Personalized Prediction of Tumor Response and Cancer Progression on Prostate Needle Biopsy

Transcription

1 Personalized Prediction of Tumor Response and Cancer Progression on Prostate Needle Biopsy Michael J. Donovan,*, Faisal M. Khan, Gerardo Fernandez, Ricardo Mesa-Tejada, Marina Sapir, Valentina Bayer Zubek, Douglas Powell, Stephen Fogarasi, Yevgen Vengrenyuk, Mikhail Teverovskiy, Mark R. Segal, R. Jeffrey Karnes, Thomas A. Gaffey, Christer Busch, Michael Haggman, Peter Hlavcak, Stephen J. Freedland, Robin T. Vollmer, Peter Albertsen, Jose Costa and Carlos Cordon-Cardo From the Aureon Laboratories, Inc. (MJD, FMK, GF, RMT, MS, VBZ, DP, SF, YV, MT, JC, CCC), Yonkers and Herbert Irving Comprehensive Cancer Center, Columbia University (RMT, CCC), New York, New York, Center for Bioinformatics and Molecular Biostatistics, University of California-San Francisco (MRS), San Francisco, California, Mayo Clinic (RJK, TAG), Rochester, Minnesota, University Hospital at Uppsala (CB, MH, PH), Uppsala, Sweden, Durham Veterans Affairs and Duke University Medical Center (SJF, RTV), Durham, North Carolina, and University of Connecticut Health Science Center (PA), Farmington and Yale University School of Medicine (JC), New Haven, Connecticut Purpose: To our knowledge in patients with prostate cancer there are no available tests except clinical variables to determine the likelihood of disease progression. We developed a patient specific, biology driven tool to predict outcome at diagnosis. We also investigated whether biopsy androgen receptor levels predict a durable response to therapy after secondary treatment. Materials and Methods: We evaluated paraffin embedded prostate needle biopsy tissue from 1,027 patients with ct1c-t3 prostate cancer treated with surgery and followed a median of 8 years. Machine learning was done to integrate clinical data with biopsy quantitative biometric features. Multivariate models were constructed to predict disease progression with the C index to estimate performance. Results: In a training set of 686 patients (total of 87 progression events) 3 clinical and 3 biopsy tissue characteristics were identified to predict clinical progression within 8 years after prostatectomy with 78% sensitivity, 69% specificity, a C index of 0.74 and a HR of Validation in an independent cohort of 341 patients (total of 44 progression events) yielded 76% sensitivity, 64% specificity, a C index of 0.73 and a HR of Increased androgen receptor in tumor cells in the biopsy highly significantly predicted resistance to therapy, ie androgen ablation with or without salvage radiotherapy, and clinical failure (p ). Conclusions: Morphometry reliably classifies Gleason pattern 3 tumors. When combined with biomarker data, it adds to the hematoxylin and eosin analysis, and prostate specific antigen values currently used to assess outcome at diagnosis. Biopsy androgen receptor levels predict the likelihood of a response to therapy after recurrence and may guide future treatment decisions. Key Words: prostate; prostatic neoplasms; biopsy; receptors, androgen; biological markers PROSTATE cancer remains the most commonly diagnosed nonskin cancer in American men and it causes approximately 29,000 deaths each year. 1 Treatment options are radical prostatectomy, radiotherapy and watchful waiting with Abbreviations and Acronyms AMACR -methyl-acylcoenzyme A racemase AR androgen receptor bgg dominant biopsy Gleason grade bgs biopsy Gleason sum CF clinical failure CK18 cytokeratin 18 FP favorable pathology MST minimal spanning tree pakt phosphorylated SVRc censored data support vector regression Submitted for publication December 18, Study received Durham Veterans Affairs Medical Center, Mayo Clinic, University of Connecticut Health Science Center, University of Graz and University Hospital at Uppsala institutional review board approval. Supported by an Aureon Laboratories sponsored research agreement. * Correspondence: Aureon Laboratories, Inc., 28 Wells Ave., Yonkers, New York (telephone: ; FAX: ; Michael.Donovan@aureon.com). Financial interest and/or other relationship with Aureon Laboratories. Financial interest and/or other relationship with Sanofi, Bristol Meyers, Galaxo and Blue Cross Blue Shield. Supplementary material for this article can be obtained at jason.alter@aureon.com. For another article on a related topic see page /09/ /0 Vol. 182, , July 2009 THE JOURNAL OF UROLOGY Printed in U.S.A. Copyright 2009 by AMERICAN UROLOGICAL ASSOCIATION DOI: /j.juro

2 126 PREDICTING PROSTATE CANCER PROGRESSION no apparent consensus on how to maximize disease control and survival, especially in men with intermediate risk prostate cancer (PSA 10 to 20 ng/ml, clinical stage T2b-c and Gleason score 7). The only randomized clinical study to compare observation vs surgery showed a lower overall rate of death in men with T1 or T2 disease treated with radical prostatectomy, although results must be weighed against quality of life issues and comorbidity. 2,3 Furthermore, PSA screening has challenged traditional prognostic models due to the over diagnosis of indolent tumors, lead time bias, grade inflation and longer life expectancy. 4 8 Several groups have developed methods to predict prostate cancer outcomes based on information at diagnosis. The updated Partin tables predict the risk of pathological stage (extracapsular extension, and seminal vesicle and lymph node invasion), 9 while the 10- year preoperative nomogram 10 provides the probability of freedom from biochemical recurrence within 10 years after radical prostatectomy. A recognized limitation is that these tools rely on clinical data and predict a biochemical recurrence outcome that does not invariably relate to systemic disease, suggesting that additional end points are required for optimal patient specific risk stratification. 11 We previously used systems pathology to identify quantitative features associated with prostate cancer progression. 12,13 Our prostatectomy CF model used a complex end point comparable to that in the biopsy study, including castrate PSA increase, bone metastasis and prostate cancer specific death with the androgen resistance PSA end point denoting early progression as a surrogate for systemic metastasis. 13 Incorporating the castrate PSA increase end point in the biopsy model maximized the number of patients at high risk and more importantly optimized our ability to identify men with rapidly progressive disease for potential early intervention. We now report a biopsy tool developed in a multi-institutional cohort followed a median of 8 years postoperatively. METHODS Patients and Samples This study was approved by the Durham Veterans Affairs Medical Center, Mayo Clinic, University of Connecticut Health Science Center, University of Graz and University Hospital at Uppsala institutional review boards. Information was compiled on 1,487 patients treated with radical prostatectomy between 1989 and 2003 for localized or locally advanced prostate cancer (ct1c-t3) for which formalin fixed, paraffin embedded tissue samples were available. We excluded from study patients treated with neoadjuvant therapy. Researchers elsewhere who were not involved in the study randomized and split the cohort between the training and validation sets (67% vs 33%) with a similar proportion of CF events and demographic balance. CF was prespecified as any of 3 events, including 1) unequivocal radiographic or pathological evidence of metastasis, castrate or noncastrate (including skeletal or soft tissue disease in lymph nodes or solid organs), 2) increasing PSA in a castrate state, ie androgen ablation with or without salvage radiotherapy, or 3) death from prostate cancer, as documented by a review of the medical record and death certificate. Time to CF was defined as from prostatectomy to the first of these events. If a patient did not experience CF as of the last visit or the outcome at the most recent visit was unknown, the outcome was censored. Hormonal therapy or salvage radiotherapy was done at treating physician discretion. The castrate PSA increase was the first PSA increase in a trajectory regardless of treatment dose, type and duration. Gleason score and bgg were obtained after reevaluating the primary diagnostic biopsy at each institution. Clinical stage was assessed by chart review. Only patients with complete clinicopathological, morphometric and molecular data, including outcome information, were further studied for a total of 686 for training and 341 for validation (table 1). Characteristics in these 1,027 patients were similar to those in the original 1,487 (data not shown). Patients were excluded from analysis primarily due to poor biopsy specimen quality (crush or artifact), poor antigen quality due to over fixation and/or auto-fluorescence, too little usable tumor content (6 or fewer glands) and/or incomplete clinical data. Exclusion parameters were evenly distributed among the different cohorts. Up to 7 unstained slides and/or paraffin blocks Table 1. Characteristics of patients in training and validation sets Characteristics No. Training (%) No. Validation (%) Overall Mean age Preop PSA (ng/ml): 10 or Less 460 (67.1) 231 (67.7) Greater than (32.9) 110 (32.3) bgg: 2 25 (3.6) 8 (2.3) (76.4) 246 (72.1) (19.0) 85 (24.9) 5 7 (1.0) 2 (0.6) Gleason score: 4 5 (0.7) 4 (1.2) 5 31 (4.5) 7 (2.1) (42.9) 159 (46.6) (41.8) 137 (40.2) 8 46 (6.7) 25 (7.3) 9 17 (2.5) 8 (2.3) 10 6 (0.9) 1 (0.3) Clinical stage: T1a 6 (0.9) 3 (0.9) T1c 263 (38.3) 116 (34.0) T2 374 (54.5) 198 (58.1) T3 27 (3.9) 15 (4.4) Missing 16 (2.3) 9 (2.6) CF events: 87 (12.7) 44 (12.9) Castrate PSA increase 77 (11.2) 40 (11.7) Pos bone scan 9 (1.3) 4 (1.2) Death from prostate Ca 1 (0.1) 0

3 PREDICTING PROSTATE CANCER PROGRESSION 127 were retrieved per patient with hematoxylin and eosin stained sections to identify tumor content, representative Gleason score and highest Gleason grade. Image Analysis Up to 3 hematoxylin and eosin images were acquired and assessed for tumor content, Gleason grade and quality (staining, morphology and artifacts) by 3 of us (MJD, GF and RMT). Using a digital masking tool infiltrating tumor, including tumor gland lumina, was outlined and included for morphometric analysis. Image analysis software was used as previously described 12,13 to generate quantitative histological features based on prostate cancer cellular properties, eg the relationship of epithelial nuclear area to gland lumen area. The final result of each morphometric feature was its median value derived from available tumor. Quantitative Multiplex Immunofluorescence Multiple antigens were quantified in single tissue sections by immunofluorescence as previously reported with individual assays developed according to biological parameters and technical capabilities. 12,13 Briefly, we performed 2 multiplex assays in prostate needle biopsies with Alexa fluorochrome labeled antibodies for certain antigens, including 1) multiplex 1 AR, racemase (AMACR), CK18, TP73L (p63) and high molecular weight keratin, and 2) multiplex 2 Ki67, pakt, CD34, CK18 and AMACR. Each multiplex contained DAPI (4=-6-diamidino-2-phenylindole) for nuclei. Image analysis algorithms were used to identify cell types/ compartments, eg luminal epithelial cells and epithelial/ stromal nuclei, and quantify AR, Ki67, pakt, CD34 and AMACR in prostate specimens. Machine learning was used to determine optimal thresholds and assign positive and negative profiles. The final result of each immunofluorescence feature was its median value derived from available tumor. Statistical Analysis The predictive model was constructed using SVRc, 12 a method that applies support vector regression to high dimensional data adapted to use with censored records. If a patient did not experience CF as of the last visit or the outcome at the most recent visit was unknown, the patient outcome was censored. All other patients were considered events or noncensored. Our experience with SVRc showed that this approach can increase model predictive accuracy over that of the Cox model. 12,13 RESULTS Training Set Patient Characteristics In the training set of 686 patients 87 (12.7%) had CF after prostatectomy, including 9 with positive bone scan, 77 with a castrate increase in PSA and 1 who died of prostate cancer. These 686 patients were followed a median of 96 months (range 2.5 to 237.5) after prostatectomy. Of the men 33% had preoperative PSA greater than 10 ng/ml, 42% had a biopsy Gleason score of 7 or greater and 55% had ct2 disease. A total of 219 patients (32%) received standard androgen ablation with or without salvage radiotherapy. Table 1 lists patient characteristics. On univariate analysis preoperative PSA, biopsy Gleason score and bgg were associated with CF (C index 0.4 or less, or 0.6 or more). Prostate Tumor Sample Quantitative Analysis Histological image analysis. Morphometric features were generated from hematoxylin and eosin stained tumor areas, reflecting overall tissue architecture, including the tumor cell distribution and their relationship to glandular structures. On univariate analysis 27 histological features were significantly associated with CF (C index 0.4 or less, or 0.6 or more). Quantitative immunofluorescence. We previously noted the usefulness of AMACR to identify and characterize individual tumor cells. 13 In the current study we quantified and generated AR, Ki67 and pakt features in AMACR positive and negative epithelial tumor cells, and used the endothelial marker CD34 to capture vessel content. We also used DAPI and CK18 on tumor morphometry to quantify the proximity between tumor cells and their distribution with respect to glands and each other (MST functions). 14 MST, AR and Ki67 features incorporated bgg for marker assessment, ie for bgg 3 or less the AR feature was used and for bgg greater than 3 the Ki67 feature was used. In the training set 36 of 303 noncensored patients (10%) with bgg 3 or less had clinical progression within 8 years of prostatectomy. Of the 36 cases 19 (52%) had high AR, suggesting that AR expression discriminates significant from indolent disease, especially for low grade cancer. In contrast, 31 of 55 noncensored patients (56%) with bgg greater than 3 had clinical progression within 8 years of prostatectomy. In this group increasing Ki67 was additive with bgg in regard to shortened time to clinical progression. Model Development A SVRc model to predict CF was developed in the 686 training set patients. Modeling began with the 40 variables associated with CF on univariate analysis. Supervised multivariate learning resulted in an optimized model containing 6 independent (nonintercorrelated) features. Table 2 lists these features in order of importance in the model. The clinical features were preoperative PSA, bgs and bgg. The 2 imaging features, infiltrating cells and cellular topology, reflected cellular and tissue architecture at the transition between Gleason patterns 3 and 4. The hematoxylin and eosin feature quantified the proportion of tumor epithelial cells not directly associated with an intact gland structure. The MST combined feature (for bgg 3 or less use MST function and for bgg greater than 3 use actual Gleason grade) quantified tumor cell spatial

4 128 PREDICTING PROSTATE CANCER PROGRESSION Table 2. Features selected in model Features Univariate p Value (signed log rank test) Model Wt Preop PSA bgg Biopsy Gleason score AR dynamic range at low bgg total Ki67 at high bgg Mean distance between epithelial tumor cells at low bgg grade at high bgg Isolated tumor epithelial cell area/total tumor area distribution, reflecting differentiation and stromal content. Kaplan-Meier curves for the 2 imaging features showed their ability to accurately stratify patients (fig. 1, A and B). From biomarker based features the SVRc algorithm selected only the combined feature of dynamic range of AR and total Ki67 content. Shorter time to CF was predicted by increasing proportion of tumor cells with high AR expression in specimens with clinical bgg 3 or less and high Ki67 in specimens with bgg 4 and 5. Of note, combined features leveraged the clinical bgg to identify where the biomarker was most predictive with respect to outcome and while triggered by clinical bgg, the combined molecular feature was more predictive than bgg or bgs alone (univariate C index vs 0.37 and 0.336). Figure 1, C shows the Kaplan-Meier curve in patients stratified by this feature. Figure 2 shows representative AR and Ki67 profiles in tumors with low and high bgg, respectively. The primary measure of model accuracy was the C index, which is similar in interpretation to the ROC AUC. The C index estimates the probability that of a pair of randomly chosen comparable patients the patient with the higher predicted score in the model will experience progression within a shorter time than the other patient. The C index in a multivariable model is 0.5 (model performs the same as a coin toss) to 1.0 (model has perfect ability to discriminate). The training model had a C index of When patients were stratified by model score below vs above 30.19, corresponding to a 13.82% model predicted probability of CF, the HR was 5.12, sensitivity was 78% and specificity was 69% for correctly predicting CF within 8 years. Kaplan-Meier curves in patients by model score showed the ability of the model to separate patients according to risk (fig. 3, A). Validation The model was validated at University of California-San Francisco using data on 341 patients with a median followup of 72 months (range 9.5 to 200). Of the patients 44 (12.9%) had CF, including 4 with a positive bone scan and 40 with a castrate increase in PSA. Of the men 32% had preoperative PSA greater than 10 ng/ml, 40% had biopsy Gleason score greater than 7 and 58% had ct2 disease. A Probability of remaining Progression Free post RP B Probability of remaining Progression Free post RP C Probability of remaining Progression Free post RP Months to Disease Progression post RP Months to Disease Progression post RP Months to Disease Progression post RP Figure 1. Kaplan-Meier curves show freedom from CF in training set by select morphometric imaging features in prostate needle biopsy specimen. A, H & E feature shows area of nongland associated/single and multiple infiltrating cells (cutoff 0.31, p ). B, combined feature incorporating cellular topology by MST (cutoff 3.93, p ). C, combined feature of dynamic AR range in bgg 3 or less tumors, or Ki67 with bgg 4 to 5 (combined feature cutoff 0.943, p ).

5 PREDICTING PROSTATE CANCER PROGRESSION 129 Of the patients 113 (33%) received standard androgen ablation with or without salvage radiotherapy. The model resulted in 76% sensitivity, 64% specificity, a C index of 0.73 and a HR of 3.47 for predicting CF. Separate Kaplan-Meier curves were generated in patients with a score of above or below (fig. 3, B). These 2 patient groups differed significantly in time to CF (log rank test p ). Biopsy AR Predicted Durable Response to Therapy After Recurrence Previously we noted that AR in prostatectomy samples predicted the response to therapy after recurrence. 13 We confirmed these results in biopsy specimens from 219 patients (training cohort) who received hormonal therapy with or without salvage radiotherapy, of Probabitlity of Remaining Clinical Failure Free Probabitlity of Remaining Clinical Failure Free Months Post Radical Prostatectomy Months Post Radical Prostatectomy Figure 3. Kaplan-Meier curves show probability of freedom from clinical progression by SVRc model score (log rank test p ). Green lines indicate low predicted risk (model score or less). Red lines indicate high predicted risk (model score greater than 30.19). A, training set (HR 5.12). B, validation set (HR 3.47). whom 77 (35%) progressed with castrate PSA increase. On this analysis increasing AR was significantly associated with decreased time to failure (log rank test p ). In addition, a hematoxylin and eosin lumen feature was associated with outcome (log rank test p 0.05), supporting a further role for morphometry as a surrogate for Gleason grading (table 3). Clinical variables, including preoperative PSA, and biopsy Gleason grade and score, were not significant. Figure 2. Typical multiplex immunofluorescence results in patients at high risk. A, AR in bgg 3 tumor epithelial nuclei with intensity increasing from least (blue areas) to moderate (red areas) to high (yellow areas). B, Ki67 (yellow areas) in tumor epithelial nuclei (blue areas). Purple areas indicate stromal nuclei. Reduced from 200. Table 3. AR immunofluorescence, hematoxylin and eosin morphometric features associated with time to clinical progression after hormonal therapy Feature p Value Tumor epithelial cells: AR area in AMACR pos cells AR dynamic intensity range Hematoxylin eosin median lumen area 0.01

6 130 PREDICTING PROSTATE CANCER PROGRESSION FP Prediction in Prostatectomy Specimen With surgical outcome data as a clinical end point we investigated whether pretreatment features in the disease progression model predicted FP, defined as ptnm T2 or less, prostatectomy Gleason score 6 or less (no Gleason 4) and undetectable PSA (nadir PSA 0.02 ng/ml or less). In a training set of 628 patients 2 clinical and 3 biopsy tissue characteristics were identified with 75% sensitivity, 69% specificity and an AUC of 0.78 to predict FP. Validation in an independent cohort of 280 patients yielded 74% sensitivity, 65% specificity and an AUC of Table 4 lists these features in order of importance in the model. Of note, 4 features overlapped with the CF model, including preoperative PSA, biopsy Gleason score, a hematoxylin and eosin morphometric feature assessing isolated tumor cells and bgg combined with AR and Ki67. The additional selected hematoxylin and eosin feature measured tumor epithelial nuclei within a specified distance from gland lumina and it was most related to tumor differentiation. DISCUSSION A challenge when treating patients with localized prostate cancer is determining who is at high risk for death from the disease. To address this issue our approach has been to generate predictive, tumor specific tools using technological advances in morphometry and quantitative immunofluorescence. Earlier we generated postoperative models to predict the likelihood of PSA recurrence 12 and CF 13 based on clinicopathological characteristics and prostatectomy specimen features. We have now generated a pretreatment model using clinical variables and prostate needle biopsy specimen features to predict CF after prostatectomy. The model was validated with 76% sensitivity, 64% specificity, a C index of 0.73 and a HR of 3.47 (p ). Incorporating morphometry with biomarker data adds to existing clinical tools for predicting outcome, ie hematoxylin and eosin analysis, and PSA, and provides vigorous assessment of Gleason grading. In contrast, the 5-year preoperative biochemical recurrence nomogram 15 for this cohort yielded a C index of 0.69 and a Table 4. Features selected in FP stage model Features* Preop PSA Isolated tumor epithelial nuclei area/total tumor area Biopsy Gleason score Epithelial nuclei area differential distance from gland lumen/total tumor area AR dynamic range at low bgg total Ki67 at high bgg * Univariate signed log rank p Model wt Table 5. Univariate and multivariate results for predicting CF within 8 years in validation cohort Predictor c Index HR HR Univariate p Value (signed log rank test) Age at biopsy Preop PSA Clinical stage bgg Gleason score Kattan PSA recurrence nomogram: 5 Yrs Yrs SVRc systems pathology model HR of 2.34 (p 0.005), supporting improved accuracy with the systems approach. To address the robustness of our current results we compared our risk stratification with traditional clinicopathological factors independently and in the Kattan nomograms (table 5). Furthermore, sensitivity and specificity analysis of the various nomograms vs our systems method in low and intermediate risk groups, as defined by American Urological Association criteria, indicated that the systems method is twice as effective for identifying patients at high risk for CF within 8 years but who appear to be at intermediate risk based on clinical profiles (sensitivity 62% vs 31% for the nomogram). Notable is the predominance of the castrate nonmetastatic PSA increase event for clinical progression. This dynamic point proved to be important for assessing disease outcome after definitive therapy. 13,16,17 Previously we observed that a castrate PSA increase was uniformly associated with distant metastasis. In the current biopsy study 10 of the 77 patients (14%) with a castrate PSA increase showed metastasis, of whom the model identified 9 (90%) as being at high risk for progression. We anticipate that the biological basis for the high risk classification, including transition to a Gleason 4 pattern and high AR, will be used as part of the therapeutic decision process. We believe that a systems model using multiple robust tumor characteristics will yield more objective risk assessment in contemporary patients, particularly in a community practice, where select pathological variables are prone to subjectivity. Clinical variables in the model were pretreatment PSA, bgs and bgg. As anticipated, higher bgg was associated with worse outcome on univariate analysis but it was associated with better outcome in the multivariate model. This phenomenon shows the reversal paradox, in which the variable acts as a control for other factors during modeling. 18 We hypothesize that the reversal in our model resulted from the impact of the 2 combined features containing bgg as a trigger, ie for bgg 3 or less use MST or AR. Although bgg appears useful for prognosis, the

7 PREDICTING PROSTATE CANCER PROGRESSION 131 variable usefulness associated with outcome shows the need for additional measures. This observation is further supported by the importance of morphometric variables in the progression and FP models, including MST, and hematoxylin and eosin features reflecting the distribution and organization of tumor epithelial cells and nuclei. The features identify a transition phase between Gleason grades 3 to 4 and increasing levels were associated with disease progression. We and others reported a central role for AR and Ki67 in prostate cancer growth and progression. 12,13,22 25 Our current model identifies these markers as part of a biological tumor grade for intermediate cancer and shows that aberrant AR, especially in bgg 3 or less tumors, may effect downstream signaling pathways and contribute to castrate/metastatic disease. The complex role of AR in disease progression is evidenced by its significance in our 2 prognostic models but also with respect to durability of the response to hormonal therapy by the shortened time to castrate PSA increase. We confirmed our prior observation and propose that this high AR expressing tumor cell population is prone to metastasis, clonal expansion and potentially resistant to standard treatments. Evidence in biopsy and prostatectomy specimens previously linked Ki67 with bgg and outcome. However, as with AR, clinical adoption has been challenged by the lack of reproducibility, poor standardized assays and the need for an accurate cutoff. Our approach of using an immunofluorescence based, quantifiable assay integrated with image analysis and mathematical models appears to have circumvented these limitations. Finally, although pakt was associated with outcome, it was not selected in the multivariate model. In addition, the features derived from CD34 vessel content did not attain univariate statistical significance. Several studies showed pakt involvement in prostate cancer proliferation and survival pathways and linked increased pakt with Ki-67, activated AR and a hormone refractory phenotype The role of CD34 is more controversial due to differing methods of identifying and counting vessels in various sample types. 29 We continue to investigate pakt and CD34 to clarify their prognostic/predictive significance for prostate cancer. CONCLUSIONS We developed an accurate tool to predict disease progression and FP in the prostatectomy specimen at diagnosis. Given the retrospective nature of the study design and the potential for inherent bias, we are investigating access to randomized clinical trial biopsy materials. Biological and morphological attributes of the model represent a phenotype that will most likely supplement current practice to determine appropriate treatment options and patient followup. ACKNOWLEDGMENTS Drs. Vijay Aggarwal, Robert Shovlin, Jason Alter and Charles DiComo contributed to the manuscript. Dr. William M. Pottenger, Rutgers University performed the randomized training/test split of the data. Mark Clayton, and Drs. Stefan Hamann, Paola Capodieci and Ali Tabesh, Aureon Laboratories; Dr. Angelique W. Levi, Aureon Laboratories and Yale University School of Medicine; Dr. Mona Norberg, University Hospital at Uppsala, Dr. Jayakrishnan Jayachandran and Leah Gerber, Durham Veterans Affairs and Duke University Medical Center; and Judith Fine, University of Connecticut Health Science Center provided clinical database and technical support. The cohort was randomized and split at Rutgers University. REFERENCES 1. Jemal A, Siegel R, Ward E, Hao Y, Xu J, Murray T et al: Cancer statistics, CA Cancer J Clin 2008; 58: Holmberg L, Bill-Axelson A, Helgesen F, Salo JO, Folmerz P, Haggman M et al: A randomized trial comparing radical prostatectomy with watchful waiting in early prostate cancer. N Engl J Med 2002; 347: Bill-Axelson A, Holmberg L, Ruutu M, Haggman M, Andersson SO, Bratell S et al: Radical prostatectomy versus watchful waiting in early prostate cancer. N Engl J Med 2005; 352: Klotz L: Active surveillance versus radical treatment for favorable-risk localized prostate cancer. Curr Treat Options Oncol 2006; 7: Dall Era MA, Cooperberg MR, Chan JM, Davies BJ, Albertsen PC, Klotz LH et al: Active surveillance for early-stage prostate cancer: review of the current literature. Cancer 2008; 112: Albertsen PC, Hanley JA and Fine J: 20-Year outcomes following conservative management of clinically localized prostate cancer. JAMA 2005; 293: Telesca D, Etzioni R and Gulati R: Estimating lead time and overdiagnosis associated with PSA screening from prostate cancer incidence trends. Biometrics 2008; 64: Barry MJ, Kaufman DS and Wu CL: Case : a 55 year-old-man with an elevated prostate-specific antigen level and early-stage prostate cancer. N Engl J Med 2008; 358: Makarov DV, Trock BJ, Humphreys EB, Mangold LA, Walsh PC, Epstein JI 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 Urology 2007; 69: Stephenson AJ, Scardino PT, Eastham JA, Bianco FJ Jr, Dotan ZA, Fearn PA et al: Preoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Natl Cancer Inst 2006; 98: 715.

8 132 PREDICTING PROSTATE CANCER PROGRESSION 11. Freedland SJ, Humphreys EB, Mangold LA, Eisenberger M, Dorey FJ, Walsh PC et al: Risk of prostate cancer-specific mortality following biochemical recurrence after radical prostatectomy. JAMA 2005; 294: Cordon-Cardo C, Kotsianti A, Verbel DA, Teverovsky M, Capodieci P, Hamann S et al: Improved prediction of prostate cancer recurrence through systems pathology. J Clin Invest 2007; 117: Donovan MJ, Hamann S, Clayton M, Khan FM, Sapir M, Bayer-Zubek V et al: A systems pathology approach for the prediction of prostate cancer progression after radical prostatectomy. J Clin Oncol 2008; 26: van Diest PJ, Fleege JC and Baak JP: Syntactic structure analysis in invasive breast cancer: analysis of reproducibility, biologic background, and prognostic value. Hum Pathol 1992; 23: Kattan MW, Eastham JA, Stapleton AM, Wheeler TM and Scardino PT: A preoperative nomogram for disease recurrence following radical prostatectomy for prostate cancer. J Natl Cancer Inst 1998; 90: Bianco FJ: Paradigms in androgen/castrate resistant states of prostate cancer in a biomarker era. Urol Oncol Semin Orig Invest 2008; 26: Smaletz O, Scher HI, Small EJ, Verbel D, McMillan A, Regan K et al: Nomogram for overall survival of patients with progressive metastatic prostate cancer after castration. J Clin Oncol 2002; 20: Tu YK, Gunnell D and Gilthorpe MS: Simpson s paradox, Lord s paradox, and Suppression Effects are the same phenomenon the reversal paradox. Emerg Themes Epidemiol 2008; 5: Gonzalgo ML, Bastian PJ, Mangold LA, Trock BJ, Epstein JI, Walsh PC et al: Relationship between primary Gleason pattern on needle biopsy and clinicopathologic outcomes among men with Gleason score 7 adenocarcinoma of the prostate. Urology 2006; 67: Grober ED, Tsihlias J, Jewett MA, Sweet JM, Evans AJ, Trachtenberg J et al: Correlation of the primary Gleason pattern on prostate needle biopsy with clinico-pathological factors in Gleason 7 tumors. Can J Urol 2004; 11: Muntener M, Epstein JI, Hernandez DJ, Gonzalgo ML, Mangold L, Humpreys E et al: Prognostic significance of Gleason score discrepancies between needle biopsy and radical prostatectomy. Eur Urol 2008; 53: Bubendorf L, Tapia C, Gasser TC, Casella R, Grunder B, Moch H et al: Ki67 labeling index in core needle biopsies independently predicts tumor-specific survival in prostate cancer. Hum Pathol 1998; 29: Mucci NR, Rubin MA, Strawderman MS, Montie JE, Smith DC and Pienta KJ: Expression of nuclear antigen Ki-67 in prostate cancer needle biopsy and radical prostatectomy specimens. J Natl Cancer Inst 2000; 92: Pollack A, DeSilvio M, Khor LY, Li R, Al-Saleem TI, Hammond ME et al: Ki-67 staining is a strong predictor of distant metastasis and mortality for men with prostate cancer treated with radiotherapy plus androgen deprivation: Radiation Therapy Oncology Group Trial J Clin Oncol 2004; 22: Aaltomaa S, Karja V, Lipponen P, Isotalo T, Kankkunen JP, Talja M et al: Expression of Ki-67, cyclin D1 and apoptosis markers correlated with survival in prostate cancer patients treated by radical prostatectomy. Anticancer Res 2006; 26: Shimizu Y, Segawa T, Inoue T, Shiraishi T, Yoshida T, Toda Y et al: Increased Akt and phosphorylated Akt expression are associated with malignant biological features of prostate cancer in Japanese men. BJU Int 2007; 100: Wang Y, Kreisberg JI and Ghosh PM: Cross-talk between the androgen receptor and the phosphatidylinositol 3-kinase/Akt pathway in prostate cancer. Curr Cancer Drug Targets 2007; 7: McCall P, Gemmell LK, Mukherjee R, Bartlett J and Edwards J: Phosphorylation of the androgen receptor is associated with reduced survival in hormone-refractory prostate cancer patients. Br J Cancer 2008; 98: Khatami A, Pihl CG, Norrby K, Hugosson J and Damber JE: Is tumor vascularity in prostate core biopsies a predictor of PSA recurrence after radical prostatectomy? Acta Oncol 2005; 44: 362.