Development and Validation of a Clinical Prognostic Stage Group System for Nonmetastatic Prostate Cancer Using Disease-Specific Mortality Results From the International Staging Collaboration for Cancer of the Prostate | Oncology | JAMA Oncology | JAMA Network
[Skip to Navigation]
Sign In
Figure 1.  Clinical Prognostic Stage Group Score System for Prostate Cancer–Specific Mortality (PCSM) Prediction in the Validation Cohort
Clinical Prognostic Stage Group Score System for Prostate Cancer–Specific Mortality (PCSM) Prediction in the Validation Cohort

Score stage group IA (0 points) included 1261 patients from the validation cohort (12.9%; 10-year PCSM estimate, 0.3%); Score stage group IB (1-2 points), 2501 patients (25.6%; 10-year PCSM estimate, 0.8%); Score stage group IC (3-4 points), 1901 patients (19.5%; 10-year PCSM estimate, 2.0%); Score stage group IIA (5-6 points), 1554 patients (15.9%; 10-year PCSM estimate, 3.3%); Score stage group IIB (7-8 points), 1208 patients (12.4%; 10-year PCSM estimate, 4.4%); Score stage group IIC (9-10 points), 719 patients (7.4%; 10-year PCSM estimate, 9.5%); Score stage group IIIA (11-12 points), 354 patients (3.6%; 10-year PCSM estimate, 11.7%); Score stage group IIIB (13-16 points), 248 patients (2.5%; 10-year PCSM estimate, 21.2%); and Score stage group IIIC (≥17 points), 23 patients (0.2%; 10-year PCSM estimate, 40.0%).

Figure 2.  Reclassification of the Current American Joint Committee on Cancer (AJCC) 8th Edition Staging System Using the Proposed Clinical Prognostic Staging Group Score System in the Validation Cohort
Reclassification of the Current American Joint Committee on Cancer (AJCC) 8th Edition Staging System Using the Proposed Clinical Prognostic Staging Group Score System in the Validation Cohort

Color coding represents percentage of patients with current AJCC 8th edition stage that are reclassified to each new Score stage. Red is 0%. Orange, yellow, and green represent increasing percentages of reclassification.

Figure 3.  Model Discrimination (C Index) of Models Over Time for Prediction of Prostate Cancer–Specific Mortality in the Validation Cohort
Model Discrimination (C Index) of Models Over Time for Prediction of Prostate Cancer–Specific Mortality in the Validation Cohort

Higher values of C index indicate better discriminatory performance. Error bars indicate bootstrap 95% CIs. AJCC indicates American Joint Committee on Cancer.

Table 1.  STAR-CAP Cohort Characteristics
STAR-CAP Cohort Characteristics
Table 2.  Fine-Gray Regression Model of Prostate Cancer–Specific Mortality in Training Cohort
Fine-Gray Regression Model of Prostate Cancer–Specific Mortality in Training Cohort
Supplement.

eMethods. STAR-CAP Cohort and Treatment Details

eTable 1. American Joint Committee on Cancer Precision Medicine Core Inclusion and Exclusion Criteria for Evaluation of Prognostic Models

eTable 2. Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) Checklist

eTable 3. Patients With Missing Data Elements in STAR-CAP Cohort

eTable 4. Univariable Fine and Gray Regression Models of Prostate Cancer Specific Mortality in Training Cohort

eTable 5. Full Fine and Gray Score System With All Regression Coefficients, and Model Intercept or Baseline Survival at a Given Time Point

eTable 6. 10-year C Index Within Subgroups for Prediction of Prostate Cancer‒Specific Mortality in the Validation Set

eTable 7. Surveillance, Epidemiology, and End Results Cohort

eTable 8. C Index and Index of Predictive Accuracy for Prediction of Prostate Cancer-Specific Mortality in the SEER Cohort

eTable 9. Multivariable Fine and Gray Regression Model of Prostate Cancer‒Specific Mortality With Interactions in Training Cohort

eTable 10. C Index and Index of Predictive Accuracy for Prediction of Prostate Cancer‒Specific Mortality in the Validation Cohort

eFigure 1. Calibration Plot for Predictions of 3-, 5-, 10-, and 15-Year Prostate Cancer‒Specific Mortality in Validation Cohort

eFigure 2. Index of Prognostic Accuracy of Models for Prediction of PCSM in the Validation Cohorts

eFigure 3. 10-Year C Index for Prediction of Prostate Cancer‒Specific Mortality in the Validation Set Compared to National Comprehensive Cancer Network (NCCN) Risk Categorization and the Prostate Cancer Risk Assessment CAPRA Score

1.
Klotz  L, Vesprini  D, Sethukavalan  P,  et al.  Long-term follow-up of a large active surveillance cohort of patients with prostate cancer.   J Clin Oncol. 2015;33(3):272-277. doi:10.1200/JCO.2014.55.1192 PubMedGoogle ScholarCrossref
2.
Tosoian  JJ, Mamawala  M, Epstein  JI,  et al.  Active surveillance of grade group 1 prostate cancer: long-term outcomes from a large prospective cohort.   Eur Urol. 2020;77(6):675-682. doi:10.1016/j.eururo.2019.12.017 PubMedGoogle ScholarCrossref
3.
Hamdy  FC, Donovan  JL, Lane  JA,  et al; ProtecT Study Group.  10-year outcomes after monitoring, surgery, or radiotherapy for localized prostate cancer.   N Engl J Med. 2016;375(15):1415-1424. doi:10.1056/NEJMoa1606220 PubMedGoogle ScholarCrossref
4.
Lawton  CA, Lin  X, Hanks  GE,  et al.  Duration of androgen deprivation in locally advanced prostate cancer: long-term update of NRG oncology RTOG 9202.   Int J Radiat Oncol Biol Physics. 2017;98(2):296-303. doi:10.1016/j.ijrobp.2017.02.004Google Scholar
5.
Rosenthal  SA, Hu  C, Sartor  O,  et al.  Effect of chemotherapy with docetaxel with androgen suppression and radiotherapy for localized high-risk prostate cancer: the randomized phase III NRG Oncology RTOG 0521 trial.   J Clin Oncol. 2019;37(14):1159-1168. doi:10.1200/JCO.18.02158 PubMedGoogle ScholarCrossref
6.
Shipley  WU, Seiferheld  W, Lukka  HR,  et al; NRG Oncology RTOG.  Radiation with or without antiandrogen therapy in recurrent prostate cancer.   N Engl J Med. 2017;376(5):417-428. doi:10.1056/NEJMoa1607529 PubMedGoogle ScholarCrossref
7.
Brierley  J, O’Sullivan  B, Asamura  H,  et al.  Global consultation on cancer staging: promoting consistent understanding and use.   Nat Rev Clin Oncol. 2019;16(12):763-771. doi:10.1038/s41571-019-0253-x PubMedGoogle ScholarCrossref
8.
Pisansky  TM, Cha  SS, Earle  JD,  et al.  Prostate-specific antigen as a pretherapy prognostic factor in patients treated with radiation therapy for clinically localized prostate cancer.   J Clin Oncol. 1993;11(11):2158-2166. doi:10.1200/JCO.1993.11.11.2158 PubMedGoogle ScholarCrossref
9.
Pisansky  TM, Kahn  MJ, Rasp  GM, Cha  SS, Haddock  MG, Bostwick  DG.  A multiple prognostic index predictive of disease outcome after irradiation for clinically localized prostate carcinoma.   Cancer. 1997;79(2):337-344. doi:10.1002/(SICI)1097-0142(19970115)79:2<337::AID-CNCR17>3.0.CO;2-1 PubMedGoogle ScholarCrossref
10.
Pisansky  TM, Kahn  MJ, Bostwick  DG.  An enhanced prognostic system for clinically localized carcinoma of the prostate.   Cancer. 1997;79(11):2154-2161. doi:10.1002/(SICI)1097-0142(19970601)79:11<2154::AID-CNCR13>3.0.CO;2-V PubMedGoogle ScholarCrossref
11.
Kattan  MW, Hess  KR, Amin  MB,  et al; members of the AJCC Precision Medicine Core.  American Joint Committee on Cancer acceptance criteria for inclusion of risk models for individualized prognosis in the practice of precision medicine.   CA Cancer J Clin. 2016;66(5):370-374. doi:10.3322/caac.21339 PubMedGoogle ScholarCrossref
12.
Buyyounouski  MK, Choyke  PL, McKenney  JK,  et al. Prostate cancer—major changes in the American Joint Committee on Cancer eighth edition cancer staging manual.  CA Cancer J Clin. 2017;67(3):245-253. doi:10.3322/caac.21391
13.
Collins  GS, Reitsma  JB, Altman  DG, Moons  KGM.  Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement.   BMJ. 2015;350:g7594-g7594. doi:10.1136/bmj.g7594 PubMedGoogle ScholarCrossref
14.
Surveillance, Epidemiology and End Results. Prostate with watchful waiting database (2010-2015). National Cancer Institute; 2018. Accessed February 12, 2020. https://seer.cancer.gov/seerstat/databases/prostate-ww/index.html
15.
Dess  RT, Hartman  HE, Mahal  BA,  et al.  Association of black race with prostate cancer–specific and other-cause mortality.   JAMA Oncol. 2019;5(7):975-983. doi:10.1001/jamaoncol.2019.0826 PubMedGoogle ScholarCrossref
16.
Kupelian  P, Thames  H, Levy  L,  et al.  Year of treatment as independent predictor of relapse-free survival in patients with localized prostate cancer treated with definitive radiotherapy in the PSA era.   Int J Radiat Oncol Biol Phys. 2005;63(3):795-799. doi:10.1016/j.ijrobp.2005.03.029Google ScholarCrossref
17.
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(6):1938-1942. doi:10.1097/01.ju.0000158155.33890.e7 PubMedGoogle ScholarCrossref
18.
Ishwaran  H, Gerds  TA, Kogalur  UB, Moore  RD, Gange  SJ, Lau  BM.  Random survival forests for competing risks.   Biostatistics. 2014;15(4):757-773. doi:10.1093/biostatistics/kxu010 PubMedGoogle ScholarCrossref
19.
National Comprehensive Cancer Network. Prostate Cancer. Version 4.2019 2020. Published August 19, 2019. Accessed February 13, 2020. https://www.nccn.org/professionals/physician_gls/default.aspx#prostate
20.
Hermanek  P, Hutter  RV, Sobin  LH.  Prognostic grouping: the next step in tumor classification.   J Cancer Res Clin Oncol. 1990;116(5):513-516. doi:10.1007/BF01613004 PubMedGoogle ScholarCrossref
21.
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(11):969-974. doi:10.1001/jama.280.11.969 PubMedGoogle ScholarCrossref
22.
Bruinsma  SM, Bangma  CH, Carroll  PR,  et al; Movember GAP3 consortium.  Active surveillance for prostate cancer: a narrative review of clinical guidelines.   Nat Rev Urol. 2016;13(3):151-167. doi:10.1038/nrurol.2015.313 PubMedGoogle ScholarCrossref
23.
Tosco  L, Briganti  A, D’amico  AV,  et al.  Systematic review of systemic therapies and therapeutic combinations with local treatments for high-risk localized prostate cancer.   Eur Urol. 2019;75(1):44-60. doi:10.1016/j.eururo.2018.07.027 PubMedGoogle ScholarCrossref
24.
Zumsteg  ZS, Zelefsky  MJ, Woo  KM,  et al.  Unification of favourable intermediate-, unfavourable intermediate-, and very high-risk stratification criteria for prostate cancer.   BJU Int. 2017;120(5B):E87-E95. doi:10.1111/bju.13903 PubMedGoogle ScholarCrossref
25.
Spratt  DE, Zhang  J, Santiago-Jiménez  M,  et al.  Development and validation of a novel integrated clinical-genomic risk group classification for localized prostate cancer.   J Clin Oncol. 2018;36(6):581-590. doi:10.1200/JCO.2017.74.2940 PubMedGoogle ScholarCrossref
26.
Fulgham  PF, Rukstalis  DB, Turkbey  IB,  et al.  AUA policy statement on the use of multiparametric magnetic resonance imaging in the diagnosis, staging and management of prostate cancer.   J Urol. 2017;198(4):832-838. doi:10.1016/j.juro.2017.04.101 PubMedGoogle ScholarCrossref
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    Original Investigation
    October 22, 2020

    Development and Validation of a Clinical Prognostic Stage Group System for Nonmetastatic Prostate Cancer Using Disease-Specific Mortality Results From the International Staging Collaboration for Cancer of the Prostate

    Author Affiliations
    • 1Department of Radiation Oncology, University of Michigan School of Medicine, Ann Arbor
    • 2School of Public Health, University of Colorado, Denver
    • 3Department of Radiation Oncology, Memorial Sloan Kettering Cancer Center, New York, New York
    • 4Division of Urology, Department of Surgery, Samuel Oschin Comprehensive Cancer Institute, Cedars-Sinai Medical Center, Los Angeles, California
    • 5Durham VA Medical Center, Durham, North Carolina
    • 6Harvard Radiation Oncology Program, Massachusetts General Hospital, Boston
    • 7Department of Urology, University of California, San Francisco, Helen Diller Family Comprehensive Cancer Center
    • 8Department of Radiation Oncology, Mayo Clinic, Rochester, Minnesota
    • 9Department of Radiation Oncology, Fox Chase Cancer Center, Philadelphia, Pennsylvania
    • 10Section of Urology, Medical College of Georgia, Augusta, Georgia
    • 11Division of Urology, Department of Surgery, Oregon Health and Science University, Portland
    • 12Department of Urology, University of California, Los Angeles, School of Medicine
    • 13Department of Urology, University of California, San Diego, Health System
    • 14Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University, Baltimore, Maryland
    • 15Department of Biostatistics, University of Michigan, Ann Arbor
    • 16Department of Pathology, University of Michigan, Ann Arbor
    • 17Department of Urology, University of Michigan, Ann Arbor
    • 18Department of Radiation Oncology, Brigham and Women’s Hospital, Dana-Farber Cancer Institute, Boston, Massachusetts
    • 19Department of Radiation Oncology, University of California, San Francisco
    • 20Department of Medicine, University of California, San Francisco
    • 21Department of Urology, Mayo Clinic, Rochester, Minnesota
    • 22Department of Radiation Oncology, Penn State Cancer Institute, Hershey, Pennsylvania
    • 23Department of Oncology, Queen’s University, Kingston, Ontario, Canada
    • 24Radiation Medicine Program, Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada
    • 25Department of Radiation Oncology, University of Toronto, Toronto, Ontario, Canada
    • 26Department of Surgical Oncology, Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada
    • 27Division of Urology, University of Toronto, Toronto, Ontario, Canada
    • 28Department of Urology, Scientific Institute and University Vita–Salute San Raffaele Hospital, Milan, Italy
    • 29Department of Quantitative Health Sciences, Cleveland Clinic Foundation, Cleveland, Ohio
    JAMA Oncol. 2020;6(12):1912-1920. doi:10.1001/jamaoncol.2020.4922
    Key Points

    Question  Can an improved clinical prognostic stage group system for nonmetastatic prostate cancer be developed?

    Findings  This cohort study used data from 19 684 men with nonmetastatic prostate cancer from 55 centers across the United States, Canada, and Europe to develop and perform 2 independent validations of a clinical prognostic stage group system that meets criteria set forth by the American Joint Committee on Cancer Precision Medicine Core committee. The proposed system outperformed the existing American Joint Committee on Cancer system and commonly used risk-stratification systems.

    Meaning  These findings suggest that the new prognostic stage group system may be used to inform therapeutic decision-making and future clinical trial design.

    Abstract

    Importance  In 2016, the American Joint Committee on Cancer (AJCC) established criteria to evaluate prediction models for staging. No localized prostate cancer models were endorsed by the Precision Medicine Core committee, and 8th edition staging was based on expert consensus.

    Objective  To develop and validate a pretreatment clinical prognostic stage group system for nonmetastatic prostate cancer.

    Design, Setting, and Participants  This multinational cohort study included 7 centers from the United States, Canada, and Europe, the Shared Equal Access Regional Cancer Hospital (SEARCH) Veterans Affairs Medical Centers collaborative (5 centers), and the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) registry (43 centers) (the STAR-CAP cohort). Patients with cT1-4N0-1M0 prostate adenocarcinoma treated from January 1, 1992, to December 31, 2013 (follow-up completed December 31, 2017). The STAR-CAP cohort was randomly divided into training and validation data sets; statisticians were blinded to the validation data until the model was locked. A Surveillance, Epidemiology, and End Results (SEER) cohort was used as a second validation set. Analysis was performed from January 1, 2018, to November 30, 2019.

    Exposures  Curative intent radical prostatectomy (RP) or radiotherapy with or without androgen deprivation therapy.

    Main Outcomes and Measures  Prostate cancer–specific mortality (PCSM). Based on a competing-risk regression model, a points-based Score staging system was developed. Model discrimination (C index), calibration, and overall performance were assessed in the validation cohorts.

    Results  Of 19 684 patients included in the analysis (median age, 64.0 [interquartile range (IQR), 59.0-70.0] years), 12 421 were treated with RP and 7263 with radiotherapy. Median follow-up was 71.8 (IQR, 34.3-124.3) months; 4078 (20.7%) were followed up for at least 10 years. Age, T category, N category, Gleason grade, pretreatment serum prostate-specific antigen level, and the percentage of positive core biopsy results among biopsies performed were included as variables. In the validation set, predicted 10-year PCSM for the 9 Score groups ranged from 0.3% to 40.0%. The 10-year C index (0.796; 95% CI, 0.760-0.828) exceeded that of the AJCC 8th edition (0.757; 95% CI, 0.719-0.792), which was improved across age, race, and treatment modality and within the SEER validation cohort. The Score system performed similarly to individualized random survival forest and interaction models and outperformed National Comprehensive Cancer Network (NCCN) and Cancer of the Prostate Risk Assessment (CAPRA) risk grouping 3- and 4-tier classification systems (10-year C index for NCCN 3-tier, 0.729; for NCCN 4-tier, 0.746; for Score, 0.794) as well as CAPRA (10-year C index for CAPRA, 0.760; for Score, 0.782).

    Conclusions and Relevance  Using a large, diverse international cohort treated with standard curative treatment options, a proposed AJCC-compliant clinical prognostic stage group system for prostate cancer has been developed. This system may allow consistency of reporting and interpretation of results and clinical trial design.

    Introduction

    Prostate cancer is one of the most common cancers diagnosed in men worldwide. Localized prostate cancer can be indolent and does not always require immediate intervention.1-3 In contrast, some men harbor aggressive localized disease with a high risk of prostate cancer–specific mortality (PCSM), even with multimodality treatment.4-6 No prospectively validated predictive biomarkers are available for clinical use in localized prostate cancer, and thus treatment recommendations are based on prognostic variables. Given the substantial heterogeneity in disease severity, accurate estimates of outcomes are critical to make informed therapeutic decisions.

    For more than 6 decades, the American Joint Committee on Cancer (AJCC) has grouped patients by stages of disease (I-IV), which are used to estimate prognosis.7 Categorizations have been based on the local extent of the primary tumor (T category), along with anatomical patterns of spread to lymph nodes (N category) and distant sites of metastasis (M category). For prostate cancer, nonanatomical factors, such as Gleason grade in 2002 and prostate-specific antigen (PSA) level in 2010, were incorporated over time into AJCC prognostic stage groups based on data supporting their significance.8-10 Nevertheless, prostate cancer remains one of the few cancers for which such stage groupings have not been adopted in national guidelines or the design of clinical trials.

    As part of its most recent update in 2016, the AJCC adopted a strategy to incorporate more advanced statistical models to better personalize prognostic estimates. To do so, its Precision Medicine Core committee established criteria for model evaluation.11 The AJCC prostate cancer expert panel followed these guidelines and evaluated 15 prognostic models published from 2011 to 2015. Two models met criteria, but both were limited to patients with metastatic disease.12 As a result, no statistical models for localized disease were incorporated into the current 8th edition.

    To address these issues, we established an international staging collaboration for cancer of the prostate (STAR-CAP). We sought to develop and validate a pretreatment clinical prognostic stage group system using readily available clinical factors. The aim was to accurately predict PCSM and improve on the existing AJCC expert consensus estimates.

    Methods
    Patients

    Our study was designed to adhere to prognostic risk model guidance set forth by the AJCC Precision Medicine Core committee, and the modeling was built according to Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guideline (eTables 1 and 2 in the Supplement).13 Institutional review board approval was obtained from the University of Michigan, and data sharing agreements were signed with each participating institution. All data were deidentified and waiver of informed consent was granted by the institutional review boards.

    The STAR-CAP data were collected from 7 North American and European cancer centers, the Shared Equal Access Regional Cancer Hospital (SEARCH) collaborative (5 medical centers within the Veterans Affairs Medical Centers system), and the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) registry (43 academic and community practices across the United States). The cohort included patients with prostate cancer treated with curative intent from January 1, 1992, through December 31, 2013; follow-up continued through December 31, 2017. Patients with metastatic disease (M1), with pretreatment serum PSA levels of greater than 200 or less than 0.5 ng/mL (to convert to μg/L, multiply by 1.0), or who received neoadjuvant androgen deprivation therapy (ADT) before radical prostatectomy (RP) were excluded. Surgical procedures included RP with or without pelvic lymph node dissection. Patients undergoing radiotherapy received external beam radiotherapy, brachytherapy, or a combination of external beam radiotherapy and brachytherapy with or without ADT. eMethods in the Supplement provides additional diagnostic, radiotherapy, and ADT details. The Surveillance, Epidemiology, and End Results (SEER) database was used as a second independent and external validation data set.14 Patients in SEER were treated from January 1, 2010, to December 31, 2015, and were managed with active surveillance, watchful waiting, RP, radiotherapy, or primary ADT.

    Procedures

    The primary end point was time to PCSM, defined as the time from treatment initiation (RP, radiotherapy, or ADT) to death due to prostate cancer censored at last follow-up. Follow-up was per institution, which generally included a PSA level measurement every 3 to 12 months for 5 years and annually thereafter. Imaging for metastasis was also per individual institution, typically triggered by symptoms or continued increase in PSA levels. Death due to prostate cancer was per individual institution, most often defined as metastatic and/or castration-resistant prostate cancer at time of death without an alternative cause.

    All variables used in the prediction model development were present at baseline before treatment. Clinical T category (T1a-T4) and clinical N category (N0 or N1) were defined per the AJCC 8th edition to allow direct comparisons.12 Primary and secondary Gleason scores were required. Gleason scores of ≤3 + 3 = 6 through 5 + 5 = 10 were included; International Society of Urologic Pathologists grade group definitions were considered, but analysis was not limited to the new categorizations (eg, primary pattern 5 evaluated separately from Gleason score of 4 + 5). Age, pretreatment serum PSA level, and percentage of positive core biopsy results among biopsies performed were evaluated as continuous and categorical variables as described below. Race and treatment type were not included in the models15 but were included in subgroup analyses of model performance.

    Statistical Analysis

    Data were analyzed from January 1, 2018, to November 30, 2019. All available data from individual cohorts meeting inclusion criteria were merged, and the STAR-CAP data set was randomly split 50-50 into training and validation data sets. This allowed for sufficient model training and a validation set large enough to evaluate model performance within treatment subgroups. Follow-up was administratively censored at 15 years. The 2 primary statisticians (K.S. and M.J.S.) were blinded to the STAR-CAP and SEER validation sets until the model development was finalized using the training data set.

    Our first aim was to create an optimized categorical system using a points-based approach to create clinical stage groupings. We used Fine-Gray regression models with age, T category, Gleason grade, percentage of positive core biopsy results, and PSA level as covariates. Noncancer death was treated as a competing risk. Treatment year was modeled to account for evolution in Gleason grading and stage migration but was not included in the final Score system.16 To identify thresholds for categorizing continuous covariables, regression splines and indicator variables for each decile were used in the training data. This process created groups for each categorical variable (eg, T category 2c-3a) and bins for each continuous variable (eg, age >70 years). Each group or bin was assigned a point value by dividing parameter estimate (log hazard ratio) by 0.30 and rounding to the nearest integer value, similar to the Cancer of the Prostate Risk Assessment (CAPRA) method.17 Total points were summed for each patient. Point totals with similar PCSM risk were grouped into 9 final groupings for the Score staging system.

    The second aim was to develop risk prediction models with individualized PCSM predictions. We considered 2 methods: (1) a Fine-Gray model, referred to as the interaction model, which treated continuous variables as linear, allowed for covariable interactions, and used stepwise selection based on Akaike information criterion, and (2) a random survival forest for competing risks.18 These models also served as a basis to compare performance of the categorical Score staging system.

    The performance of the newly developed risk prediction systems (Score, interaction, and random survival forest) were compared with the existing AJCC 8th edition. Discriminatory ability was estimated using the concordance index (C index); calibration, using calibration plots; and overall performance, using the Index of Prediction Accuracy. We also compared the Score staging system with the National Comprehensive Cancer Network (NCCN) 3- and 4-tier risk stratification system, given its common use, as well as CAPRA.19 All analyses were performed using R, version 3.5.3 (R Project for Statistical Computing), software. The eMethods in the Supplement includes more statistical method details. Two-sided P < .05 indicated statistical significance.

    Results
    Model Participants

    From 1992 to 2013, 19 684 patients (median age, 64.0 [interquartile range {IQR}, 59.0-70.0] years) with localized prostate cancer were collected with complete clinical variables (Table 1). An additional 8337 men had data collected with missing clinical variables required for the Score system. The most common missing element was percentage of positive core biopsy results among biopsies performed (n = 6246) (eTable 3 in the Supplement). Overall outcomes and follow-up distributions were similar: 166 of 8329 (2.0%) men in the missing cohort died of prostate cancer and 1021 of 8329 (12.3%) died of other causes, with a median follow-up of 72.3 (range, 32.0-133.0) months, compared with 450 of 19 684 (2.3%) and 2522 of 19 684 (12.8%), respectively, in the nonmissing cohort with a median follow up of 71.8 (IQR, 34.3-124.3) months. A total of 4078 patients (20.7%) were followed for at least 10 years.

    Age, Gleason grade, percentage of positive core biopsy results, T and N categories, and PSA levels were similar in the training and validation cohorts. As for primary treatment, 12 241 patients (63.1%) underwent RP and 7263 (36.5%) received external beam radiotherapy, brachytherapy, or a combination. Of those receiving radiotherapy, 3656 (50.3%) received radiotherapy alone, 2178 (30.0%) received short-term ADT (<12 months), 696 (9.6%) received long-term ADT (≥12 months), and 733 (10.1%) received ADT of an unknown duration. After excluding those with missing variables, the training cohort had 9915 patients with 218 PCSM events, and the validation cohort had 9769 patients with 232 PCSM events.

    Model Development and Specification

    Table 2 shows the Fine-Gray multivariable model with categorized covariables used to generate the points-based Score staging system. The cumulative baseline subdistribution hazard (reference group for all categorical covariates, treatment year 1992) at time t years, β0(t), is given as β0(3) = 0.000856, β0(5) = 0.00241, β0(10) = 0.00838, and β0(15) = 0.0163. The baseline PCSM(t) = {1 − exp[−B0(t)]} is then given by 0.0855% at 3 years, 0.241% at 5 years, 0.835% at 10 years, and 1.616% at 15 years. Treatment year was included as a continuous fixed effect (centered at year 1992) to account for evolution in Gleason grading and stage migration and other changes over time (coefficient, −0.055 [SE, 0.015]; subdistribution hazard ratio [sHR], 0.95 [95% CI, 0.92-0.98]; P < .001). Age demonstrated a bimodal risk distribution; those who were 50 years or younger and older than 70 years had increased PCSM (sHR, 1.55 [95% CI, 0.74-3.24] and 1.32 [95% CI, 0.97-1.79], respectively) compared with those aged 51 to 70 years (reference category). Bilateral prostate gland involvement (T2c) was grouped with extraprostatic extension (T3a) in the model based on similar risk. Clinical node-positive status had a substantially increased PCSM hazard compared with clinical node negative (sHR, 10.99 [95% CI, 2.85-42.38]; P < .001). Gleason grade was also highly prognostic; relative to Gleason grade 3 + 3, those with a primary pattern grade of 5 had the worst outcomes (sHR, 12.65 [95% CI, 7.05-22.71]; P < .001). The percentage of positive core biopsy results was prognostic, independent of Gleason grade. The categorical breakdown was greater than 50% to 75% (sHR, 1.61 [95% CI, 1.09-2.37]; P = .02) and greater than 75% to 100% (sHR, 2.14 [95% CI, 1.51-3.03]; P < .001) compared with 50% or less (reference category). Level of PSA was also prognostic, and a higher level conferred a worse prognosis; 5 groups were included based on threshold values of greater than 6 to 10 ng/mL (sHR, 1.42 [95% CI, 0.95-2.14]; P = .09), greater than 10 to 20 ng/mL (sHR, 1.85 [95% CI, 1.19-2.86]; P = .006), greater than 20 to 50 ng/mL (sHR, 2.37 [95% CI, 1.44-3.91]; P < .001), and greater than 50 ng/mL (sHR, 3.62 [95% CI, 1.92-6.85]; P < .001), with 6 ng/mL or less as the reference category.

    Each risk covariable was assigned a point value of 0 to 8 based on the regression coefficient transformation (Table 2). The total possible points ranged from 0 to 27. Point totals with similar PCSM predictions in the training cohort were combined, which resulted in 9 final categorical groups (0, 1-2, 3-4, 5-6, 7-8, 9-10, 11-12, 13-16, and ≥17, labeled new stages IA-IIIC) (Figure 1 and eTable 5 in the Supplement). For example, a 71-year-old man (1 point) with cT2a (1 point), N0 (0 points), Gleason grade 4 + 3 (5 points), 7 of 12 positive core results (2 points), and a PSA level of 9 ng/mL (1 point) would have a total of 10 points (stage IIC) and a 10-year PCSM estimate of 9.5%. An online calculator of the STAR-CAP Clinical Prognostic Stage Grouping is given at http://www.star-cap.org.

    The lowest Score group was 0 (new stage IA) (1261 [12.9%] of the validation cohort) and had an estimated 10-year PCSM of 0.3%. The highest Score group was at least 17 (new stage IIIC) (23 [0.2%] of the validation cohort) and an estimated 10-year PCSM of 40.0%. The Score staging system reclassified patients across every stage of the existing AJCC 8th edition. For example, in the validation cohort, of those with stage IIIA in the AJCC 8th edition (T1-2N0M0; PSA level ≥20 ng/mL; grade group 1-4 [n = 447]), 98 men (21.9%) were Score stage IC (10-year PCSM risk of 2.0%), whereas 105 (23.5%) were Score stage IIIA or IIIB (10-year PCSM risk >10%) (Figure 2).

    Score Staging System Performance

    Discrimination, calibration, and overall performance metrics of the Score staging system compared favorably with the existing AJCC 8th edition. The Score 10-year C index of 0.796 (95% CI, 0.760-0.828) exceeded the AJCC 8th edition C index of 0.757 (95% CI, 0.719-0.792) (Figure 3). Both the Score and AJCC 8th edition staging systems were well calibrated for predicting PCSM at various time points (eFigure 1 in the Supplement). The Index of Prediction Accuracy measure of overall model performance exceeded that of AJCC 8th edition for all points (eFigure 2 in the Supplement). The Score staging system outperformed the existing AJCC staging system across age, race, and treatment subgroups (eTable 6 in the Supplement). The Score staging also outperformed current NCCN 3- and 4-tier classification systems (10-year C index for NCCN 3-tier, 0.729; for NCCN 4-tier, 0.746; for Score, 0.794) as well as CAPRA (10-year C index for CAPRA, 0.760; for Score, 0.782) and a modification of CAPRA to include T3b and T4 and clinical node-positive categories (10-year C index for modified CAPRA, 0.774; for Score, 0.796) (eFigure 3 in the Supplement).

    The SEER data set is summarized in eTable 7 in the Supplement. The total cohort included 125 575 men with nonmetastatic disease and complete case information. Median follow-up was 35 (IQR, 16-53) months. In contrast to the STAR-CAP data set, 31 886 (25.4%) received no local therapy. The performance of the new Score staging system was superior to the AJCC 8th edition in this data set as well. The 5-year C index was 0.838 compared with 0.817 for the existing AJCC 8th edition. The Index of Prediction Accuracy measure of overall model performance was also superior to AJCC 8th edition (eTable 8 in the Supplement).

    Individualized Risk Prediction Model Performance

    The interaction (eTable 9 in the Supplement) and random survival forest models had similar C indexes to the Score staging system in the validation data at 10 years (interaction, 0.798; random forest, 0.797; Score, 0.796) with evidence of improved performance at 15 years (eTable 10 in the Supplement). Both the interaction and random forest models, however, had a lower Index of Prediction Accuracy at 10 years compared with the Score staging system in the validation data owing to poorer calibration. Thus, the categorical Score staging system offered simplicity and ease of use without sacrificing performance.

    Discussion

    The AJCC has long encouraged efforts to refine cancer classification.20 The development of prognostic grouping models has provided the conceptual framework for precision oncology. Early efforts established a footing for such research in prostate cancer,10,21 but to date, a prognostication model for localized disease has not yet fulfilled AJCC criteria.12 Our international consortium used a large and diverse cohort of men treated with standard curative therapies with long-term follow-up to develop and validate a well-performing, AJCC-compliant clinical prognostic group staging system using variables readily available across diverse practice settings. The new Score grouping system provides a better fit to PCSM as an outcome compared with the existing prognostic stage groups from the AJCC 8th edition, thus providing important data to inform patients of their expected outcome. The Score grouping system also provides a common foundation for future clinical trial designs that may originate in different research settings, such as in the various National Clinical Trials Network groups.

    Prognosis was highly heterogeneous. The difference in estimated 10-year PCSM between the lowest and highest Score stages was 100-fold. Score stages IA and IB (3762 patients [38.5%] of the cohort) had a 10-year PCSM estimate of less than 1% with treatment. These patients are now routinely managed with active surveillance.22 On the other end of the spectrum, Score stages IIC to IIIC (1344 [13.8%] of the cohort) had an estimated 10-year PCSM that exceeded 10% with standard-of-care treatment. These men may be candidates for treatment intensification.23

    The 10-year C index of the Score staging system (0.796) exceeded that of the AJCC 8th edition (0.757). This improved performance was evident across age, race, and treatment subgroups. The Score staging system also outperformed in a second SEER validation data set. We identified patients within the same AJCC prognostic stage group with dramatically different estimated PCSM risk. Our system also outperformed the NCCN 3- and 4-tier risk groups for the clinically meaningful end point of PCSM. The NCCN adopted the risk groups defined by D’Amico et al,21 which were initially based on a biochemical failure end point. Over time, NCCN risk group subdivision with variable validation and a nonharmonious incorporation of clinical factors has been undertaken. Subsequent work has shown many of these subgroups have similar long-term outcomes.24 In addition, age, an important prognostic variable, is not included in NCCN risk stratification.19 Our work also builds on the CAPRA system, which was designed based on patients undergoing surgery in a similar manner to the Score staging system.17 Given the size of our cohort and the inclusion of patients treated with radiotherapy, we now include patients with clinical T3b-T4 and node-positive disease. Using more granular cut points, the Score system also outperformed CAPRA.

    Although our interaction and random forest risk models provide individualized prognosis estimates, both performed similarly to the categorical Score staging system. This suggests that clinical and pathologic prognostic information is largely captured in the Score system. Thus, to further improve the performance of our system, additional variables providing independent information will be needed.

    Previously, Spratt et al25 demonstrated that prognostic gene expression classifiers consistently improve prognostic performance over clinicopathologic variables alone. Although genomics and other molecular diagnostics could improve prognostication in subsets of patients with localized prostate cancer, the staging system developed herein is generalizable and based on readily available prognostic factors routinely acquired in initial workup. Genomic tests are not available in many countries, so ours is a system that may be used worldwide. Similarly, magnetic resonance imaging is increasingly used for initial staging of prostate cancer; however, to our knowledge there are no validated models to date demonstrating that magnetic resonance imaging findings independently risk stratify PCSM.26 Furthermore, clinical factors alone are likely sufficient in certain patients, because approximately 40% of the STAR-CAP cohort had a PCSM risk of less than 1% at 10 years. In this cohort, genomics and/or magnetic resonance imaging may have limited prognostic value when active intervention is undertaken.

    Limitations

    There are several limitations to discuss. The T category was based on physical examination results; N category, on computed tomographic findings; and Gleason grade and percentage of positive core biopsy results, on systematic transrectal ultrasonographic biopsy findings. Diagnostic pathways are evolving to include multiparametric magnetic resonance imaging–based targeted biopsy findings and local tumor staging as well as molecular imaging for distant staging. Only with long-term follow-up will we understand how these additions add to the prognostic value of the STAR-CAP data set. Gleason grading was not centrally reviewed. Follow-up and PCSM definitions were per institutional standards and not prospectively defined. Although this results in heterogeneity, it could also be considered as a strength, given that the system performed well across both validation cohorts. The discretizing of variables required to create the STAR-CAP clinical prognostic stage groupings may have resulted in loss of information. However, the performance was comparable to the continuous models. Finally, although the STAR-CAP data set identifies patients with increased PCSM risk who receive standard treatments, no patient in this data set underwent conservative management. However, the Prostate Testing for Cancer and Treatment (ProtecT) trial has not shown a difference in PCSM between active monitoring and radical treatment in patients at low and intermediate risk with 10 years of follow-up.3

    Conclusions

    We developed and validated a proposed AJCC-compliant clinical prognostic group staging system that can be used globally for patients with nonmetastatic prostate cancer and has better performance than the current AJCC 8th edition. Our Score system represents high-quality evidence to support endorsement as a new staging system for prostate cancer.

    Back to top
    Article Information

    Accepted for Publication: July 20, 2020.

    Corresponding Authors: Daniel E. Spratt, MD (sprattda@med.umich.edu), and Robert T. Dess, MD (rdess@med.umich.edu), Department of Radiation Oncology, University of Michigan School of Medicine, 1500 E Medical Center Dr, Ann Arbor, MI 48109.

    Published Online: October 22, 2020. doi:10.1001/jamaoncol.2020.4922

    Author Contributions: Drs Dess and Suresh contributed equally as first authors. Drs Schipper and Spratt contributed equally as senior authors. Drs Dess and Spratt had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

    Concept and design: Dess, Suresh, Jackson, Kaffenberger, Feng, Tran, Stish, Moraes, Finelli, Gandaglia, Briganti, Carroll, Schipper, Spratt.

    Acquisition, analysis, or interpretation of data: Dess, Suresh, Zelefsky, Freedland, Mahal, Cooperberg, Davis, Horwitz, Terris, Amling, Aronson, Kane, Hearn, Deville, DeWeese, Greco, McNutt, Song, Sun, Mehra, Kaffenberger, Morgan, Nguyen, Feng, Sharma, Tran, Pisansky, Zaorsky, Berlin, Fossati, Briganti, Karnes, Kattan, Schipper, Spratt.

    Drafting of the manuscript: Dess, Suresh, Horwitz, Mehra, Feng, Finelli, Gandaglia, Briganti, Schipper, Spratt.

    Critical revision of the manuscript for important intellectual content: Dess, Suresh, Zelefsky, Freedland, Mahal, Cooperberg, Davis, Horwitz, Terris, Amling, Aronson, Kane, Jackson, Hearn, Deville, DeWeese, Greco, McNutt, Song, Sun, Mehra, Kaffenberger, Morgan, Nguyen, Feng, Sharma, Tran, Stish, Pisansky, Zaorsky, Moraes, Berlin, Finelli, Fossati, Briganti, Carroll, Karnes, Kattan, Schipper, Spratt.

    Statistical analysis: Dess, Suresh, Mahal, Sun, Feng, Moraes, Gandaglia, Briganti, Kattan, Schipper, Spratt.

    Obtained funding: Spratt.

    Administrative, technical, or material support: Jackson, McNutt, Feng, Tran, Zaorsky, Fossati, Carroll, Karnes, Spratt.

    Supervision: Dess, Zelefsky, Terris, Amling, Mehra, Morgan, Sharma, Finelli, Fossati, Gandaglia, Briganti, Spratt.

    Conflict of Interest Disclosures: Dr Mahal reported receiving grants from Prostate Cancer Foundation, American Society for Radiation Oncology, and the Department of Defense and other from Genzyme Corporation, The Exeter Group, and Prostate Health Education Network outside the submitted work. Dr Cooperberg reported receiving personal fees from Astellas Pharma US, Inc, Bayer AG, MDxHealth, Myriad Genetics, Inc, Dendreon Pharmaceutical LLC, Steba biotech, AstraZeneca, and AbbVie, Inc, outside the submitted work. Dr McNutt reported being the founder of Oncospace, Inc, outside the submitted work and having a patent GPU Convolution Superposition with royalties paid to Sun Nuclear, XStrahl, and a patent Shape based autoplanning with royalties paid to Oncospace, Inc. Dr Song reported receiving grants and consulting support from BioProtect Ltd and grants from Bristol Myers Squibb and Bayer AG outside the submitted work. Dr Kaffenberger reported receiving personal fees from MDxHealth and Clovis Oncology and nonfinancial support from Bristol Myers Squibb outside the submitted work. Dr Morgan reported receiving grants from Myriad Genetics, Inc, and Decipher Biosciences, Inc, outside the submitted work. Dr Nguyen reported receiving personal fees from COTA, Ferring Pharmaceuticals, Astellas Pharma US, Inc, Dendreon Pharmaceutical LLC, Blue Earth Diagnostics, and Boston Scientific Corporation, grants and personal fees from Astellas Pharma US, Inc, Bayer AG, and Janssen Pharmaceuticals, Inc, and personal fees and divested equity from Augmenix, Inc, outside the submitted work. Dr Feng reported receiving personal fees from Janssen Oncology, Sanofi, Bayer AG, Celgene Corporation, Blue Earth Diagnostics, Genentech USA Inc, Myovant, Roivant, and Astellas, grants from Zenith Epigenetics Ltd, and other from PFS Genomics and SerImmune outside the submitted work. Dr Tran reported receiving grants from Astellas Pharma US, Inc, and Bayer Healthcare, grants, personal fees, and travel expenses from RefleXion, and personal fees from Noxopharm outside the submitted work and having a patent for Compounds and Methods of Use in Ablative Radiotherapy, Patient 9114158, licensed to Natsar Pharmaceuticals. Dr Finelli reported receiving personal fees from Astellas Pharma US, Inc, Amgen, Inc, Bayer AG, Janssen Pharmaceuticals, Inc, TerSera Therapeutics LLC, Sanofi, AbbVie, Inc, Knight Therapeutics, Inc, and Verity outside the submitted work. Dr Carroll reported serving on the advisory board for Nutcracker Therapeutics (messenger RNA therapy), Francis Medical (tissue ablation), and Insightec (tissue ablation). Dr Karnes reported receiving personal fees and institutional royalties from Decipher Biosciences outside the submitted work. Dr Kattan reported receiving personal fees from Exosome Diagnostics, Inc, and Stratify Genomics outside the submitted work. Dr Schipper reported receiving personal fees from Innovative Analytics outside the submitted work. Dr Spratt reported receiving personal fees from Janssen Pharmaceuticals, Inc, Blue Earth Diagnostics, and AstraZeneca outside the submitted work.

    Funding/Support: This study was supported by the Prostate Cancer Foundation (Dr Spratt); grants CA186786 (Dr Spratt), R01 CA240991-05 (Dr Spratt), RO1 R01CA231219 (Dr Aronson), and R01CA240991-05 (Dr Spratt) from the Prostate Cancer National Institutes of Health Specialized Programs of Research Excellence; and generous philanthropic gifts from patients for partially funding this study.

    Role of the Funder/Sponsor: The sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

    References
    1.
    Klotz  L, Vesprini  D, Sethukavalan  P,  et al.  Long-term follow-up of a large active surveillance cohort of patients with prostate cancer.   J Clin Oncol. 2015;33(3):272-277. doi:10.1200/JCO.2014.55.1192 PubMedGoogle ScholarCrossref
    2.
    Tosoian  JJ, Mamawala  M, Epstein  JI,  et al.  Active surveillance of grade group 1 prostate cancer: long-term outcomes from a large prospective cohort.   Eur Urol. 2020;77(6):675-682. doi:10.1016/j.eururo.2019.12.017 PubMedGoogle ScholarCrossref
    3.
    Hamdy  FC, Donovan  JL, Lane  JA,  et al; ProtecT Study Group.  10-year outcomes after monitoring, surgery, or radiotherapy for localized prostate cancer.   N Engl J Med. 2016;375(15):1415-1424. doi:10.1056/NEJMoa1606220 PubMedGoogle ScholarCrossref
    4.
    Lawton  CA, Lin  X, Hanks  GE,  et al.  Duration of androgen deprivation in locally advanced prostate cancer: long-term update of NRG oncology RTOG 9202.   Int J Radiat Oncol Biol Physics. 2017;98(2):296-303. doi:10.1016/j.ijrobp.2017.02.004Google Scholar
    5.
    Rosenthal  SA, Hu  C, Sartor  O,  et al.  Effect of chemotherapy with docetaxel with androgen suppression and radiotherapy for localized high-risk prostate cancer: the randomized phase III NRG Oncology RTOG 0521 trial.   J Clin Oncol. 2019;37(14):1159-1168. doi:10.1200/JCO.18.02158 PubMedGoogle ScholarCrossref
    6.
    Shipley  WU, Seiferheld  W, Lukka  HR,  et al; NRG Oncology RTOG.  Radiation with or without antiandrogen therapy in recurrent prostate cancer.   N Engl J Med. 2017;376(5):417-428. doi:10.1056/NEJMoa1607529 PubMedGoogle ScholarCrossref
    7.
    Brierley  J, O’Sullivan  B, Asamura  H,  et al.  Global consultation on cancer staging: promoting consistent understanding and use.   Nat Rev Clin Oncol. 2019;16(12):763-771. doi:10.1038/s41571-019-0253-x PubMedGoogle ScholarCrossref
    8.
    Pisansky  TM, Cha  SS, Earle  JD,  et al.  Prostate-specific antigen as a pretherapy prognostic factor in patients treated with radiation therapy for clinically localized prostate cancer.   J Clin Oncol. 1993;11(11):2158-2166. doi:10.1200/JCO.1993.11.11.2158 PubMedGoogle ScholarCrossref
    9.
    Pisansky  TM, Kahn  MJ, Rasp  GM, Cha  SS, Haddock  MG, Bostwick  DG.  A multiple prognostic index predictive of disease outcome after irradiation for clinically localized prostate carcinoma.   Cancer. 1997;79(2):337-344. doi:10.1002/(SICI)1097-0142(19970115)79:2<337::AID-CNCR17>3.0.CO;2-1 PubMedGoogle ScholarCrossref
    10.
    Pisansky  TM, Kahn  MJ, Bostwick  DG.  An enhanced prognostic system for clinically localized carcinoma of the prostate.   Cancer. 1997;79(11):2154-2161. doi:10.1002/(SICI)1097-0142(19970601)79:11<2154::AID-CNCR13>3.0.CO;2-V PubMedGoogle ScholarCrossref
    11.
    Kattan  MW, Hess  KR, Amin  MB,  et al; members of the AJCC Precision Medicine Core.  American Joint Committee on Cancer acceptance criteria for inclusion of risk models for individualized prognosis in the practice of precision medicine.   CA Cancer J Clin. 2016;66(5):370-374. doi:10.3322/caac.21339 PubMedGoogle ScholarCrossref
    12.
    Buyyounouski  MK, Choyke  PL, McKenney  JK,  et al. Prostate cancer—major changes in the American Joint Committee on Cancer eighth edition cancer staging manual.  CA Cancer J Clin. 2017;67(3):245-253. doi:10.3322/caac.21391
    13.
    Collins  GS, Reitsma  JB, Altman  DG, Moons  KGM.  Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement.   BMJ. 2015;350:g7594-g7594. doi:10.1136/bmj.g7594 PubMedGoogle ScholarCrossref
    14.
    Surveillance, Epidemiology and End Results. Prostate with watchful waiting database (2010-2015). National Cancer Institute; 2018. Accessed February 12, 2020. https://seer.cancer.gov/seerstat/databases/prostate-ww/index.html
    15.
    Dess  RT, Hartman  HE, Mahal  BA,  et al.  Association of black race with prostate cancer–specific and other-cause mortality.   JAMA Oncol. 2019;5(7):975-983. doi:10.1001/jamaoncol.2019.0826 PubMedGoogle ScholarCrossref
    16.
    Kupelian  P, Thames  H, Levy  L,  et al.  Year of treatment as independent predictor of relapse-free survival in patients with localized prostate cancer treated with definitive radiotherapy in the PSA era.   Int J Radiat Oncol Biol Phys. 2005;63(3):795-799. doi:10.1016/j.ijrobp.2005.03.029Google ScholarCrossref
    17.
    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(6):1938-1942. doi:10.1097/01.ju.0000158155.33890.e7 PubMedGoogle ScholarCrossref
    18.
    Ishwaran  H, Gerds  TA, Kogalur  UB, Moore  RD, Gange  SJ, Lau  BM.  Random survival forests for competing risks.   Biostatistics. 2014;15(4):757-773. doi:10.1093/biostatistics/kxu010 PubMedGoogle ScholarCrossref
    19.
    National Comprehensive Cancer Network. Prostate Cancer. Version 4.2019 2020. Published August 19, 2019. Accessed February 13, 2020. https://www.nccn.org/professionals/physician_gls/default.aspx#prostate
    20.
    Hermanek  P, Hutter  RV, Sobin  LH.  Prognostic grouping: the next step in tumor classification.   J Cancer Res Clin Oncol. 1990;116(5):513-516. doi:10.1007/BF01613004 PubMedGoogle ScholarCrossref
    21.
    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(11):969-974. doi:10.1001/jama.280.11.969 PubMedGoogle ScholarCrossref
    22.
    Bruinsma  SM, Bangma  CH, Carroll  PR,  et al; Movember GAP3 consortium.  Active surveillance for prostate cancer: a narrative review of clinical guidelines.   Nat Rev Urol. 2016;13(3):151-167. doi:10.1038/nrurol.2015.313 PubMedGoogle ScholarCrossref
    23.
    Tosco  L, Briganti  A, D’amico  AV,  et al.  Systematic review of systemic therapies and therapeutic combinations with local treatments for high-risk localized prostate cancer.   Eur Urol. 2019;75(1):44-60. doi:10.1016/j.eururo.2018.07.027 PubMedGoogle ScholarCrossref
    24.
    Zumsteg  ZS, Zelefsky  MJ, Woo  KM,  et al.  Unification of favourable intermediate-, unfavourable intermediate-, and very high-risk stratification criteria for prostate cancer.   BJU Int. 2017;120(5B):E87-E95. doi:10.1111/bju.13903 PubMedGoogle ScholarCrossref
    25.
    Spratt  DE, Zhang  J, Santiago-Jiménez  M,  et al.  Development and validation of a novel integrated clinical-genomic risk group classification for localized prostate cancer.   J Clin Oncol. 2018;36(6):581-590. doi:10.1200/JCO.2017.74.2940 PubMedGoogle ScholarCrossref
    26.
    Fulgham  PF, Rukstalis  DB, Turkbey  IB,  et al.  AUA policy statement on the use of multiparametric magnetic resonance imaging in the diagnosis, staging and management of prostate cancer.   J Urol. 2017;198(4):832-838. doi:10.1016/j.juro.2017.04.101 PubMedGoogle ScholarCrossref
    ×