int 2014; 113: Van Vugt H, Roobol MJ, Busstra M et al. Compliance with biopsy recommendations of a prostate cancer risk calculator. BJU Int 2012; 109: Mikolajczyk SD, Catalona WJ, Evans CL et al. Proenzyme forms of prostate-specific antigen in serum improve the detection of prostate cancer. Clin Chem 2004; 50: CatalonaWJ, Bartsch G, Rittenhouse HG et al. Serum pro-prostate specific antigen preferentially detects aggressive prostate cancers in men with 2 to 4 ng/ml prostate specific antigen.j Urol 2004; 171(6 Pt 1): Fossati N, Lazzeri M, Haese et al. Clinical performance of serum isoform [-2]proPS (p2ps) and its derivatives, namely %p2ps and PHI (Prostate Health Index) in men younger than 60 years of age: results from a multicentric European study. BJU Int 2015; 115: Mearini L, Ferri C, Lazzeri M et al. Evaluation of prostate-specific antigen isoform p2ps and its derivates, %p2ps, prostate health index and prostate dimension-adjusted related index in the detection of prostate cancer at first biopsy: an exploratory, prospective study. Urol Int 2014; 93: Perdona S, Bruzzese D, Ferro M et al. Prostate health index (phi) and prostate cancer antigen 3 (PC3) significantly improve diagnostic accuracy in patients undergoing prostate biopsy. Prostate 2013; 73: Ferro M, Bruzzese D, Perdona S et al. Prostate Health Index (Phi) and Prostate Cancer ntigen 3 (PC3) significantly improve prostate cancer detection at initial biopsy in a total PS range of 2-10 ng/ml. PLoS One 2013; 8: e Chun FK-H, Briganti, Graefen M et al. Development and external validation of an extended 10-core biopsy nomogram. Eur Urol 2007; 52: Guazzoni G, Nava L, Lazzeri M et al. Prostate-specific antigen (PS) isoform p2ps significantly improves the prediction of prostate cancer at initial extended prostate biopsies in patients with total PS between 2.0 and 10 ng/ml: results of a prospective study in a clinical setting. Eur Urol 2011; 60: Epstein JI, llsbrook WC, min MB, Egevad LL. The 2005 International Society of Urological Pathology (ISUP) Consensus Conference on gleason grading of prostatic carcinoma. m J Surg Pathol 2005; 29: Nguyen CT, Kattan MW. How to tell if a new marker improves prediction. Eur Urol 2011; 60: Vickers J, Cronin M. Traditional statistical methods for evaluating prediction models are uninformative as to clinical value: towards a decision analytic framework. Semin Oncol 2010; 37: DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988; 44: PCPT. Individualized Risk ssessment of Prostate Cancer, vailable at: ccessed July R Core Team. R: Language and Environment for Statistical Computing. ustria: Foundation for Statistical Computing, Robin X, Turck N, Hainard et al. proc: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 2011; 12: Fox J, Weisberg S. n R Companion to pplied Regression, 2nd edn. Thousand Oaks, C: Sage, Lazzeri M, Haese, de la Taille et al. Serum isoform [-2]proPS derivatives significantly improve prediction of prostate cancer at initial biopsy in a total PS range of 2-10 ng/ml: a multicentric European study. Eur Urol 2013; 63: Loeb S, Sokoll LJ, Broyles DL et al. Prospective multicenter evaluation of the Beckman Coulter Prostate Health Index using WHO calibration. J Urol 2013; 189: Lughezzani G, Lazzeri M, Larcher et al. Development and internal validation of a Prostate Health Index based nomogram for predicting prostate cancer at extended biopsy. J Urol 2012; 188: Catalona WJ, Partin W, Sanda MG et al. multicenter study of [- 2]pro-prostate specific antigen combined with prostate specific antigen and free prostate specific antigen for prostate cancer detection in the 2.0 to 1 ng/ml prostate specific antigen range. J Urol 2011; 185: Jansen FH, van Schaik RHN, Kurstjens J et al. Prostate-specific antigen (PS) isoform p2ps in combination with total PS and free PS improves diagnostic accuracy in prostate cancer detection. Eur Urol 2010; 57: Roobol M, zevedo N. The rotterdam prostate cancer risk calculator: improved prediction with more relevant pre-biopsy information, now in the palm of your hand. J Urol 2014; 191: e Roemeling S, Schr oder FH. Words of wisdom. Re: Needle biopsies on autopsy prostates: sensitivity of cancer detection based on true prevalence. Eur Urol 2008; 53: Numao N, Kawakami S, Sakura M et al. Characteristics and clinical significance of prostate cancers missed by initial transrectal 12-core biopsy. BJU Int 2012; 109: BJU International 2015 BJU International

9 Risk assessment incorporating PHI score Correspondence: Robert Foley, Conway Institute of Biomolecular and Biomedical Research, Univeristy College Dublin, Belfield, Dublin 4, Ireland. bbreviations: fps, free PS; PHI, Prostate Health Index; PCPT, Prostate Cancer Prevention Trial; ROC, receiveroperating characteristic; PCa, prostate cancer; UC, area under the curve; tps, total PS; ERSPC, European Randomized study for Screening for Prostate Cancer. BJU International 2015 BJU International 417