- Email: [email protected]
Contemporary identification of patients at high risk of early prostate cancer recurrence after radical retropubic prostatectomy
RAPID COMMUNICATION CME ARTICLE
CONTEMPORARY IDENTIFICATION OF PATIENTS AT HIGH RISK OF EARLY PROSTATE CANCER RECURRENCE AFTER RADICAL RETROPUBIC PROSTATECTOMY WILLIAM W. ROBERTS, ERIK J. BERGSTRALH, MICHAEL L. BLUTE, JEFFREY M. SLEZAK, MICHAEL CARDUCCI, MISOP HAN, JONATHAN I. EPSTEIN, MARIO A. EISENBERGER, PATRICK C. WALSH, AND ALAN W. PARTIN
ABSTRACT Objectives. To develop a model that will identify a contemporary cohort of patients at high risk of early prostate cancer recurrence (greater than 50% at 36 months) after radical retropubic prostatectomy for clinically localized disease. Data from this model will provide important information for patient selection and the design of prospective randomized trials of adjuvant therapies. Methods. Proportional hazards regression analysis was applied to two patient cohorts to develop and cross-validate a multifactorial predictive model to identify men with the highest risk of early prostate cancer recurrence. The model and validation cohorts contained 904 and 901 men, respectively, who underwent radical retropubic prostatectomy at Johns Hopkins Hospital. This model was then externally validated using a cohort of patients from the Mayo Clinic. Results. A model for weighted risk of recurrence was developed: RW⬘ ⫽ lymph node involvement (0/1) ⫻ 1.43 ⫹ surgical margin status (0/1) ⫻ 1.15 ⫹ modified Gleason score (0 to 4) ⫻ 0.71 ⫹ seminal vesicle involvement (0/1) ⫻ 0.51. Men with an RW⬘ greater than 2.84 (9%) demonstrated a 50% biochemical recurrence rate (prostrate-specific antigen level greater than 0.2 ng/mL) at 3 years and thus were placed in the high-risk group. Kaplan-Meier analyses of biochemical recurrence-free survival demonstrated rapid deviation of the curves based on the RW⬘. This model was cross-validated in the second group of patients and performed with similar results. Furthermore, similar trends were apparent when the model was externally validated on patients treated at the Mayo Clinic. Conclusions. We have developed a multivariate Cox proportional hazards model that successfully stratifies patients on the basis of their risk of early prostate cancer recurrence. UROLOGY 57: 1033–1037, 2001. © 2001, Elsevier Science Inc.
A
fter radical retropubic prostatectomy (RRP) for clinically localized prostate cancer, 16% of patients by 5 years have developed biochemical disease recurrence demonstrated by a detectable prostatespecific antigen (PSA) level.1 Current adjuvant therapies for prostate cancer have not demonstrated longterm survival benefit (greater than 10 years), unlike This study was supported by SPORE grant CA-58326 From the James Buchanan Brady Urological Institute and Departments of Urology, Oncology, and Pathology, Johns Hopkins Medical Institutions, Baltimore, Maryland; and Section of Biostatistics and Department of Urology, Mayo Clinic, Rochester, Minnesota Reprint requests: Alan W. Partin, M.D., Ph.D., Department of Urology, Marburg 205A, Johns Hopkins Hospital, 600 North Wolfe Street, Baltimore, MD 21287-2101 Submitted: January 11, 2001, accepted (with revisions): February 5, 2001 © 2001, ELSEVIER SCIENCE INC. ALL RIGHTS RESERVED
adjuvant therapies for breast or colorectal cancer, which have become the standard of care.2– 4 Randomized controlled trials of new adjuvant therapies for prostate cancer can most effectively be designed and evaluated using a rigorous biostatistical model to identify the subset of patients at highest risk of early disease recurrence. We used easily obtainable pathologic and clinical variables from a contemporary population of patients to construct a multivariate proportional hazards model of early prostate cancer recurrence. MATERIAL AND METHODS Between April 1982 and June 1999, 2569 men underwent anatomic RRP for clinically localized prostate cancer at Johns Hopkins Hospital by a single surgeon (P.C.W.). Patients with clinical Stage T1a and T1b disease (n ⫽ 178) were excluded, as 0090-4295/01/$20.00 PII S0090-4295(01)00978-5 1033
TABLE I. Demographic, clinical, and pathologic data for modeling and validation groups
Mean age (yr) White (n) Clinical stage (n) T1c T2a T2b T2c T3a Preoperative PSA (ng/mL) ⱕ4.0 4.1–10.0 10.1–20.0 ⬎20.0 Modified Gleason score (n) 0 Gleason sum 5 1 Gleason pattern (3⫹3) 2 Gleason pattern (3⫹4) 3 Gleason pattern (4⫹3) 4 Gleason sum 8–10 Pathologic stage (n) Organ confined Extraprostatic extension Positive surgical margins Positive seminal vesicles Positive lymph nodes
Modeling Group
Validation Group
P Value
Mayo Validation
58.2 (33—76) 871 (96)
57.8 (36–75) 849 (94)
0.19* 0.03†
65.3 (40–84) (90⫹)
379 306 150 43 26
(42) (34) (17) (5) (3)
356 327 150 45 23
(40) (36) (17) (5) (3)
203 498 153 50
(22) (55) (17) (6)
213 475 168 45
(24) (53) (19) (5)
82 445 234 75 68
(9) (49) (26) (8) (8)
81 444 234 75 67
(9) (49) (26) (8) (7)
0.99*
443 452 96 79 54
(49) (50) (11) (9) (6)
431 468 91 71 58
(48) (52) (10) (8) (6)
0.62† 0.41† 0.72† 0.51† 0.68†
0.68*
0.88*
501 547 781 373 197
(21) (23) (33) (16) (8)
379 1212 552 256
(16) (51) (21) (11)
1159 (48) 508 (21) 635 (26)‡ 97 (4) 1545 854 747 275 29
(64) (36) (31) (11) (1)
KEY: PSA ⫽ prostate-specific antigen. Numbers in parentheses are percentages, except for age for which they are the range. * t test. † Chi-square test. ‡ Sum of Gleason pattern (3 ⫹ 4) and Gleason pattern (4 ⫹ 3).
transurethral resection before RRP could have altered their PSA levels. Patients with a Gleason sum of less than 5 (n ⫽ 23) were excluded to reflect contemporary trends.5 Patients with less than 1 year of follow-up (n ⫽ 207) were excluded. Patients whose records did not document a preoperative serum PSA level (n ⫽ 255) or were incomplete regarding surgical margin status, extraprostatic extension status, seminal vesicle status, lymph node status, and prostatectomy Gleason score (n ⫽ 53) were also excluded. Patients who had undergone preoperative neoadjuvant hormonal or radiation therapy or postoperative adjuvant hormonal or radiation therapy before evidence of PSA elevation (n ⫽ 37) were not included. Additionally, 6 patients were excluded because of no evidence of cancer within the specimen and 5 patients because of known metastatic disease (Stage D1-D2) before surgery. A computerized sorting algorithm from STATA 6.0 (Stata, College Station, Tex) randomly divided the eligible men into a modeling cohort (904 men) and a validation cohort (901 men), matched for Gleason score. All men were followed up after surgery with a digital rectal examination and serum PSA determination at 3, 6, and 12 months and then annually. Biochemical recurrence was defined as a detectable PSA level (greater than 0.2 ng/mL). The prostate, seminal vesicles, and lymph nodes were examined histologically in standard fashion to determine extraprostatic extension status, surgical margin status, seminal vesicle involvement, and lymph node status.6,7 A modified Gleason score based on the primary and secondary Gleason pattern was then assigned to further distribute the midrange Gleason scores (Table I). 1034
All statistical analyses were performed using STATA 6.0. The Kaplan-Meier method was used to estimate biochemical recurrence-free survival after RRP for each risk group. Patients last known alive without biochemical recurrence and those dead without recurrence had their follow-up time censored. The Cox proportional hazards regression analysis was used to model the time to biochemical recurrence after surgery using pathologic and clinical variables. All clinical and pathologic information was included in the original multivariate regression model. The preoperative PSA level, clinical stage, and modified Gleason score were entered as continuous variables. The pathologic variables (eg, extraprostatic extension status, seminal vesicle status, lymph node status, and surgical margin status) were entered as dichotomous variables. Variables were removed in a stepwise fashion when they failed to explain the differences in the risk of recurrence. The remaining variables were then incorporated into the final risk equation. The predictive ability of this equation was then assessed using the validation cohort. A second multivariate equation was then derived using the same process on the validation cohort, then cross-validated on the original modeling group. This cross-validation process yielded results similar to the first equation (results not shown). External validation of the RW⬘ equation was performed on 2399 patients treated at the Mayo Clinic who met the same entry criteria. Patients in this series with a Gleason sum of 7 were assigned a modified Gleason score of 2.5 since the primary and secondary Gleason patterns were not available in the database. UROLOGY 57 (6), 2001
FIGURE 1. Kaplan-Meier biochemical recurrence-free survival (proportion) in the modeling and validation groups.
TABLE II. Univariate and multivariate Cox proportional hazards analyses of variables in modeling group for prediction of biochemical recurrence Hazards Ratio
95% CI
P Value
Univariate analysis Positive lymph nodes Positive extraprostatic extension Positive seminal vesicles Positive surgical margins Modified Gleason score Clinical stage PSA Age
9.56 7.58 7.48 4.25 2.50 1.47 1.06 1.03
6.58–13.90 4.27–13.44 5.23–10.68 2.96–6.09 2.17–2.88 1.27–1.70 1.04–1.07 1.01–1.07
0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.004
Multivariate analysis Positive lymph nodes Positive surgical margins Modified Gleason score Positive seminal vesicles
4.20 3.17 2.02 1.66
2.68–6.58 2.19–4.60 1.72–2.36 1.07–2.59
0.001 0.001 0.001 0.02
KEY: CI ⫽ confidence interval; PSA ⫽ prostate-specific antigen.
RESULTS
RW⬘⫽lymph node involvement (0/1) ⫻ 1.43
The demographic, clinical, and pathologic data for the modeling and validation groups are summarized in Table I. These groups were well matched for age, clinical stage, PSA level, and pathologic stage. The final pathologic analysis demonstrated that 49% and 48% of the men in the modeling and validation groups, respectively, had organ-confined disease. Overall, 271 men (15%) developed biochemical recurrence. The Kaplan-Meier survival curves for the modeling and validation groups were similar (Fig. 1). The results of the univariate and multivariate analyses are shown in Table II and yielded the following equation based on the multivariate Cox model coefficients:
⫹surgical margin status (0/1) ⫻ 1.15
UROLOGY 57 (6), 2001
⫹modified Gleason score (0 to 4) ⫻ 0.71 ⫹seminal vesicle involvement (0/1) ⫻ 0.51. Men at highest risk of early recurrence were defined as having a RW⬘ greater than 2.84. This cutoff value was selected because it resulted in a greater than 50% rate of biochemical recurrence by 3 years. This same equation was then applied to the validation cohort with the same RW⬘ cutoff. KaplanMeier recurrence-free survival rates for the modeling and validation groups are plotted in Figure 2. The modeling group biochemical recurrence1035
FIGURE 2. Kaplan-Meier biochemical recurrence-free survival (proportion) in high-risk and low-risk subgroups of the modeling and validation groups. Numbers under curves represent the number of men at the start of each interval.
free rate was 95%, 91%, and 80% in the low-risk group at 3, 5, and 10 years, respectively, and 42%, 35%, and 18% in the high-risk group at 3, 5, and 10 years, respectively. Application of this model to an independent series of patients from the Mayo Clinic (Table I) produced similar results with a biochemical recurrence-free rate of 79%, 71%, and 62% at 3, 5, and 9 years, respectively, for the lowrisk group and 50%, 39%, and 27% at 3, 5, and 9 years, respectively, for the high-risk group. The similar separation of the high and low-risk survival curves seen in these two patient populations provides validation of this model. COMMENT Previously, Partin et al.8 developed a biostatistical model (RW-weighted risk of recurrence) using the postoperative Gleason score, the specimen confinement status, and a sigmoidally transformed PSA value to predict which patients with palpable T2b or T2c prostate cancer were most likely to develop early biochemical recurrence and therefore deemed most suitable to participate in adju1036
vant clinical trials of new therapies. This validated model successfully stratified patients into high, intermediate, and low-risk groups for early disease recurrence. However, use of PSA and improved early detection methods have brought about a stage shift in the presentation of prostate cancer; a higher proportion of men now present with nonpalpable Stage T1c disease. Therefore, an updated predictive model of prostate cancer recurrence is warranted to accurately reflect this shift in patient presentation and stage migration. Other attempts to identify patients at high risk of early biochemical failure after RRP for clinically localized disease have incorporated race, quantitative nuclear grade, microvessel density, p53, retinoblastoma expression, and chromogranin A.9 –11 A neural network method has also been applied to this issue, and successful results were noted, albeit on a small, highly selected, patient population.12 Stamey et al.13 developed a multivariate regression model that incorporated PSA level, Gleason grade, cancer volume, prostate weight, vascular invasion, percentage of intraductal cancer, and lymph node UROLOGY 57 (6), 2001
status. Although the data from their analysis appear promising, significant time and expense are necessary in performing the tumor volume analysis of the prostate, rendering the technique impractical for widespread use and independent validation. Kattan et al.14 developed a predictive nomogram based on PSA level, Gleason sum, extraprostatic extension status, surgical margin status, seminal vesicle invasion, and lymph node status. This nomogram was designed to characterize the risk of biochemical recurrence in an individual patient. Unfortunately, no stratification into groups based on risk was reported. We developed a multivariate proportional hazards model based on easily obtainable clinical and pathologic information: lymph node status, seminal vesicle status, surgical margin status, and Gleason score. Extraprostatic extension did demonstrate a statistically significant hazard ratio when incorporated into the multivariate analysis. However, the addition of this fifth variable into the equation did not alter the stratification achieved using only four variables. For this reason and to avoid variability in the classification of extraprostatic extension, this variable was not incorporated into the final equation. Blute et al.15 have found the preoperative PSA level to be a significant predictor of biochemical recurrence. Interestingly, our study did not find the PSA level to add significantly to the final overall model for either the modeling (P ⫽ 0.14) or validation (P ⫽ 0.08) groups. However, among patients classified as low risk, the PSA level did provide some additional predictive information (10year event-free rate of 90%, 84%, and 67% for PSA less than 4, 4 to 10, and greater than 10 ng/mL, respectively; P ⬍0.01). The possible explanations for the decreased predictive importance of the PSA level include that there were fewer high (greater than 20 ng/mL) PSA values (5% versus 14%) compared with the study by Blute et al.,15 perhaps making it more difficult to detect a relationship between PSA and outcome. CONCLUSIONS Within 5 years of RRP for clinically localized disease, 16% of men will develop biochemical recurrence of prostate cancer. We have developed and validated a multivariate proportional hazards model that successfully stratifies patients on the basis of their risk of early disease recurrence. Using this model, the selection of men at high risk of early prostate cancer recurrence can be accomplished within 3 to 4 days of surgery, thus allowing rapid enrollment in randomized clinical trials evaluating novel adjuvant therapies. These men at high
UROLOGY 57 (6), 2001
risk of early recurrence are not only ideal candidates for clinical trials, as their enrollment will maximize the efficiency of the trials (by decreasing the sample size and the time to definitive results), but they are also the men most likely to benefit from these experimental therapies. REFERENCES 1. Han M, Partin AW, Pound CR, et al: Long term biochemical disease-free and cancer-specific survival following anatomic radical retropubic prostatectomy: the 15-year Johns Hopkins experience. Urol Clin North Am (in press). 2. McCarthy JF, Catalona WJ, and Hudson MA: Effect of radiation therapy on detectable serum prostate specific antigen levels following radical prostatectomy: early versus delayed treatment. J Urol 151: 1575–1578, 1994. 3. Fair WR, Cookson MS, Stroumbakis N, et al: The indications, rationale, and results of neoadjuvant androgen deprivation in the treatment of prostate cancer: Memorial SloanKettering Cancer Center results. Urology 49(suppl 3A): 46 – 55, 1997. 4. Hegarty NJ, Fitzpatrick JM, Richie JP, et al: Future prospects in prostate cancer. Prostate 40: 261–268, 1999. 5. Epstein JI: Gleason score 2-4 adenocarcinoma of the prostate on needle biopsy: a diagnosis that should not be made. Am J Surg Pathol 24: 477– 478, 2000. 6. Bova GS, Fox WM, and Epstein JI: Methods of radical prostatectomy specimen processing: a novel technique for harvesting fresh prostate cancer tissue and review of processing techniques. Mod Pathol 6: 201–207, 1993. 7. Hall GS, Kramer CE, and Epstein JI: Evaluation of radical prostatectomy specimens: a comparative analysis of various sampling methods. Am J Surg Pathol 16: 315–324, 1992. 8. Partin AW, Piantadosi S, Sanda MG, et al: Selection of men at high risk for disease recurrence for experimental adjuvant therapy following radical prostatectomy. Urology 45: 831– 838, 1995. 9. Bauer JJ, Connelly RR, Seterhenn IA, et al: Biostatistical modeling using traditional preoperative and pathological prognostic variables in the selection of men at high risk for disease recurrence after radical prostatectomy for prostate cancer. J Urol 159: 929 –933, 1998. 10. Veltri RW, Miller MC, Partin AW, et al: Ability to predict biochemical progression using Gleason score and a computer-generated quantitative nuclear grade derived from cancer cell nuclei. Urology 48: 685– 691, 1996. 11. Krupski T, Petroni GR, Frierson HF, et al: Microvessel density, p53, retinoblastoma, and chromogranin A immunohistochemistries as predictors of disease-specific survival following radical prostatectomy for carcinoma of the prostate. Urology 55: 743–749, 2000. 12. Potter SR, Miller C, Mangold LA, et al: Genetically engineered neural networks for predicting prostate cancer progression after radical prostatectomy. Urology 54: 791–795, 1999. 13. Stamey TA, Yemoto CY, McNeal JE, et al: Prostate cancer is highly predictable: a prognostic equation based on all morphological variables in radical prostatectomy specimens. J Urol 163: 1155–1160, 2000. 14. Kattan MW, Wheeler TM, and Scardino PT: Postoperative nomogram for disease recurrence after radical prostatectomy for prostate cancer. J Clin Oncol 17: 1499 –1507, 1999. 15. Blute ML, Bergstralh EJ, Iocca A, et al: Use of Gleason score, prostate specific antigen, seminal vesicle and margin status to predict biochemical failure after radical prostatectomy. J Urol 165: 119 –125, 2001.
1037
CONTEMPORARY IDENTIFICATION OF PATIENTS AT HIGH RISK OF EARLY PROSTATE CANCER RECURRENCE AFTER RADICAL RETROPUBIC PROSTATECTOMY WILLIAM W. ROBERTS, ERIK J. BERGSTRALH, MICHAEL L. BLUTE, JEFFREY M. SLEZAK, MICHAEL CARDUCCI, MISOP HAN, JONATHAN I. EPSTEIN, MARIO A. EISENBERGER, PATRICK C. WALSH, AND ALAN W. PARTIN
ABSTRACT Objectives. To develop a model that will identify a contemporary cohort of patients at high risk of early prostate cancer recurrence (greater than 50% at 36 months) after radical retropubic prostatectomy for clinically localized disease. Data from this model will provide important information for patient selection and the design of prospective randomized trials of adjuvant therapies. Methods. Proportional hazards regression analysis was applied to two patient cohorts to develop and cross-validate a multifactorial predictive model to identify men with the highest risk of early prostate cancer recurrence. The model and validation cohorts contained 904 and 901 men, respectively, who underwent radical retropubic prostatectomy at Johns Hopkins Hospital. This model was then externally validated using a cohort of patients from the Mayo Clinic. Results. A model for weighted risk of recurrence was developed: RW⬘ ⫽ lymph node involvement (0/1) ⫻ 1.43 ⫹ surgical margin status (0/1) ⫻ 1.15 ⫹ modified Gleason score (0 to 4) ⫻ 0.71 ⫹ seminal vesicle involvement (0/1) ⫻ 0.51. Men with an RW⬘ greater than 2.84 (9%) demonstrated a 50% biochemical recurrence rate (prostrate-specific antigen level greater than 0.2 ng/mL) at 3 years and thus were placed in the high-risk group. Kaplan-Meier analyses of biochemical recurrence-free survival demonstrated rapid deviation of the curves based on the RW⬘. This model was cross-validated in the second group of patients and performed with similar results. Furthermore, similar trends were apparent when the model was externally validated on patients treated at the Mayo Clinic. Conclusions. We have developed a multivariate Cox proportional hazards model that successfully stratifies patients on the basis of their risk of early prostate cancer recurrence. UROLOGY 57: 1033–1037, 2001. © 2001, Elsevier Science Inc.
A
fter radical retropubic prostatectomy (RRP) for clinically localized prostate cancer, 16% of patients by 5 years have developed biochemical disease recurrence demonstrated by a detectable prostatespecific antigen (PSA) level.1 Current adjuvant therapies for prostate cancer have not demonstrated longterm survival benefit (greater than 10 years), unlike This study was supported by SPORE grant CA-58326 From the James Buchanan Brady Urological Institute and Departments of Urology, Oncology, and Pathology, Johns Hopkins Medical Institutions, Baltimore, Maryland; and Section of Biostatistics and Department of Urology, Mayo Clinic, Rochester, Minnesota Reprint requests: Alan W. Partin, M.D., Ph.D., Department of Urology, Marburg 205A, Johns Hopkins Hospital, 600 North Wolfe Street, Baltimore, MD 21287-2101 Submitted: January 11, 2001, accepted (with revisions): February 5, 2001 © 2001, ELSEVIER SCIENCE INC. ALL RIGHTS RESERVED
adjuvant therapies for breast or colorectal cancer, which have become the standard of care.2– 4 Randomized controlled trials of new adjuvant therapies for prostate cancer can most effectively be designed and evaluated using a rigorous biostatistical model to identify the subset of patients at highest risk of early disease recurrence. We used easily obtainable pathologic and clinical variables from a contemporary population of patients to construct a multivariate proportional hazards model of early prostate cancer recurrence. MATERIAL AND METHODS Between April 1982 and June 1999, 2569 men underwent anatomic RRP for clinically localized prostate cancer at Johns Hopkins Hospital by a single surgeon (P.C.W.). Patients with clinical Stage T1a and T1b disease (n ⫽ 178) were excluded, as 0090-4295/01/$20.00 PII S0090-4295(01)00978-5 1033
TABLE I. Demographic, clinical, and pathologic data for modeling and validation groups
Mean age (yr) White (n) Clinical stage (n) T1c T2a T2b T2c T3a Preoperative PSA (ng/mL) ⱕ4.0 4.1–10.0 10.1–20.0 ⬎20.0 Modified Gleason score (n) 0 Gleason sum 5 1 Gleason pattern (3⫹3) 2 Gleason pattern (3⫹4) 3 Gleason pattern (4⫹3) 4 Gleason sum 8–10 Pathologic stage (n) Organ confined Extraprostatic extension Positive surgical margins Positive seminal vesicles Positive lymph nodes
Modeling Group
Validation Group
P Value
Mayo Validation
58.2 (33—76) 871 (96)
57.8 (36–75) 849 (94)
0.19* 0.03†
65.3 (40–84) (90⫹)
379 306 150 43 26
(42) (34) (17) (5) (3)
356 327 150 45 23
(40) (36) (17) (5) (3)
203 498 153 50
(22) (55) (17) (6)
213 475 168 45
(24) (53) (19) (5)
82 445 234 75 68
(9) (49) (26) (8) (8)
81 444 234 75 67
(9) (49) (26) (8) (7)
0.99*
443 452 96 79 54
(49) (50) (11) (9) (6)
431 468 91 71 58
(48) (52) (10) (8) (6)
0.62† 0.41† 0.72† 0.51† 0.68†
0.68*
0.88*
501 547 781 373 197
(21) (23) (33) (16) (8)
379 1212 552 256
(16) (51) (21) (11)
1159 (48) 508 (21) 635 (26)‡ 97 (4) 1545 854 747 275 29
(64) (36) (31) (11) (1)
KEY: PSA ⫽ prostate-specific antigen. Numbers in parentheses are percentages, except for age for which they are the range. * t test. † Chi-square test. ‡ Sum of Gleason pattern (3 ⫹ 4) and Gleason pattern (4 ⫹ 3).
transurethral resection before RRP could have altered their PSA levels. Patients with a Gleason sum of less than 5 (n ⫽ 23) were excluded to reflect contemporary trends.5 Patients with less than 1 year of follow-up (n ⫽ 207) were excluded. Patients whose records did not document a preoperative serum PSA level (n ⫽ 255) or were incomplete regarding surgical margin status, extraprostatic extension status, seminal vesicle status, lymph node status, and prostatectomy Gleason score (n ⫽ 53) were also excluded. Patients who had undergone preoperative neoadjuvant hormonal or radiation therapy or postoperative adjuvant hormonal or radiation therapy before evidence of PSA elevation (n ⫽ 37) were not included. Additionally, 6 patients were excluded because of no evidence of cancer within the specimen and 5 patients because of known metastatic disease (Stage D1-D2) before surgery. A computerized sorting algorithm from STATA 6.0 (Stata, College Station, Tex) randomly divided the eligible men into a modeling cohort (904 men) and a validation cohort (901 men), matched for Gleason score. All men were followed up after surgery with a digital rectal examination and serum PSA determination at 3, 6, and 12 months and then annually. Biochemical recurrence was defined as a detectable PSA level (greater than 0.2 ng/mL). The prostate, seminal vesicles, and lymph nodes were examined histologically in standard fashion to determine extraprostatic extension status, surgical margin status, seminal vesicle involvement, and lymph node status.6,7 A modified Gleason score based on the primary and secondary Gleason pattern was then assigned to further distribute the midrange Gleason scores (Table I). 1034
All statistical analyses were performed using STATA 6.0. The Kaplan-Meier method was used to estimate biochemical recurrence-free survival after RRP for each risk group. Patients last known alive without biochemical recurrence and those dead without recurrence had their follow-up time censored. The Cox proportional hazards regression analysis was used to model the time to biochemical recurrence after surgery using pathologic and clinical variables. All clinical and pathologic information was included in the original multivariate regression model. The preoperative PSA level, clinical stage, and modified Gleason score were entered as continuous variables. The pathologic variables (eg, extraprostatic extension status, seminal vesicle status, lymph node status, and surgical margin status) were entered as dichotomous variables. Variables were removed in a stepwise fashion when they failed to explain the differences in the risk of recurrence. The remaining variables were then incorporated into the final risk equation. The predictive ability of this equation was then assessed using the validation cohort. A second multivariate equation was then derived using the same process on the validation cohort, then cross-validated on the original modeling group. This cross-validation process yielded results similar to the first equation (results not shown). External validation of the RW⬘ equation was performed on 2399 patients treated at the Mayo Clinic who met the same entry criteria. Patients in this series with a Gleason sum of 7 were assigned a modified Gleason score of 2.5 since the primary and secondary Gleason patterns were not available in the database. UROLOGY 57 (6), 2001
FIGURE 1. Kaplan-Meier biochemical recurrence-free survival (proportion) in the modeling and validation groups.
TABLE II. Univariate and multivariate Cox proportional hazards analyses of variables in modeling group for prediction of biochemical recurrence Hazards Ratio
95% CI
P Value
Univariate analysis Positive lymph nodes Positive extraprostatic extension Positive seminal vesicles Positive surgical margins Modified Gleason score Clinical stage PSA Age
9.56 7.58 7.48 4.25 2.50 1.47 1.06 1.03
6.58–13.90 4.27–13.44 5.23–10.68 2.96–6.09 2.17–2.88 1.27–1.70 1.04–1.07 1.01–1.07
0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.004
Multivariate analysis Positive lymph nodes Positive surgical margins Modified Gleason score Positive seminal vesicles
4.20 3.17 2.02 1.66
2.68–6.58 2.19–4.60 1.72–2.36 1.07–2.59
0.001 0.001 0.001 0.02
KEY: CI ⫽ confidence interval; PSA ⫽ prostate-specific antigen.
RESULTS
RW⬘⫽lymph node involvement (0/1) ⫻ 1.43
The demographic, clinical, and pathologic data for the modeling and validation groups are summarized in Table I. These groups were well matched for age, clinical stage, PSA level, and pathologic stage. The final pathologic analysis demonstrated that 49% and 48% of the men in the modeling and validation groups, respectively, had organ-confined disease. Overall, 271 men (15%) developed biochemical recurrence. The Kaplan-Meier survival curves for the modeling and validation groups were similar (Fig. 1). The results of the univariate and multivariate analyses are shown in Table II and yielded the following equation based on the multivariate Cox model coefficients:
⫹surgical margin status (0/1) ⫻ 1.15
UROLOGY 57 (6), 2001
⫹modified Gleason score (0 to 4) ⫻ 0.71 ⫹seminal vesicle involvement (0/1) ⫻ 0.51. Men at highest risk of early recurrence were defined as having a RW⬘ greater than 2.84. This cutoff value was selected because it resulted in a greater than 50% rate of biochemical recurrence by 3 years. This same equation was then applied to the validation cohort with the same RW⬘ cutoff. KaplanMeier recurrence-free survival rates for the modeling and validation groups are plotted in Figure 2. The modeling group biochemical recurrence1035
FIGURE 2. Kaplan-Meier biochemical recurrence-free survival (proportion) in high-risk and low-risk subgroups of the modeling and validation groups. Numbers under curves represent the number of men at the start of each interval.
free rate was 95%, 91%, and 80% in the low-risk group at 3, 5, and 10 years, respectively, and 42%, 35%, and 18% in the high-risk group at 3, 5, and 10 years, respectively. Application of this model to an independent series of patients from the Mayo Clinic (Table I) produced similar results with a biochemical recurrence-free rate of 79%, 71%, and 62% at 3, 5, and 9 years, respectively, for the lowrisk group and 50%, 39%, and 27% at 3, 5, and 9 years, respectively, for the high-risk group. The similar separation of the high and low-risk survival curves seen in these two patient populations provides validation of this model. COMMENT Previously, Partin et al.8 developed a biostatistical model (RW-weighted risk of recurrence) using the postoperative Gleason score, the specimen confinement status, and a sigmoidally transformed PSA value to predict which patients with palpable T2b or T2c prostate cancer were most likely to develop early biochemical recurrence and therefore deemed most suitable to participate in adju1036
vant clinical trials of new therapies. This validated model successfully stratified patients into high, intermediate, and low-risk groups for early disease recurrence. However, use of PSA and improved early detection methods have brought about a stage shift in the presentation of prostate cancer; a higher proportion of men now present with nonpalpable Stage T1c disease. Therefore, an updated predictive model of prostate cancer recurrence is warranted to accurately reflect this shift in patient presentation and stage migration. Other attempts to identify patients at high risk of early biochemical failure after RRP for clinically localized disease have incorporated race, quantitative nuclear grade, microvessel density, p53, retinoblastoma expression, and chromogranin A.9 –11 A neural network method has also been applied to this issue, and successful results were noted, albeit on a small, highly selected, patient population.12 Stamey et al.13 developed a multivariate regression model that incorporated PSA level, Gleason grade, cancer volume, prostate weight, vascular invasion, percentage of intraductal cancer, and lymph node UROLOGY 57 (6), 2001
status. Although the data from their analysis appear promising, significant time and expense are necessary in performing the tumor volume analysis of the prostate, rendering the technique impractical for widespread use and independent validation. Kattan et al.14 developed a predictive nomogram based on PSA level, Gleason sum, extraprostatic extension status, surgical margin status, seminal vesicle invasion, and lymph node status. This nomogram was designed to characterize the risk of biochemical recurrence in an individual patient. Unfortunately, no stratification into groups based on risk was reported. We developed a multivariate proportional hazards model based on easily obtainable clinical and pathologic information: lymph node status, seminal vesicle status, surgical margin status, and Gleason score. Extraprostatic extension did demonstrate a statistically significant hazard ratio when incorporated into the multivariate analysis. However, the addition of this fifth variable into the equation did not alter the stratification achieved using only four variables. For this reason and to avoid variability in the classification of extraprostatic extension, this variable was not incorporated into the final equation. Blute et al.15 have found the preoperative PSA level to be a significant predictor of biochemical recurrence. Interestingly, our study did not find the PSA level to add significantly to the final overall model for either the modeling (P ⫽ 0.14) or validation (P ⫽ 0.08) groups. However, among patients classified as low risk, the PSA level did provide some additional predictive information (10year event-free rate of 90%, 84%, and 67% for PSA less than 4, 4 to 10, and greater than 10 ng/mL, respectively; P ⬍0.01). The possible explanations for the decreased predictive importance of the PSA level include that there were fewer high (greater than 20 ng/mL) PSA values (5% versus 14%) compared with the study by Blute et al.,15 perhaps making it more difficult to detect a relationship between PSA and outcome. CONCLUSIONS Within 5 years of RRP for clinically localized disease, 16% of men will develop biochemical recurrence of prostate cancer. We have developed and validated a multivariate proportional hazards model that successfully stratifies patients on the basis of their risk of early disease recurrence. Using this model, the selection of men at high risk of early prostate cancer recurrence can be accomplished within 3 to 4 days of surgery, thus allowing rapid enrollment in randomized clinical trials evaluating novel adjuvant therapies. These men at high
UROLOGY 57 (6), 2001
risk of early recurrence are not only ideal candidates for clinical trials, as their enrollment will maximize the efficiency of the trials (by decreasing the sample size and the time to definitive results), but they are also the men most likely to benefit from these experimental therapies. REFERENCES 1. Han M, Partin AW, Pound CR, et al: Long term biochemical disease-free and cancer-specific survival following anatomic radical retropubic prostatectomy: the 15-year Johns Hopkins experience. Urol Clin North Am (in press). 2. McCarthy JF, Catalona WJ, and Hudson MA: Effect of radiation therapy on detectable serum prostate specific antigen levels following radical prostatectomy: early versus delayed treatment. J Urol 151: 1575–1578, 1994. 3. Fair WR, Cookson MS, Stroumbakis N, et al: The indications, rationale, and results of neoadjuvant androgen deprivation in the treatment of prostate cancer: Memorial SloanKettering Cancer Center results. Urology 49(suppl 3A): 46 – 55, 1997. 4. Hegarty NJ, Fitzpatrick JM, Richie JP, et al: Future prospects in prostate cancer. Prostate 40: 261–268, 1999. 5. Epstein JI: Gleason score 2-4 adenocarcinoma of the prostate on needle biopsy: a diagnosis that should not be made. Am J Surg Pathol 24: 477– 478, 2000. 6. Bova GS, Fox WM, and Epstein JI: Methods of radical prostatectomy specimen processing: a novel technique for harvesting fresh prostate cancer tissue and review of processing techniques. Mod Pathol 6: 201–207, 1993. 7. Hall GS, Kramer CE, and Epstein JI: Evaluation of radical prostatectomy specimens: a comparative analysis of various sampling methods. Am J Surg Pathol 16: 315–324, 1992. 8. Partin AW, Piantadosi S, Sanda MG, et al: Selection of men at high risk for disease recurrence for experimental adjuvant therapy following radical prostatectomy. Urology 45: 831– 838, 1995. 9. Bauer JJ, Connelly RR, Seterhenn IA, et al: Biostatistical modeling using traditional preoperative and pathological prognostic variables in the selection of men at high risk for disease recurrence after radical prostatectomy for prostate cancer. J Urol 159: 929 –933, 1998. 10. Veltri RW, Miller MC, Partin AW, et al: Ability to predict biochemical progression using Gleason score and a computer-generated quantitative nuclear grade derived from cancer cell nuclei. Urology 48: 685– 691, 1996. 11. Krupski T, Petroni GR, Frierson HF, et al: Microvessel density, p53, retinoblastoma, and chromogranin A immunohistochemistries as predictors of disease-specific survival following radical prostatectomy for carcinoma of the prostate. Urology 55: 743–749, 2000. 12. Potter SR, Miller C, Mangold LA, et al: Genetically engineered neural networks for predicting prostate cancer progression after radical prostatectomy. Urology 54: 791–795, 1999. 13. Stamey TA, Yemoto CY, McNeal JE, et al: Prostate cancer is highly predictable: a prognostic equation based on all morphological variables in radical prostatectomy specimens. J Urol 163: 1155–1160, 2000. 14. Kattan MW, Wheeler TM, and Scardino PT: Postoperative nomogram for disease recurrence after radical prostatectomy for prostate cancer. J Clin Oncol 17: 1499 –1507, 1999. 15. Blute ML, Bergstralh EJ, Iocca A, et al: Use of Gleason score, prostate specific antigen, seminal vesicle and margin status to predict biochemical failure after radical prostatectomy. J Urol 165: 119 –125, 2001.
1037