EUROPEAN UROLOGY 60 (2011)

Size: px

Start display at page:

Download "EUROPEAN UROLOGY 60 (2011)"

James Thompson
5 years ago
Views:

EUROPEAN UROLOGY 60 (2011) 920 930 available at www.sciencedirect.com journal homepage: www.europeanurology.com Platinum Priority Prostate Cancer Editorial by Monique J. Roobol and Eveline A.M. Heijnsdijk on pp.

Population-Based Cohort Firas Abdollah a,b,1, Maxine Sun a,1, *, Jan Schmitges a,c, Zhe Tian a, Claudio Jeldres a, Alberto Briganti b, Shahrohk F.

Karakiewicz a,e a Cancer Prognostics and Health Outcomes Unit, University of Montreal Health Centre, Montreal, Canada; b Department of Urology, Vita Salute San Raffaele University, Milan, Italy; c

1 EUROPEAN UROLOGY 60 (2011) available at journal homepage: Platinum Priority Prostate Cancer Editorial by Monique J. Roobol and Eveline A.M. Heijnsdijk on pp of this issue Cancer-Specific and Other-Cause Mortality After Radical Prostatectomy Versus Observation in Patients with Prostate Cancer: Competing-Risks Analysis of a Large North American Population-Based Cohort Firas Abdollah a,b,1, Maxine Sun a,1, *, Jan Schmitges a,c, Zhe Tian a, Claudio Jeldres a, Alberto Briganti b, Shahrohk F. Shariat d, Paul Perrotte e, Francesco Montorsi b, Pierre I. Karakiewicz a,e a Cancer Prognostics and Health Outcomes Unit, University of Montreal Health Centre, Montreal, Canada; b Department of Urology, Vita Salute San Raffaele University, Milan, Italy; c Martini-Clinic, Prostate Cancer Center Hamburg-Eppendorf, Hamburg, Germany; d Department of Urology, Weill Medical College of Cornell University, New York, NY, USA; e Department of Urology, University of Montreal Health Center, Montreal, Canada Article info Article history: Accepted June 20, 2011 Published online ahead of print on June 29, 2011 Keywords: Prostate cancer Cancer-specific mortality Other-cause mortality Competing-risks regression Nomogram Please visit europeanurology to read and answer questions on-line. The EU-ACME credits will then be attributed automatically. Abstract Background: Initial treatment options for low-risk clinically localized prostate cancer (PCa) include radical (RP) or observation. Objective: To examine cancer-specific mortality (CSM) after accounting for other-cause mortality (OCM) in PCa patients treated with either RP or observation. Design, setting, and participants: Using the Surveillance Epidemiology and End Results Medicare-linked database, a total of patients 65 yr with localized (T1/2) PCa were identified ( ). Intervention: RP and observation. Measurements: Propensity-score matching was used to adjust for potential selection biases associated with treatment type. The matched cohort was randomly divided into the development and validation sets. Competing-risks regression models were fitted and a competing-risks nomogram was developed and externally validated. Results and limitations: Overall, (49.8%) patients were treated with RP versus (50.2%) with observation. Propensity score matched analyses derived matched pairs. In the development cohort, the 10-yr CSM rate was 2.8% ( %) for RP versus 5.8% ( %) for observation (absolute risk reduction: 3.0%; relative risk reduction: 0.5%; p < 0.001). In multivariable analyses, the CSM hazard ratio for RP was 0.48 ( ) relative to observation (p < 0.001). The competing-risks nomogram discrimination was 73% and 69% for prediction of CSM and OCM, respectively, in external validation. The nature of observational data may have introduced a selection bias. Conclusions: On average RP reduces the risk of CSM by half in patients aged 65 yr, relative to observation. The individualized protective effect of RP relative to observation may be quantified with our nomogram. Crown Copyright # 2011 Published by Elsevier B.V. on behalf of European Association of Urology. All rights reserved. 1 Both authors contributed equally to this article. * Corresponding author. Cancer Prognostics and Health Outcomes Unit, University of Montreal Health Center, Montreal, Quebec, H2X 3J4, Canada. Tel ext ; Fax: address: mcw.sun@umontreal.ca (M. Sun) /$ see back matter Crown Copyright # 2011 Published by Elsevier B.V. on behalf of European Association of Urology. All rights reserved. doi: /j.eururo

2 EUROPEAN UROLOGY 60 (2011) Introduction Prostate cancer (PCa) is the second-leading cause of cancer death in men [1]. However, most contemporary patients have clinically localized and potentially indolent tumors at diagnosis [2]. Of these, the majority succumb to death from causes other than PCa [3,4]. This raises concerns about overtreatment [5]. To date, one contemporary randomized trial has tested the effect of radical (RP) versus watchful waiting on cancer-control outcomes in PCa patients [6]. This trial showed improved overall survival and cancerspecific mortality (CSM) when RP was provided instead of watchful waiting. An update of the trial was recently published and included a subanalysis demonstrating that the benefit of RP was limited to younger men (65 yr) [7]. Conversely, Wong et al. [8] examined patients 65 yr old diagnosed with PCa within the Surveillance Epidemiology and End Results (SEER) Medicare-linked database between 1991 and In that report, active treatment (RP or radiotherapy) improved CSM in comparison to observation. However, the authors did not account for the effect of other-cause mortality (OCM) on CSM. Consequently, benefits of active treatment may have been overestimated [9,10]. To address this important limitation, we relied on competing-risks methodology [10] to test the relationship between RP versus observation and CSM, after accounting for OCM. We hypothesized that significant CSM differences may exist according to treatment type. Subsequently, we developed and externally validated a competing-risks nomogram to provide clinicians with a tool capable of estimating the individual probability of CSM and OCM according to treatment type. 2. Methods 2.1. Data source We relied on the SEER-Medicare insurance program linked database, which is 98% complete for case ascertainment. The SEER regions represented approximately 14% of the US population before 2000 and 26% thereafter. The Medicare insurance program encompasses approximately 97% of the US population aged 65 yr. Linkage to the SEER database is complete for approximately 93% of patients [11] Study population Between 1992 and 2005, we identified men aged 65 yr with nonmetastatic PCa and both Medicare Part A and Part B claims available and who were not enrolled in a health maintenance organization throughout the duration of the study. Patients were not included if PCa was diagnosed at autopsy or on death certificate only, or if their original or current reason for Medicare entitlement was listed as disability or a Medicare status code including disability. Patients with T3/T4 tumors (n = 8556), anaplastic or unknown grade (n = 5657), unknown stage (n = 227), aged 80 yr at diagnosis (n = 18880), or missing socioeconomic information (n = 971) were excluded. This resulted in patients included in our study Variable definition Patient age was obtained from the Medicare file, while race, marital status, population density, and SEER registry were provided by the SEER data. We used the percentage of the 2000 whole US population Census tract with a 4-yr college education and the median income per census tract as proxies for socioeconomic status. The Charlson comorbidity index (CCI) was derived from Medicare claims during the year prior to PCa diagnosis [12]. Tumor grade is reported in the SEER database as well differentiated (Gleason score, 2 4), moderately differentiated (Gleason score, 5 7), or poorly differentiated (Gleason score, 8 10). Clinical extension information provided by the SEER was used to define tumor clinical stage [8]. Treatment type was identified by searching outpatient claims and Part A and Part B Medicare files for the appropriate International Classification of Disease (ICD) 9th revision, and Healthcare Common Procedure Coding System codes during the 6 mo following PCa diagnosis [8]. RP was identified using the Current Procedural Terminology, 4th edition (CPT-4) codes: 55840, 55842, 55845, and [13]. Observation was identified via the absence of active treatment codes (eg, RP, external beam radiation, radiation implants, brachytherapy, hormonal therapy, orchiectomy) Outcomes Underlying cause of death information was provided by the SEER data and, where previously validated in the context of PCa, was shown to be highly reliable (91%) [14]. Patients who succumbed to PCa (ICD or ICD-10 C619) were classified as CSM, while patients who succumbed to all other causes were classified as OCM. Data on cause-specific mortality were available throughout the end of The duration of survival was defined as the interval from the date of diagnosis to the Medicare date of mortality Statistical analyses Frequencies and proportions were reported for categorical variables. Means, medians, and ranges were reported for continuously coded variables. The independent sample t test and chi-square test compared the statistical significance of differences in means and proportions, respectively. First, we attempted to adjust for the selection bias inherent in observational data using propensity-score matching [15,16]. The propensity to undergo RP was calculated using a logistic regression model that adjusted for age at diagnosis, race, marital status, socioeconomic status, CCI, population density, clinical stage, tumor grade, year of diagnosis, and SEER registry. The nearest-neighbor method, with a caliper width of 0.2 of the standard deviation of the logit, was used. This method results in a modest residual bias and highest precision [17]. The standardized difference measure was used to test how well the controls match the cases [18]. Second, we randomly divided the propensity score matched cohort (n = ) into two equally sized groups: the development and the external validation cohorts. In the development cohort, cumulative incidence plots were used to graphically depict CSM and OCM rates. Statistical significance of differences in survival rates was assessed with the Gray test [19]. Due to the protracted natural history of PCa, a large proportion of patients may succumb to deaths other than cancer [3,4,9,20]. Thus competing-risks regression analyses were used. These avoid overestimation of CSM, as censoring due to OCM may artificially reduce the pool of individuals at risk of CSM events [10]. Covariates were age at diagnosis, race, CCI, clinical stage, and tumor grade. A sensitivity analysis was used to measure the potential effect of an unmeasured confounder on the relationship between treatment type and mortality (CSM and OCM) [21,22] Coefficients of the multivariable analyses were used to construct a competing-risks nomogram. Its calibration and

3 922 EUROPEAN UROLOGY 60 (2011) Table 1 Descriptive characteristics of patients treated with radical versus observation for prostate cancer between 1992 and 2005 within the Surveillance Epidemiology and End Results (SEER) Medicare-linked database and treatment typepropensity score matched patients Characteristics Entire cohort (N = ) Propensity score matched cohort (n = ) Radical n = (49.8%) Observation n = (50.2%) p value Radical n = (50%) Observation n = (50%) Standardized difference Age, yr Mean < Median Range Race (%) White (89.6) (82.2) < (86.4) 9947 (85.2) Black 1334 (6.0) 2466 (11.0) 956 (8.2) 1059 (9.1) 0.8 Other 984 (4.4) 1521 (6.8) 627 (5.4) 663 (5.7) 0.3 Marital status (%) Married (82.7) (65.8) < (76.1) 8657 (74.2) Unmarried * 3847 (17.3) 7668 (34.2) 2788 (23.9) 3012 (25.8) 1.9 Annual median income, USD (%) (21.7) 6349 (28.3) < (24.1) 2956 (25.3) (26.1) 5657 (25.2) 2990 (25.6) 2974 (25.5) (26.3) 5302 (23.6) 3001 (25.7) 2938 (25.2) (25.8) 5142 (22.9) 2863 (24.5) 2801 (24.0) 0.5 College education (% of persons) (22.0) 6023 (26.8) < (23.5) 2834 (24.3) (25.0) 5693 (25.4) 2946 (25.2) 2994 (25.7) (25.9) 5277 (23.5) 2934 (25.1) 2831 (24.3) (27.1) 5457 (24.3) 3047 (26.1) 3010 (25.8) 0.3 Charlson comorbidity index (%) (50.6) 9642 (42.9) < (48.6) 5715 (49.0) (29.6) 5882 (26.2) 3333 (28.6) 3181 (27.3) (12.2) 3344 (14.9) 1535 (13.2) 1521 (13.0) (7.6) 3582 (16.0) 1131 (9.7) 1252 (10.7) 0.1 Population density (%) Metropolitan (85.6) (84.6) (85.6) 9971 (85.4) Nonmetropolitan 3206 (14.4) 3455 (15.4) 1680 (14.4) 1698 (14.6) 0.1 Clinical stage (%) T (33.5) (51.8) < (37.1) 4707 (40.4) T2a/b (50.9) 9293 (41.4) 5776 (49.5) 5759 (49.4) 0.1 T2c 3474 (15.6) 1528 (6.8) 1561 (13.4) 1203 (10.3) 3.0 Tumor grade, Gleason score (%) (4.9) 3941 (17.6) < (7.8) 1072 (9.2) (68.2) (67.6) 8053 (69.0) 8292 (71.1) (26.9) 3328 (14.8) 2701 (23.1) 2305 (19.8) 3.3 Year of diagnosis (%) (8.7) 1429 (6.4) < (8.4) 877 (7.5) (6.1) 1500 (6.7) 715 (6.1) 760 (6.5) (5.7) 1365 (6.1) 642 (5.5) 649 (5.6) (6.2) 1407 (6.3) 727 (6.2) 729 (6.2) (5.8) 1264 (5.6) 679 (5.8) 648 (5.6) (5.9) 1224 (5.5) 667 (5.7) 645 (5.5) (4.9) 1105 (4.9) 553 (4.7) 557 (4.8) (5.2) 1101 (4.9) 624 (5.3) 582 (5.0) (8.4) 1911 (8.5) 1003 (8.6) 1018 (8.7) (9.0) 2030 (9.0) 1036 (8.9) 1065 (9.1) (9.3) 2190 (9.8) 1127 (9.7) 1156 (9.9) (9.0) 2003 (8.9) 1008 (8.6) 1067 (9.1) (8.3) 1929 (8.6) 984 (8.4) 965 (8.3) (7.6) 1992 (8.9) 929 (8.0) 951 (8.1) 0.1 SEER registry (%) San Francisco 1268 (5.7) 1328 (5.9) < (6) 715 (6.1) Connecticut 1205 (5.4) 2087 (9.3) 748 (6.4) 829 (7.1) Detroit 2327 (10.5) 3119 (13.9) 1370 (11.7) 1412 (12.1) 0.3 Hawaii 309 (1.4) 347 (1.5) 169 (1.4) 173 (1.5) 0 Iowa 2178 (9.8) 1914 (8.5) 986 (8.4) 981 (8.4) 0 New Mexico 978 (4.4) 992 (4.4) 545 (4.7) 533 (4.6) 0.1 Seattle 2353 (10.6) 1953 (8.7) 1172 (10) 1138 (9.8) 0.2 Utah 1665 (7.5) 1221 (5.4) 819 (7.0) 786 (6.7) 0.2 Atlanta 802 (3.6) 751 (3.3) 388 (3.3) 393 (3.4) 0 San Jose 680 (3.1) 744 (3.3) 379 (3.2) 390 (3.3) 0 Los Angeles 3255 (14.6) 2373 (10.6) 1661 (14.2) 1482 (12.7) 1.5 Rural Georgia 49 (0.2) 71 (0.3) 24 (0.2) 26 (0.2) 0

4 EUROPEAN UROLOGY 60 (2011) Table 1 (Continued ) Characteristics Entire cohort (N = ) Propensity score matched cohort (n = ) Radical n = (49.8%) Observation n = (50.2%) p value Radical n = (50%) Observation n = (50%) Standardized difference Greater California 2619 (11.8) 2237 (10) 1293 (11.1) 1313 (11.3) 0.1 Kentucky 652 (2.9) 823 (3.7) 350 (3.0) 363 (3.1) 0.1 Louisiana 980 (4.4) 777 (3.5) 478 (4.1) 467 (4.0) 0 New Jersey 924 (4.2) 1713 (7.6) 588 (5.0) 668 (5.7) 0.6 * Includes single, separated, divorced, or widowed. discrimination (c-index) was tested in the external validation cohort [23 25]. 3. Results 3.1. Patients baseline characteristics Between 1992 and 2005, (47.8%) and (50.2%) patients were treated with RP or observation, respectively. Baseline patient, socioeconomic, and regional differences were recorded between the two treatment types (Table 1). Moreover, RP patients harbored higher clinical stage (T2c: 16% vs 7%; p < 0.001) and tumor grade (Gleason score 8 10: 27% vs 15%; p < 0.001) relative to observation patients. Propensity-score matching resulted in a cohort of matched pairs with mean standardized differences of < 10% between the two groups (Table 1). The development and validation cohorts were further stratified (Table 2). Patient and clinical characteristics between the two groups were comparable Survival analyses During the study, the number of patients who died of PCa was 224 versus 518 in the RP and observation groups, respectively (Table 3). For the same respective groups, the number of patients who died of OCM was 2224 versus CSM rates at 5 and 10 yr were, respectively, 0.6% and 2.8% for RP versus 1.8% and 5.8% for observation in the development cohort (p < 0.001) (Fig. 1). OCM rates were Table 2 Descriptive statistics of treatment type-propensity score matched patients with prostate cancer treated with radical versus observation between 1992 and 2005 and within the Surveillance Epidemiology and End Results (SEER) Medicarelinked database * Characteristics Development cohort External validation cohort p value Age, yr Mean Median Range Treatment type (%) Observation 5760 (49.4) 5909 (50.6) 0.1 Radical 5909 (50.6) 5760 (49.4) Race (%) White (86.0) 9998 (85.7) 0.6 Black 1006 (8.6) 1009 (8.6) Other 628 (5.4) 662 (5.7) Marital status (%) Married 8737 (74.9) 8801 (75.4) 0.3 Unmarried ** 2932 (25.1) 2868 (24.6) Annual median income, USD (%) (25.1) 2845 (24.4) (25.0) 3048 (26.1) (25.7) 2941 (25.2) (24.2) 2835 (24.3) College education (% of persons) (24.0) 2774 (23.8) (25.8) 2933 (25.1) (24.4) 2915 (25.0) (25.8) 3047 (26.1) Charlson comorbidity index (%) (49.0) 5665 (48.5) (27.9) 3260 (27.9) (12.6) 1584 (13.6) (10.5) 1160 (9.9) Population density (%) Metropolitan 9954 (85.3) (85.7) 0.3 Nonmetropolitan 1715 (14.7) 1663 (14.3)

5 924 EUROPEAN UROLOGY 60 (2011) Table 2 (Continued ) Characteristics Development cohort External validation cohort p value Clinical stage (%) T (38.3) 4566 (39.1) 0.3 T2a/b 5819 (49.9) 5716 (49.0) T2c 1377 (11.8) 1387 (11.9) Tumor grade, Gleason score (%) (8.3) 1015 (8.7) (70.0) 8171 (70.0) (21.6) 2483 (21.3) Year of diagnosis (%) (7.7) 952 (8.2) (6.4) 731 (6.3) (5.8) 612 (5.2) (6.4) 709 (6.1) (5.8) 650 (5.6) (5.5) 666 (5.7) (4.8) 551 (4.7) (5.2) 595 (5.1) (8.7) 1002 (8.6) (8.9) 1067 (9.1) (9.8) 1134 (9.7) (8.8) 1050 (9) (8.2) 997 (8.5) (7.9) 953 (8.2) SEER registry (%) San Francisco 693 (5.9) 721 (6.2) 0.5 Connecticut 811 (7.0) 766 (6.6) Detroit 1429 (12.2) 1353 (11.6) Hawaii 158 (1.4) 184 (1.6) Iowa 962 (8.2) 1005 (8.6) New Mexico 547 (4.7) 531 (4.6) Seattle 1136 (9.7) 1174 (10.1) Utah 823 (7.1) 782 (6.7) Atlanta 405 (3.5) 376 (3.2) San Jose 384 (3.3) 385 (3.3) Los Angeles 1554 (13.3) 1589 (13.6) Rural Georgia 24 (0.2) 26 (0.2) Greater California 1313 (11.3) 1293 (11.1) Kentucky 341 (2.9) 372 (3.2) Louisiana 480 (4.1) 465 (4.0) New Jersey 609 (5.2) 647 (5.5) * Patients were randomly stratified to a development cohort that consisted of (50%) patients and an external validation cohort that consisted of (50%) patients. ** Includes single, divorced, separated, divorced, or widowed. 7.0% and 21.5% for RP versus 15.6% and 37.0% for observation for the same respective time points (p < 0.001). In the multivariable competing-risks analyses (Table 4) that predicted CSM in the development cohort, the hazard ratio (HR) for RP was 0.48 (95% confidence interval [CI], ) relative to observation. In the multivariable competing-risks analyses that predicted OCM, the HR for RP was 0.57 (95% CI, ) relative to observation. Table 3 Survival status and cause of death information (n = ) Overall cohort (N = ) Development cohort (n = ) Validation cohort (n = ) RP Observation RP Observation RP Observation Patients, no Total deaths, no Causes of death Prostate cancer Other causes Follow-up, yr Mean Median Range RP = radical.

6 Table 4 Univariable and multivariable competing-risks regression models for prediction of cancer-specific mortality (CSM; after accounting for other-cause mortality [OCM]) and OCM (after accounting for CSM) in patients with organ-confined prostate disease within the development cohort, treated with radical (RP), versus observation between 1992 and 2005 and within the Surveillance Epidemiology and End Results (SEER) Medicare-linked database Predictors CSM OCM Univariable Multivariable Univariable Multivariable HR (95% CI) p value HR (95% CI) p value HR (95% CI) p value HR (95% CI) p value Treatment type Observation 1.00 (ref.) 1.00 (ref.) 1.00 (ref.) 1.00 (ref.) RP 0.48 ( ) < ( ) < ( ) < ( ) <0.001 Age, yr 1.07 ( ) < ( ) ( ) < ( ) <0.001 Race White 1.00 (Ref.) 1.00 (ref.) 1.00 (ref.) 1.00 (ref.) Black 1.26 ( ) ( ) ( ) < ( ) <0.001 Other 0.94 ( ) ( ) ( ) ( ) Charlson comorbidity index (ref.) 1.00 (ref.) 1.00 (ref.) 1.00 (ref.) ( ) ( ) ( ) < ( ) < ( ) ( ) ( ) < ( ) < ( ) ( ) ( ) < ( ) <0.001 Clinical stage T (ref.) 1.00 (ref.) 1.00 (ref.) 1.00 (ref.) T2a/b 1.08 ( ) ( ) ( ) ( ) 0.1 T2c 1.68 ( ) < ( ) ( ) < ( ) Tumor grade, Gleason score (ref.) 1.00 (ref.) 1.00 (ref.) 1.00 (ref.) ( ) < ( ) ( ) ( ) ( ) < ( ) < ( ) ( ) 0.3 HR = hazard ratio; 95% CI = 95% confidence interval; ref = referent category. EUROPEAN UROLOGY 60 (2011)

7 926 [(Fig._1)TD$FIG] EUROPEAN UROLOGY 60 (2011) Fig. 1 Cumulative incidence plots depicting cancer-specific mortality (CSM) and other-cause mortality (OCM) in prostate cancer patients from the development cohort treated with radical versus observation within the Surveillance Epidemiology and End Results Medicare-linked database. Results are represented as rates with the corresponding 95% confidence intervals. Obs = observation; RP = radical Sensitivity analyses Sensitivity analysis revealed that for virtually all examined HRs, the statistical significance of the reported protective effect of RP on either CSM or OCM would no longer be detectable if the prevalence of this unmeasured confounder was at least 30% higher in one group of patients versus another (Table 5) Competing-risks nomogram Within the nomogram, grade and treatment type represent the two most influential variables for CSM (Fig. 2). In predicting OCM, age and CCI were the most influential. In the validation cohort, nomogram discrimination in predicting 10-yr CSM and OCM was, respectively, 73% and 69%. Overestimation of CSM and OCM predictions was 2.4% and 16.5%, respectively (Fig. 3A). When nomogram calibration was restricted to patients with a complete follow-up of 10 yr, the model discrimination in predicting 10-yr CSM and OCM was respectively 75% and 70%, respectively, with maximum overestimation rates of 1.1% and 3.2%, respectively (Fig. 3B). 4. Discussion A recent randomized trial demonstrated an overall and CSM survival benefit of RP relative to watchful waiting in Scandinavian men [6,7]. Wong et al. relied on a North American observation cohort of individuals aged 65 yr and demonstrated a benefit in patients managed with active treatment relative to observation [8]. However, no adjustment was made for the effect of censoring due to OCM on CSM estimates. Lack of adjustment for OCM may overestimate the benefit of active treatment versus observation [9,10]. Our hypothesis stated that significant CSM differences may exist according to treatment type (RP vs observation)

8 EUROPEAN UROLOGY 60 (2011) Table 5 Sensitivity analyses estimating the effect of an unmeasured confounder on the hazard ratio of death Prevalence in radical patients, % Prevalence in observation patients, % Cancer-specific mortality Treated HR adjusted for unmeasured confounder (95% CI) confounder HR 1.50 confounder HR 2.0 confounder HR ( ) 0.61 ( ) 0.64 ( ) ( ) 0.79 ( ) 0.85 ( ) ( ) 1.01 ( ) 1.11 ( ) ( ) 0.60 ( ) 0.62 ( ) ( ) 0.76 ( ) 0.79 ( ) ( ) 0.95 ( ) 0.99 ( ) ( ) 0.59 ( ) 0.60 ( ) ( ) 0.73 ( ) 0.74 ( ) ( ) 0.90 ( ) 0.91 ( ) Prevalence in radical patients, % Prevalence in observation patients, %) Other-cause mortality Treated HR adjusted for unmeasured confounder (95% CI) confounder HR 1.50 confounder HR 2.0 confounder HR ( ) 0.74 ( ) 0.77 ( ) ( ) 0.92 ( ) 0.93 ( ) ( ) 1.13 ( ) 1.16 ( ) ( ) 0.72 ( ) 0.74 ( ) ( ) 0.91 ( ) 0.95 ( ) ( ) 1.14 ( ) ( ) 0.71 ( ) 0.72 ( ) ( ) 0.88 ( ) 0.89 ( ) ( ) 1.08 ( ) 1.09 ( ) HR = hazard ratio; CI = confidence interval. in PCa patients. To test this hypothesis, we examined the rates of RP versus observation, as well as the associated CSM, using competing-risks analyses [10], after accounting for OCM. Moreover, we developed the first competing-risks nomogram to assist clinicians in providing individual CSM and OCM risk estimates. The rate of RP use differed according to patient and socioeconomic factors. Moreover, regional differences were recorded between treatment types. Patients treated with RP had lower comorbidity and higher clinical stage and grade relative to observation patients. Second, CSM was significantly lower in RP patients. The relative estimate (HR: 0.48; 95% CI, ) recorded in our cohort corroborates those previously reported by Bill-Axelson et al. (HR: 0.65; 95% CI, ) [7] and Wong et al. (HR: 0.50; 95% CI, ) [8]. However, the consideration of OCM in our analyses provides more realistic CSM estimates. Taken together, our observations imply that the survival benefit of RP applies to individuals aged 65 yr. This represents an important consideration, since at diagnosis most PCa patients are within this age stratum [26], which makes it essential to verify the efficacy of active treatment in these patients. Unlike previous analyses, our results also focused on OCM. The latter was significantly lower after RP than after observation. Several factors may account for the protective effect of RP on OCM relative to observation. For example, it is possible that the rates of hormonal therapy and its side effects are lower after RP than observation [6,27]. It may also be possible that other forms of salvage therapy delivered at a more advanced age to observation patients result in higher OCM rates. OCM benefit of RP may be also attributable to an unmeasured confounder. To address this hypothesis, we performed a sensitivity analysis. It showed that given an unmeasured confounder with a strong effect on CSM (HR: 3.0), it should be at least 30% more frequent in observation to obliterate the protective effect of RP on CSM. It is noteworthy that for all measured covariates, the differences in prevalence between RP and observation patients did not exceed 18%. Consequently, it is highly unlikely that an unmeasured confounder would be 30% more prevalent in one patient group than the other. Similar findings were observed when OCM was targeted as an end point. Our analyses provide an estimate of the protective effect of RP on CSM and OCM for an average patient. However, the magnitude of this effect may vary according to individual patient characteristics. To account for those individual differences, we developed and externally validated the first competing-risks nomogram predicting CSM and/or OCM between two competitive treatment alternatives for PCa. To date, such comparisons could not be made with existing studies or nomograms [28]. Our study is not devoid of limitations. First, the SEER- Medicare-linked database is based on claim files; it is primarily a billing system, which precludes detailed clinical information that might have contributed to the findings we observed. Second, findings originating from observational data contain an inherent selection bias. To address this issue, we relied on propensity score matched analysis, which is a powerful method that can overcome some of the limitations associated with the conventional covariate-based multivariable modeling. This allowed us to balance all potential confounding

9 928 [(Fig._2)TD$FIG] EUROPEAN UROLOGY 60 (2011) Fig. 2 Nomogram evaluating 10-yr cancer-specific mortality (CSM) and other-cause mortality (OCM) rates in patients with prostate cancer treated with radical versus observation. Total point values are independently calculated for each cause of death and then applied to the corresponding probability scale at the bottom of the figure. To obtain nomogram-predicted probability of CSM and OCM: (1) Locate patient values at each axis; (2) drawa vertical line to the Point axis to determine how many points are attributed for each variable value; (3) total the points for all variables; (4) locate the sum on the Total Point line to assess the individual probability of CSM and OCM. covariates according to treatment groups. Nonetheless, a validation of the current findings using other means of statistical analyses and in future randomized controlled studies is warranted [29,30]. It is possible that patients with localized PCa were likely diagnosed early with prostate-specific antigen (PSA) and, consequently, differ from the symptomatic population relative to the lower urinary tract. Previous studies relying on referral populations were also limited by this factor [8]. In addition, as in all studies comparing surgical intervention to another treatment option, the possibility of upstaging and upgrading should be considered. To reduce the bias associated with upstaging, we used the clinical extension classification according to the SEER coding system, which relies exclusively on preoperative clinical examination and omits the information obtained from the RP pathologic specimen. That being said, exclusive clinical stage data were available only for patients diagnosed from 1994 onward.

10 [(Fig._3)TD$FIG] EUROPEAN UROLOGY 60 (2011) treatments were also limited by the availability of biopsy Gleason versus pathologic Gleason scores [8]. No distinction was made in the type of RP that patients received (minimally invasive vs open approach), and we were not able to adjust for procedures that were performed with or without robotic assistance. Moreover, the rate of pelvic lymph node dissection and its extent were also unaccounted for. Nevertheless, a survival benefit of surgical intervention over watchful waiting was observed in this patient cohort regardless of the type of RP and absence or presence of PLND. We were unable to adjust for PSA, thus it was not possible to further stratify patients according to this variable. However, the latter has been shown to have only an intermediate effect on survival [33,34]. Moreover, we stratified patients according to tumor stage and grade, which represent important proxies of the extent of the disease. We did not examine data on sequential treatments. However, it is unlikely that patients treated with RP had a higher rate of sequential treatment than patients treated with observation [6]. Last but not least, the discrimination of the nomogram, especially for the prediction of OCM is not perfect (69%). However, it is comparable to previous competing-risks nomogram accuracy estimates [23]. Moreover, it is possible that an even longer follow-up is needed to improve the discrimination accuracy as more events occur and the effect of censored data will be minimized. This was demonstrated in our subanalyses that focused on patients diagnosed before 1998 who had follow-up through the end of Conclusions In patients aged 65 yr, on average, RP reduces the risk of CSM by half relative to observation. The individualized protective effect of RP relative to observation may be quantified with our nomogram. Author contributions: Maxine Sun had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Fig. 3 Calibration plot depicting the relationship between the nomogram predicted versus observed 10-yr cancer-specific mortality (triangles) and other-cause mortality (circles) (A) in the external validation cohort and (B) in patients with complete 10 yr follow up. Individuals were grouped by decile of nomogram-predicted 10-yr probabilities. Between 1992 and 1993, pathologic stage, rather than clinical stage, was recorded, when the former was more advanced. Therefore, RP patients in these 2 yr may have had their pathologic stage recorded rather than their clinical stage, similar to previous reports [8,31,32]. However, we performed a subanalysis using exclusively clinical stage data available from 1994 onward. Results remained virtually the same (data not shown). Similarly, it may be argued that upgrading has become less likely due to the now common practice of collecting 10 to 12 biopsies, which results in a high correlation between pre- and postsurgical tumor grade. Moreover, previous studies comparing surgery to other Study concept and design: Abdollah. Acquisition of data: Abdollah, Sun, Tian, Karakiewicz. Analysis and interpretation of data: Abdollah, Sun, Schmitges, Tian, Karakiewicz. Drafting of the manuscript: Abdollah, Karakiewicz. Critical revision of the manuscript for important intellectual content: Jeldres, Shariat, Briganti, Perrotte, Karakiewicz. Statistical analysis: Abdollah, Tian. Obtaining funding: Perrotte, Montorsi, Karakiewicz. Administrative, technical, or material support: Karakiewicz. Supervision: Montorsi, Karakiewicz. Other (specify): None. Financial disclosures: I certify that all conflicts of interest, including specific financial interests and relationships and affiliations relevant to the subject matter or materials discussed in the manuscript (eg, employment/affiliation, grants or funding, consultancies, honoraria, stock ownership or options, expert testimony, royalties, or patents filed, received, or pending), are the following: None. Funding/Support and role of the sponsor: Pierre I. Karakiewicz is partially supported by the University of Montreal Health Centre Urology

11 930 EUROPEAN UROLOGY 60 (2011) Specialists, Fonds de la Recherche en Sante du Quebec, the University of Montreal Department of Surgery and the University of Montreal Health Centre (CHUM) Foundation. Acknowledgment statement: We would like to thank Dr. Yu-Ning Wong for her help in providing the codes for treatment type. References [1] Jemal A, Siegel R, Ward E, et al. Cancer statistics, CA Cancer J Clin 2009;59: [2] Shao YH, Demissie K, Shih W, et al. Contemporary risk profile of prostate cancer in the United States. J Natl Cancer Inst 2009;101: [3] Abdollah F, Sun M, Thuret R, et al. A competing-risks analysis of survival after alternative treatment modalities for prostate cancer patients: Eur Urol 2011;59: [4] Albertsen PC, Hanley JA, Fine J. 20-year outcomes following conservative management of clinically localized prostate cancer. JAMA 2005;293: [5] Stattin P, Holmberg E, Johansson J, et al. Outcomes in localized prostate cancer: National Prostate Cancer Register of Sweden follow-up study. J Natl Cancer Inst 2009;102: [6] Bill-Axelson A, Holmberg L, Ruutu M, et al. Radical versus watchful waiting in early prostate cancer. N Engl J Med 2005;352: [7] Bill-Axelson A, Holmberg L, Filén F, et al. Radical versus watchful waiting in localized prostate cancer: the Scandinavian Prostate Cancer Group-4 randomized trial. J Natl Cancer Inst 2008;100: [8] Wong YN, Mitra N, Hudes G, et al. Survival associated with treatment vs observation of localized prostate cancer in elderly men. JAMA 2006;296: [9] Albertsen PC, Hanley JA, Murphy-Setzko M. Statistical considerations when assessing outcomes following treatment for prostate cancer. J Urol 1999;162: [10] Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 1999;94: [11] Warren JL, Klabunde CN, Schrag D, et al. Overview of the SEER- Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care 2002;40(Suppl 8), IV [12] Klabunde CN, Potosky AL, Legler JM, et al. Development of a comorbidity index using physician claims data. J Clin Epidemiol 2000;53: [13] American Medical Association. Current procedural terminology, ed 4. Chicago, IL: American Medical Association; [14] Penson DF, Albertsen PC, Nelson PS, et al. Determining cause of death in prostate cancer: are death certificates valid? J Natl Cancer Inst 2001;93: [15] Rosenbaum PR, Rubin DB. Reducing bias in observational studies using subclassification on the propensity score. J Amer Statistical Assoc 1984;79: [16] D Agostino Jr RB, D Agostino Sr RB. Estimating treatment effects using observational data. JAMA 2007;297: [17] Austin PC. Some methods of propensity-score matching had superior performance to others: results of an empirical investigation and Monte Carlo simulations. Biom J 2009;51: [18] Stukel TA, Fisher ES, Wennberg DE, et al. Analysis of observational studies in the presence of treatment selection bias: effects of invasive cardiac management on AMI survival using propensity score and instrumental variable methods. JAMA 2007;297: [19] Gray JR. A class of K-sample tests for comparing the cumulative incidence of a competing risk. Ann Statist 1988;16: [20] Berglund A, Garmo H, Tishelman C, et al. Comorbidity, treatment and mortality: a population-based cohort study of prostate cancer in PCBaSe Sweden. J Urol 2011;185: [21] Lin DY, Psaty BM, Kronmal RA. Assessing the sensitivity of regression results to unmeasured confounders in observational studies. Biometrics 1998;54: [22] Mitra N, Heitjan DF. Sensitivity of the hazard ratio to nonignorable treatment assignment in an observational study. Stat Med 2007;26: [23] Kutikov A, Egleston BL, Wong YN, et al. Evaluating overall survival and competing risks of death in patients with localized renal cell carcinoma using a comprehensive nomogram. J Clin Oncol 2010; 28: [24] Kattan MW, Heller G, Brennan MF. A competing-risks nomogram for sarcoma-specific death following local recurrence. Stat Med 2003;22: [25] Lughezzani G, Sun M, Budäus L, et al. Population-based external validation of a competing-risks nomogram for patients with localized renal cell carcinoma. J Clin Oncol 2010;28:e , author reply e301. [26] Miller DC, Gruber SB, Hollenbeck BK, et al. Incidence of initial local therapy among men with lower-risk prostate cancer in the United States. J Natl Cancer Inst 2006;98: [27] Isbarn H, Boccon-Gibod L, Carroll PR, et al. Androgen deprivation therapy for the treatment of prostate cancer: consider both benefits and risks. Eur Urol 2009;55: [28] Stephenson AJ, Kattan MW, Eastham JA, et al. Prostate cancer-specific mortality after radical for patients treated in the prostate-specific antigen era. J Clin Oncol 2009;27: [29] Hadley J, Yabroff R, Barrett M, et al. Comparative effectiveness of prosate cancer treatments: evaluating statistical adjustments for confounding in observational data. J Natl Cancer Inst 2010;102: [30] Wilt T, Brawer M, Barry M, et al. The prostate cancer intervention versus observation trial: VA/NCI/AHRQ Cooperative Studies Program #407 (PIVOT): design and baseline results of a randomized controlled trial comparing radical to watchful waiting for men with clinically localized prostate cancer. Contemp Clin Trials 2009;30:81 7. [31] Liu L, Coker AL, Du XL, et al. Long-term survival after radical compared to other treatments in older men with local/regional prostate cancer. J Surg Oncol 2008;97: [32] Cooperberg M, Vickers AJ, Broering J, et al. Comparative risk-adjusted mortality outcomes after primary surgery, radiotherapy, or androgen-deprivation therapy for localized prostate cancer. Cancer 2010;116: [33] Kim-Sing C, Pickles T. Intervention after PSA failure: examination of intervention time and subsequent outcomes from a prospective patient database. Int J Radiat Oncol Biol Phys 2004;60: [34] Pound C, Partin A, Eisenberger M, et al. Natural history of progression after PSA elevation following radical. JAMA 1999;281:

A Non Cancer-Related Survival Benefit Is Associated With Partial Nephrectomy

A Non Cancer-Related Survival Benefit Is Associated With Partial Nephrectomy EUROPEAN UROLOGY 61 (2012) 725 731 available at www.sciencedirect.com journal homepage: www.europeanurology.com Platinum Priority Kidney Cancer Editorial by Alexander Kutikov, Marc C. Smaldone, Brian L.