Predictive Models for Newly Diagnosed Prostate Cancer Patients

Management Update

3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 117 MANAGEMENT UPDATE Predictive Models for Newly Diagnosed Prostate Cancer Patients William T. Lowrance, MD, Peter T. Scardino, MD Department of Surgery, Urology Service, Memorial Sloan-Kettering Cancer Center, New York, NY Accurate risk assessment is of paramount importance to newly diagnosed prostate cancer patients and their physicians. Risk prediction models help identify those at high (or low) risk of disease progression and guide discussions about prognosis and treatment. Widely used, well-validated prediction tools are based on standard, readily available clinical and pathologic parameters, but do not include biomarkers, some of which may have an important role in predicting prognosis or determining therapeutic options. A new approach, known as systems pathology, may improve the accuracy of traditional prediction methods and provide patients with a more personalized risk assessment of clinically relevant outcomes. The ultimate goal of prediction models is to improve medical decision making. [Rev Urol. 2009;11(3):117-126 doi: 10.3909/riu0456] © 2009 MedReviews®, LLC Key words: Prostate cancer • Prognosis • Statistical model • Nomogram • Biological marker rostate cancer remains the most common solid organ malignancy in American men, with over 186,000 new cases diagnosed in 2008 and an estimated 28,000 deaths.1 With the widespread use of prostate-specific antigen (PSA) screening in the 1990s, there has been a dramatic stage migration for prostate cancer, with more patients diagnosed at earlier clinical stages and lower serum PSA values. Because of enhanced screening, the number of clinically insignificant prostate cancers diagnosed has increased, and, as a result, there is rising concern over treatment of such cancers. P VOL. 11 NO. 3 2009 REVIEWS IN UROLOGY 117 3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 118 Predictive Models for Prostate Cancer continued The recent European and American prostate cancer screening trials reported in the New England Journal of Medicine provide new insights into cancer screening and the associated overdetection and overtreatment of clinically insignificant tumors.2,3 According to data from the European Randomized Study of Screening for Prostate Cancer, 48 cases of prostate Currently there is a host of prostate cancer prediction tools to gauge the risk of specified oncologic outcomes available for use in clinical practice. Multiple prediction methods were developed for use in the posttreatment setting to estimate the risk of PSA recurrence or disease progression after definitive treatment (radical prostatectomy or radiation therapy). We are in need of better tools for risk assessment of newly diagnosed prostate cancer patients, enabling physicians to focus treatment efforts on those most likely to benefit from aggressive therapy. cancer need to be treated to prevent a single death from prostate cancer within 10 to 12 years.3 Clearly, we are in need of better tools for risk assessment of newly diagnosed prostate cancer patients, enabling physicians to focus treatment efforts on those most likely to benefit from aggressive therapy. Risk estimation from such posttreatment prediction tools typically relies heavily on PSA, pathologic Gleason score, and detailed pathologic stage information available after radical prostatectomy (eg, surgical margin status, presence of extracapsular extension, and seminal vesicle or lymph node involvement). Other prediction tools are available for use in the pretreatment setting, allowing for risk assessment in newly diagnosed prostate cancer patients contemplating treatment options. These prediction tools also primarily rely on PSA, Gleason score, and clinical stage to estimate risk, but lack the detailed pathologic variables accounted for in most posttreatment models. Risk categorization, probability tables, risk scores, and nomograms are some examples of widely used risk analysis methods (Table 1). These risk estimators use clinical and pathologic features and do not account for novel molecular biomarkers or genomic variants that may contribute to improved predictive accuracy. This article reviews the value of select traditional predictive tools for guiding treatment decisions and prognosis discussions in patients with clinically localized prostate cancer, and discusses the role of novel predictive Table 1 Tools for Risk Estimation in Newly Diagnosed Prostate Cancer Patients Tool Description Example Risk categories Uses categorized clinical and pathologic variables to separate patients into broad risk groups D’Amico risk groups4-6 (low, intermediate, or high risk of BCR) Probability tables Shares some features of risk categories and nomograms; uses categorized clinical and pathologic predictor variables to calculate the probability of specified outcomes Partin tables11-13 (probability of organconfined cancer, extracapsular extension, seminal vesicle, or lymph node involvement) Risk score Similar to risk categories, but also incorporates % of positive biopsy cores and age; more personalized risk assessment compared with risk groups UCSF-CAPRA score14,15 (risk score 1-10, denoting different probabilities of recurrence) Nomograms Most widely used risk prediction tools; calculate probability of event based on continuous and categorical input; most individualized risk assessment compared with risk groups, scores, or probability tables Kattan nomograms20-22,30,31 (probability of 5and 10-year freedom from BCR; indolent cancer or pathologic stage) Systems pathology Combines common clinical and pathologic parameters with quantitative tissue morphology and biomarkers obtained through advance imaging analysis Prostate Px34 (probability of clinical failure or favorable pathology/indolent disease) BCR, biochemical recurrence; UCSF-CAPRA, University of California, San Francisco-Cancer of the Prostate Risk Assessment. Prostate Px®, Aureon Laboratories, Yonkers, NY. 118 VOL. 11 NO. 3 2009 REVIEWS IN UROLOGY 3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 119 Predictive Models for Prostate Cancer tools utilizing systems pathology and genomic analyses. Risk Categories A straightforward and logical way for clinicians to estimate risk of disease recurrence after treatment is to stratify patients into distinct risk categories. In the pretreatment setting, this approach is attractive for many physicians, as they can easily categorize patients based on a few ubiquitous clinical parameters. The most widely used risk grouping system was developed by D’Amico and colleagues,4-6 where men with localized prostate cancer are grouped into categories according to whether their risk of biochemical recurrence (BCR) after definitive treatment was low, intermediate, or high. Patients are primarily stratified according to their biopsy Gleason score, serum PSA level, and clinical stage, and defined as: (1) low risk, clinical stage T1c and T2a, PSA level  10 ng/mL, and biopsy Gleason score  6; (2) intermediate risk, clinical stage T2b or biopsy Gleason score of 7 or PSA level  10 and  20 ng/mL; (3) high risk, clinical stage  T2c or PSA level  20 ng/mL or biopsy Gleason score  8. Note that a patient must meet all 3 criteria to be included in the low-risk group, but any 1 criterion can move him to a higher risk group. The American Urological Association (AUA) has incorporated D’Amico’s risk groups into their most recent prostate cancer clinical guidelines.7 Another example of pretreatment risk categorization is described by Zelefsky and colleagues. Patients are divided into favorable, intermediate, and unfavorable prognostic groups based on whether PSA is  10 ng/mL, clinical stage  T2, and Gleason sum  6.8 If all 3 conditions are met, the patient is considered to have a favorable prognosis. If 1 of the conditions is not met, then the patient falls into the intermediate group, and if 2 or 3 are unmet, then an unfavorable prognosis is assigned. The National Comprehensive Cancer Network (NCCN) adopted a modified version of Zelefsky’s and D’Amico’s risk grouping systems for their prostate cancer clinical practice guidelines.9 The NCCN guidelines panel recommends risk stratification incorporating “available predictive features included in the guidelines, risk tables, and nomograms when discussing options for the treatment of clinically localized prostate cancer.”9 The enthusiasm for risk groupings is primarily a result of their ease of application, but this enthusiasm advising newly diagnosed prostate cancer patients. Probability Tables Created in 1997 and updated in 2001 and 2007, the Partin11-13 tables are the most well-known and widely used probability tables in urology. Using PSA value, Gleason score, and clinical stage, these probability tables separately predict the likelihood of organconfined disease, extracapsular extension, seminal vesicle, or lymph node involvement. The Partin tables were originally based on the pathologic findings of over 4000 patients who underwent radical prostatectomy at 1 of 3 different academic institutions from 1982 through 1996. The enthusiasm for risk groupings is primarily a result of their ease of application, but this enthusiasm should be tempered by the loss of predictive power associated with collapsing variables into broad categories. should be tempered by the loss of predictive power associated with collapsing variables into broad categories. For example, when 2 newly diagnosed prostate cancer patients are classified as intermediate risk according to D’Amico’s grouping strategy, one assumes they are at equal risk of disease recurrence after treatment. In fact, patients within each of the 3 risk groups are a heterogeneous group and the individual risk may vary widely.10 Additionally, the clinical variables used to assign risk groups are weighted equally in their potential to result in a given outcome, when 1 variable (eg, grade) may have a much greater effect on prognosis than another (eg, stage). Inappropriate weighting results in a mixture of patients with very different individualized risks of recurrence all lumped into a broad category. Despite their limitations, many clinicians continue to rely on these methods of risk characterization when Like risk grouping, probability tables are relatively easy to understand and use, and they have been widely validated. However, their ease of use does not come without a cost. By collapsing data such as PSA and Gleason score into broad categories, the Partin tables lose some of their predictive accuracy for the 4 pathologic outcomes described above. Therefore, risk estimates are not as personalized as they could be, and instead are more akin to the low-, intermediate-, and high-risk categories discussed earlier. Additional drawbacks to the use of probability tables include the fact that they only predict pathologic stage and not prognosis. The pathologic outcome predictions are mutually exclusive; the Partin table–predicted probability of extracapsular extension is too low because those with seminal vesicle or lymph node involvement and extracapsular extension are excluded. Lastly, the tables do not use quantitative biopsy results, and there VOL. 11 NO. 3 2009 REVIEWS IN UROLOGY 119 3. RIU0456_10-22.qxd 10/22/09 9:57 PM Page 120 Predictive Models for Prostate Cancer continued is no way to estimate location of extracapsular extension, which would be helpful in surgical planning. Risk Score Investigators from the University of California, San Francisco (UCSF) developed a novel risk assessment tool that provided more detailed risk prediction than the previous 3-level risk categories.14 Known as the UCSFCancer of the Prostate Risk Assessment (CAPRA) score, it predicts cancer recurrence after radical prostatectomy based on PSA, biopsy Gleason score, clinical T stage, percentage of positive biopsy cores, and age. The UCSFCAPRA prediction index was derived from Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) data and validated in an independent cohort that showed accurate preoperative prediction of cancer recurrence.15 All input variables are categorized and weighted according to their predictive power. As with risk categories, some of the predictive ability of the input variables is lost after collapsing them into categories, but the effect is less with this scoring system compared with D’Amico’s groupings because more categories are allowed (eg, D’Amico’s system incorporates 3 PSA categories vs 5 PSA categories used by the UCSF-CAPRA score). The UCSF-CAPRA score ranges from 0 through 10, and the risk of recurrence roughly doubles per 2-point increase (Table 2). Nomograms Using traditional statistical methods, nomograms are graphical devices that allow for the approximate calculation of a given function. In medicine, that function tends to be the probability of a certain endpoint based on specified predictor variables. The number of available prostate cancer prediction nomograms can be overwhelming for the practicing physician. There seems 120 VOL. 11 NO. 3 2009 to be a nomogram for every possible clinical setting and endpoint. For example, there are multiple tools for predicting cancer prevalence at the time of prostate biopsy.16-18 Other tools allow risk assessment after diagnosis, but prior to treatment; they predict outcomes like pathologic stage, pathologic Gleason sum, BCR, or prostate cancer-specific–mortality after various treatment options.19-26 Additionally, there are prediction models for the posttreatment setting, providing risk information about important clinical endpoints (eg, BCR, metastasis, or survival).27-29 Useful prediction models, regardless of the clinical setting and outcome of interest, must be accurate, easy to use, generalizable, and have strong performance characteristics. For the purposes of this review, we focus primarily on nomograms predicting prostate cancer outcomes in the pretreatment setting. Table 2 Scoring System for the UCSF-CAPRA Risk Score Prediction Tool With Corresponding 5-Year Recurrence-Free Survival Estimates According to Score II. Gleason Score I. PSA 2.1-6.0 6.1-10 10.1-20 20.1-30 30      0 1 2 3 4 1-3/1-3  0 1-3/4-5  1 4-5/1-5  3 III. Clinical T Stage T1/T2  0 T3a  1 IV. % Positive Biopsies V. Age 34%  0 34%  1 50 years  0 50 years  1 UCSF-CAPRA Score 5-Year Recurrence-Free Survival Estimate (95% CI) 0-1 points 85% (73-92) 2 points 81% (69-89) 3 points 66% (54-76) 4 points 59% (40-74) 5 points 60% (37-77) 6 points 34% (12-57)  7 points 8% (0-28) PSA, prostate specific antigen; UCSF-CAPRA, University of California, San Francisco-Cancer of the Prostate Risk Assessment. A point value from each of the 5 categories is summed to calculate the CAPRA score. Scores range from 0-10. Recurrence defined as PSA  0.2 ng/mL on 2 consecutive postprostatectomy measurements or a 2nd cancer treatment  6 months after prostatectomy. Adapted with permission from The Journal of Urology, Vol 173, Cooperberg MR et al., “The University of California, San Francisco Cancer of the Prostate Risk Assessment score: a straightforward and reliable preoperative predictor of disease recurrence after radical prostatectomy," pp. 1938-1942, Copyright 2005, with permission from the American Urological Association.14 REVIEWS IN UROLOGY 3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 121 Predictive Models for Prostate Cancer rence-free probability. The graphical nature of nomograms facilitates their clinical use and interpretation. Kattan and colleagues also developed nomograms predicting the probability of newly diagnosed prostate cancer patients harboring clinically insignificant disease, as well as those to predict pathologic stage.21,30,31 Useful websites are available where multiple predictor values can be entered and precise risk calculations made based on easy-to-use, freely available computerized versions of these nomograms (Table 3). Unlike probability tables, risk scores, or risk groupings where predictor variables must be collapsed into categories, predictors can be incorporated into nomograms as continuous variables. The full original value of the predictor variable is preserved, leading to improved risk estimation. The different weight given to each variable can be adjusted to fit the data. As a result, nomograms provide individualized risk prediction for newly diagnosed prostate cancer patients. A preoperative prostate cancer risk assessment nomogram developed by Kattan and colleagues in 1998 and last updated in 2009 now estimates the 10-year freedom from BCR.20,22 Figure 1 demonstrates the updated nomogram where surgeon experience is factored along with PSA, clinical stage, and Gleason score. Points are summed for each of the parameters and then equated to a PSA recur- Comparing Risk Prediction Methods With so many decision aids available for use, how do clinicians choose? Prediction tools should be judged on their accuracy in a given patient population, usability, and generalizability. Usability refers to how userfriendly a prediction tool is in the Figure 1. Preoperative nomogram predicting 10-year freedom from biochemical recurrence for use in patients who have chosen radical prostatectomy. Reprinted with permission from Kattan MW et al.22 0 20 40 60 80 100 clinic. Accuracy can be measured using a concordance index (CI) or area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and the hazard ratio. These are similar statistically derived values used to describe the discriminatory ability (accuracy) of a test or model. Simply stated, the CI represents the probability of concordance between predicted and observed events. Likewise, the AUC measures the ability of a test to correctly differentiate between those with or without a condition. Whether prediction models are assessed using CI or AUC is dependent on the outcome predicted by the model. Models predicting time-toevent outcomes are evaluated via a CI, whereas models predicting a binary outcome use AUC. The range of values for both CI and AUC is from 0.5 to 1, with 1 being perfect discrimination and 0.5 being equal to a coin flip or random chance. These values can be used to directly compare the performance of different prediction models on a common dataset; the more accurate prediction tool in terms of discrimination will have the higher CI or AUC. Points Systems Pathology Prostate Specific Antigen 0 1 2 3 T2a 4 5 6 7 8 10 15 30 50 70 90 100 T2c Clinical Stage T2b T1c T3 Surgeon Experience 2000 750 250 0 5 7 9 Biopsy Gleason Sum 4 6 8 10 Total Points 0 10-y Biochemical RecurrenceFree Probability 0.98 40 0.95 80 0.9 120 0.8 0.7 160 0.5 0.3 200 0.1 0.01 240 Unfortunately, none of the above prediction methods for a prostate cancer–related endpoint is perfect, largely because the input variables are not sufficiently informative and/or reproducible. For example, Gleason score, which is a heavily weighted predictor factored into most risk assessment models, is determined by the subjective interpretation of the microscopic morphology of prostate glands by a single pathologist. This amount of subjectivity naturally results in significant variability in outcome predictions. Systems pathology is a novel multidisciplinary methodology created to eliminate the subjectivity inherent in VOL. 11 NO. 3 2009 REVIEWS IN UROLOGY 121 3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 122 Predictive Models for Prostate Cancer continued Table 3 Web-Based Resources for Pretreatment Personalized Prostate Cancer Risk Estimation Probability Tables http://urology.jhu.edu/prostate/partintables.php Nomograms http://www.mskcc.org/applications/nomograms/prostate/ Prostate Px http://www.aureon.com/prognostic-tests-prostate-px.htm Prostate Px®, Aureon Laboratories, Yonkers, NY. how pathologists currently evaluate the morphologic and molecular characteristics of tumor specimens.32,33 Systems pathology strives to replace subjective analyses with quantitative ones to increase the accuracy of Pretreatment Setting Using a systems pathology platform similar to that developed for postprostatectomy outcome predictions, Aureon Laboratories developed Prostate Px®, a test focusing on risk Systems pathology strives to replace subjective analyses with quantitative ones to increase the accuracy of current prediction tools used in prostate cancer and other diseases. current prediction tools used in prostate cancer and other diseases. Through machine learning (specifically support vector regression for censored data), the systems pathology approach was used to create several new prostate cancer prediction models by Aureon Laboratories (Yonkers, NY). The new prediction tools involve the use of standard clinical and pathologic features combined with advanced image analysis techniques that allow for the quantitation of biomarkers using immunofluorescence or immunohistochemistry and quantitation of specific microanatomic cellular characteristics. This new prediction approach fuses clinicopathologic features with state-of-the-art molecular and cellular markers utilizing advanced image analysis and artificial intelligence (Figure 2). 122 VOL. 11 NO. 3 2009 assessment in the newly diagnosed prostate cancer patient (prior to treatment). By combining clinical parameters with comprehensive molecular and microanatomical image analyses of prostate core biopsy tissue, Prostate Px provides physicians and patients with personalized outcome predictions to aid in treatment decision making. The goal of this technology is to increase the predictive accuracy of the typical pretreatment risk assessment methods currently in use that may be limited by only factoring in clinicopathologic variables. Under the direction of Aureon Laboratories, a multi-institutional group developed and validated this pretreatment prediction model.34 The model used 3 clinicopathologic variables, 1 biomarker variable, and 2 advanced imaging features from the biopsy tissue. It predicted clinical failure (as defined earlier) within 8 years of radical prostatectomy. Table 4 specifies the variables included in the pretreatment model predicting clinical failure. The model was validated on an independent cohort of patients and yielded a concordance index of 0.73. Within their validation cohort, Donovan and colleagues compared the performance of their systems pathology–based model with the Kattan 5- and 10-year BCR nomograms and found, in addition to a slightly higher concordance Figure 2. Overview of the systems pathology approach to risk prediction. REVIEWS IN UROLOGY Standard Clinical and Pathologic Data Quantitative Tissue Morphology Molecular and Genetic Biomarkers Predictive Model (developed from artificial intelligence) ⴝ Personalized Risk Estimate for Specified Outcome 3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 123 Predictive Models for Prostate Cancer Table 4 Variables Included in Pretreatment Prediction Model for Clinical Failure Based on Systems Pathology Platform (in Order of Weighted Significance) Clinical/Pathologic Variables Preoperative PSA Dominant Gleason grade Biopsy Gleason score Biomarker Variable Combined androgen receptor dynamic range at dominant Gleason grade  3, total Ki-67 at dominant Gleason grade  3 Advanced Imaging Variables Combined mean distance between epithelial tumor cells at dominant Gleason grade  3, actual grade at dominant Gleason grade  3 Area of isolated (non–lumen associated) tumor epithelial cells relative to total tumor area PSA, prostate-specific antigen. Data from Donovan et al.34 Table 5 Variables Included in Pretreatment Prediction Model for Favorable Pathology Based on Systems Pathology Platform (in Order of Weighted Significance) Clinical/Pathologic Variables Preoperative PSA Biopsy Gleason score Advanced Imaging Variables Area of isolated (non–lumen associated) tumor epithelial nuclei relative to total tumor area Area of epithelial nuclei with differential distance from gland lumens relative to total tumor area Biomarker Variable Combined androgen receptor dynamic range at dominant Gleason grade  3, total Ki-67 at dominant Gleason grade  3 PSA, prostate-specific antigen. Data from Donovan et al.34 index for the systems pathology model (0.73 vs 0.69), the systems approach was twice as sensitive as the Kattan nomogram in identifying patients initially designated by the AUA classification as intermediate risk who were actually high risk, that is, who developed clinical recurrence within 8 years.33 Also included in the Prostate Px test is another pretreatment model developed by the same authors and recently reported in the Journal of Urology.34 This model predicts indolent disease, defined by favorable pathology at the time of radical prostatectomy. They specified favorable pathology to include: (1) pathologic T stage  2; (2) pathologic Gleason score  6 (no Gleason 4 or 5 pattern); (3) an undetectable PSA ( 0.2 ng/ mL). This model uses 5 predictors, 4 of which overlap with the clinical failure model described above. The differing variable is associated with tumor differentiation and is measured using specialized hematoxylin and eosin–stained image analysis, specifically measuring tumor epithelial nuclei area within a certain distance from gland lumens. Table 5 outlines the variables analyzed in the favorable pathology model. Validation studies showed this prediction tool to be accurate, with an AUC of 0.74. The combining of the 2 pretreatment prostate cancer prediction models developed through the systems pathology approach provides a new risk assessment method that can be used when counseling newly diagnosed prostate cancer patients. It is hoped that with the discovery of more informative biomarkers, all patients and physicians will benefit from improved risk prediction accuracy. Genomic Advances Genomic and proteomic studies are likely to yield exciting new biomarkers to help distinguish clinically significant prostate cancers from VOL. 11 NO. 3 2009 REVIEWS IN UROLOGY 123 3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 124 Predictive Models for Prostate Cancer continued Table 6 Top 10 Most Frequently Selected Genes in Leave-One-Out-Cross-Validation of Combined Modeling Approach Gene Symbol Gene Title Models* Mean Expression in Recurrent Tumors EI24 Etoposide-induced 2.4 mRNA 79 Overexpressed EPB49 Erythrocyte membrane protein band 4.9 78 Underexpressed MAP4K4 Mitogen-activated protein kinase kinase kinase kinase 4 78 Overexpressed GMCL Germ cell-loss homolog (Drosophila) 50 Overexpressed HNRPC Heterogeneous nuclear ribonucleoprotein C (C1/C2) 22 Overexpressed PCOLN3 Procollagen (type III) N-endopeptidase 22 Underexpressed SIL TAL1 (SCL) interrupting locus 21 Overexpressed APP Amyloid beta (A4) precursor protein 20 Overexpressed SSRI Signal sequence receptor, alpha 13 Overexpressed BTF Bcl-2-associated transcription factor 9 Overexpressed *Number of models in which variable was selected. Reprinted with permission from Stephenson AJ et al.40 those that pose little threat to their host. A systems pathology platform allows the integration of such markers into clinical risk assessment tools. Recently, expression array analyses have identified common somatic mutations in prostate cancer that may have prognostic value. The most common example is the fusion of the androgen responsive transmembrane protease serine 2 (TMPRSS2) gene with the erythroblast-transformation specific (ETS) gene (eg, ERG, ETV1, or ETV4) discovered by Tomlins and colleagues.35 In some studies the TMPRSS2:ERG fusion has been associated with more aggressive prostate cancer and poor prognosis,36 whereas other reports question its clinical prognostic capabilities.37 More research is needed to better define its role in clinical practice, but the TMPRSS2:ERG fusion certainly represents an exciting new genetic marker. In an attempt to provide prognostic information to prostate cancer patients beyond what is available from well-established clinical and patho- 124 VOL. 11 NO. 3 2009 logic variables, prediction models based on multiple gene signatures have been created. For example, Cheville and colleagues developed a multivariable model for the postprostatectomy setting to distinguish men who develop systemic progression from those who do not by accounting for gene expression of topoisomerase2a, cadherin-10, predicted TMPRSS2 (ERG, ETV1, or ETV4) fusion status, and aneuploidy.38 In validation studies, this model yielded an AUC of 0.79. Likewise, Glinksy and colleagues were able to accurately predict biochemical recurrence in the posttreatment setting using a predictive algorithm of gene signatures.39 Combining gene expression signatures with traditional clinical and pathologic variables may create even more accurate prostate cancer recurrence prediction models. Stephenson and coworkers used such a combined modeling approach, integrating multiple prognostic gene markers with Kattan’s postoperative nomogram to correctly classify recurrence status in REVIEWS IN UROLOGY 89% of prostate cancer specimens (concordance index, 0.89).40 Table 6 shows the 10 most frequently used genes in the combined modeling approach, and Figure 3 depicts the probability of BCR for men classified as nonrecurrent versus recurrent by this combined model.40 Based on certain gene expression patterns, Febbo and Sellers were also able to separate patients into recurrent and nonrecurrent groups.41 Currently available prostate cancer risk prediction tools incorporating novel genetic markers are only available for the posttreatment setting. Their translation into the pretreatment arena has been slow and will continue to be difficult because of the limited tumor tissue provided by a transrectal biopsy for advanced genetic or proteomic analyses. Nonetheless, the future of risk prediction methods in prostate cancer will most likely involve the thoughtful combination of well-established clinical and pathologic features with novel cellular, molecular, and genetic biomarkers. 3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 125 Predictive Models for Prostate Cancer features—have guided risk assessment for prostate cancer patients. Prostate Px, a risk prediction tool based on a systems pathology approach, demonstrates good potential for improving the way risk is assessed in these patients. Methods of risk prediction will continue to evolve as more biomarkers are discovered and the understanding of genetic variations and prostate cancer progresses. 1.0 Proportion Free of Progression 0.9 0.8 0.7 0.6 0.5 0.4 0.3 This article was supported by NIH T32CA082088 Urology Oncology training grant, NCI grant CA092629 SPORE in prostate cancer, Sidney Kimmel Center for Prostate and Urologic Cancers, David H. Koch, and Aureon Laboratories. 0.2 0.1 Nonrecurrent Recurrent 0 0 6 12 18 42 48 54 24 30 36 Months From Prostatectomy 60 40 17 38 9 66 72 At Risk 43 36 41 23 40 13 39 11 References 31 3 1. 2. Figure 3. Kaplan-Meier estimates of the probability of disease recurrence for patients classified as nonrecurrent and recurrent by model combining gene expression signatures and clinical variables. Reprinted with permission from Stephenson AJ et al.40 3. Conclusion The accurate assessment of risk for disease progression or treatment failure in a newly diagnosed prostate cancer patient is of critical importance and should guide all treatment discussions and recommendations. Traditionally, risk categorization, probability tables, and nomograms—all based on established clinical and pathologic 4. Jemal A, Siegel R, Ward E, et al. Cancer statistics, 2008. CA Cancer J Clin. 2008;58:71-96. Andriole GL, Crawford ED, Grubb RL 3rd, et al. Mortality results from a randomized prostatecancer screening trial. N Engl J Med. 2009; 360:1310-1319. Schröder FH, Hugosson J, Roobol MJ, et al. Screening and prostate-cancer mortality in a randomized European study. N Engl J Med. 2009;360:1320-1328. D’Amico AV, Whittington R, Malkowicz SB, et al. Pretreatment nomogram for prostate-specific antigen recurrence after radical prostatectomy or external-beam radiation therapy for clinically Main Points • Risk assessment methods currently used for newly diagnosed prostate cancer patients are not perfect, primarily because input variables are not sufficiently informative. • Widely used risk prediction tools in the pretreatment setting include D’Amico risk categories, Partin probability tables, University of California, San Francisco-Cancer of the Prostate Risk Assessment risk score, and Kattan nomograms; all rely heavily on traditional clinical variables (eg, serum prostate-specific antigen level, Gleason score, and clinical stage) to estimate risk of various outcomes. • In general, nomograms maximize the predictive ability of each input variable, allowing for a more individualized characterization of risk compared with risk categories or probability tables. • The systems pathology approach is a new prediction method that fuses clinicopathologic features with state-of-the-art molecular and cellular markers utilizing advanced image analysis and artificial intelligence. With the goal of improving the accuracy of current risk prediction methods, this approach decreases subjectivity inherent in traditional pathologic tumor specimen processing and incorporates novel biomarkers into the prediction process. • Including new gene expression signatures in risk assessment models can also improve prediction accuracy. • Future risk prediction methods in prostate cancer will likely involve the thoughtful combination of clinical and pathologic features with novel molecular and genetic biomarkers in either a systems pathology–based approach or via a traditional nomogram. VOL. 11 NO. 3 2009 REVIEWS IN UROLOGY 125 3. RIU0456_10-22.qxd 10/22/09 4:24 PM Page 126 Predictive Models for Prostate Cancer continued 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 126 localized prostate cancer. J Clin Oncol. 1999; 17:168-172. D’Amico AV, Whittington R, Malkowicz SB, et al. The combination of preoperative prostate specific antigen and postoperative pathological findings to predict prostate specific antigen outcome in clinically localized prostate cancer. J Urol. 1998;160:2096-2101. D’Amico AV, Whittington R, Malkowicz SB, et al. Biochemical outcome after radical prostatectomy, external beam radiation therapy, or interstitial radiation therapy for clinically localized prostate cancer. JAMA. 1998;280: 969-974. Thompson I, Thrasher JB, Aus G, et al. Guideline for the management of clinically localized prostate cancer: 2007 update. J Urol. 2007; 177:2106-2131. Zelefsky MJ, Leibel SA, Gaudin PB, et al. Dose escalation with three-dimensional conformal radiation therapy affects the outcome in prostate cancer. Int J Radiat Oncol Biol Phys. 1998;41: 491-500. The NCCN Clinical Practice Guidelines in Oncology™ Prostate Cancer, ver. 2.2009. Fort Washington, PA: National Comprehensive Cancer Network, Inc.; 2009. Mitchell JA, Cooperberg MR, Elkin EP, et al. Ability of 2 pretreatment risk assessment methods to predict prostate cancer recurrence after radical prostatectomy: data from CaPSURE. J Urol. 2005;173:1126-1131. Makarov DV, Trock BJ, Humphreys EB, et al. Updated nomogram to predict pathologic stage of prostate cancer given prostate-specific antigen level, clinical stage, and biopsy Gleason score (Partin tables) based on cases from 2000 to 2005. Urology. 2007;69:1095-1101. Partin AW, Kattan MW, Subong EN, et al. Combination of prostate-specific antigen, clinical stage, and Gleason score to predict pathological stage of localized prostate cancer. A multi-institutional update. JAMA. 1997;277:1445-1451. Partin AW, Mangold LA, Lamm DM, et al. Contemporary update of prostate cancer staging nomograms (Partin Tables) for the new millennium. Urology. 2001; 58:843-848. Cooperberg MR, Pasta DJ, Elkin EP, et al. The University of California, San Francisco Cancer of the Prostate Risk Assessment score: a straightforward and reliable preoperative predictor of disease recurrence after radical prostatectomy. J Urol. 2005;173:1938-1942. Cooperberg MR, Freedland SJ, Pasta DJ, et al. Multiinstitutional validation of the UCSF cancer of the prostate risk assessment for prediction of recurrence after radical prostatectomy. Cancer. 2006;107:2384-2391. VOL. 11 NO. 3 2009 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. REVIEWS IN UROLOGY Chun FK, Briganti A, Graefen M, et al. Development and external validation of an extended 10-core biopsy nomogram. Eur Urol. 2007;52: 436-444. Eastham JA, May R, Robertson JL, et al. Development of a nomogram that predicts the probability of a positive prostate biopsy in men with an abnormal digital rectal examination and a prostate-specific antigen between 0 and 4 ng/mL. Urology. 1999;54:709-713. Karakiewicz PI, Benayoun S, Kattan MW, et al. Development and validation of a nomogram predicting the outcome of prostate biopsy based on patient age, digital rectal examination and serum prostate specific antigen. J Urol. 2005; 173:1930-1934. Chun FK, Steuber T, Erbersdobler A, et al. Development and internal validation of a nomogram predicting the probability of prostate cancer Gleason sum upgrading between biopsy and radical prostatectomy pathology. Eur Urol. 2006; 49:820-826. Kattan MW, Eastham JA, Stapleton AM, et al. A preoperative nomogram for disease recurrence following radical prostatectomy for prostate cancer. J Natl Cancer Inst. 1998;90:766-771. Kattan MW, Eastham JA, Wheeler TM, et al. Counseling men with prostate cancer: a nomogram for predicting the presence of small, moderately differentiated, confined tumors. J Urol. 2003;170:1792-1797. Kattan MW, Vickers AJ, Yu C, et al. Preoperative and postoperative nomograms incorporating surgeon experience for clinically localized prostate cancer. Cancer. 2009;115:1005-1010. Stephenson AJ, Scardino PT, Eastham JA, et al. Preoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Natl Cancer Inst. 2006; 98:715-717. Kattan MW, Potters L, Blasko JC, et al. Pretreatment nomogram for predicting freedom from recurrence after permanent prostate brachytherapy in prostate cancer. Urology. 2001;58:393-399. Kattan MW, Zelefsky MJ, Kupelian PA, et al. Pretreatment nomogram for predicting the outcome of three-dimensional conformal radiotherapy in prostate cancer. J Clin Oncol. 2000;18: 3352-3359. Kattan MW, Cuzick J, Fisher G, et al. Nomogram incorporating PSA level to predict cancerspecific survival for men with clinically localized prostate cancer managed without curative intent. Cancer. 2008;112:69-74. Kattan MW, Wheeler TM, Scardino PT. Postoperative nomogram for disease recurrence after radical prostatectomy for prostate cancer. J Clin Oncol. 1999;17:1499-1507. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. Stephenson AJ, Scardino PT, Eastham JA, et al. Postoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Clin Oncol. 2005;23: 7005-7012. Stephenson AJ, Scardino PT, Kattan MW, et al. Predicting the outcome of salvage radiation therapy for recurrent prostate cancer after radical prostatectomy. J Clin Oncol. 2007;25: 2035-2041. Koh H, Kattan MW, Scardino PT, et al. A nomogram to predict seminal vesicle invasion by the extent and location of cancer in systematic biopsy results. J Urol. 2003;170:1203-1208. Ohori M, Kattan MW, Koh H, et al. Predicting the presence and side of extracapsular extension: a nomogram for staging prostate cancer. J Urol. 2004;171:1844-1849; discussion 1849. Cordon-Cardo C, Kotsianti A, Verbel DA, et al. Improved prediction of prostate cancer recurrence through systems pathology. J Clin Invest. 2007;117:1876-1883. Donovan MJ, Hamann S, Clayton M, et al. A systems pathology approach for the prediction of prostate cancer progression after radical prostatectomy. J Clin Oncol. 2008;26:3923-3929. Donovan MJ, Khan FM, Fernandez G, et al. Personalized prediction of tumor response and cancer progression on prostate needle biopsy. J Urol. 2009;182:125-132. Tomlins SA, Rhodes DR, Perner S, et al. Recurrent fusion of TMPRSS2 and ETS transcription factor genes in prostate cancer. Science. 2005; 310:644-648. Kumar-Sinha C, Tomlins SA, Chinnaiyan AM. Recurrent gene fusions in prostate cancer. Nat Rev Cancer. 2008;8:497-511. Gopalan A, Leversha MA, Satagopan JM, et al. TMPRSS2-ERG gene fusion is not associated with outcome in patients treated by prostatectomy. Cancer Res. 2009;69:1400-1406. Cheville JC, Karnes RJ, Therneau TM, et al. Gene panel model predictive of outcome in men at high-risk of systemic progression and death from prostate cancer after radical retropubic prostatectomy. J Clin Oncol. 2008;26:3930-3936. Glinsky GV, Glinskii AB, Stephenson AJ, et al. Gene expression profiling predicts clinical outcome of prostate cancer. J Clin Invest. 2004;113: 913-923. Stephenson AJ, Smith A, Kattan MW, et al. Integration of gene expression profiling and clinical variables to predict prostate carcinoma recurrence after radical prostatectomy. Cancer. 2005; 104:290-298. Febbo PG, Sellers WR. Use of expression analysis to predict outcome after radical prostatectomy. J Urol. 2003;170:S11-S19; discussion S19-S20.