Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder

EUO-165; No. of Pages 8 E U R O P E A N U R O L O GY O N C O L O G Y X X X ( 2 0 18 ) X X X – X X X

available at www.sciencedirect.com journal homepage: euoncology.europeanurology.com

Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder Devin N. Patel a, Michael Luu b, Zachary S. Zumsteg b,c, Timothy J. Daskivich a,* a

Department of Surgery, Division of Urology, Cedars-Sinai Medical Center, Los Angeles, CA, USA;

b

Samuel Oschin Comprehensive Cancer Institute,

Cedars-Sinai Medical Center, Los Angeles, CA, USA; c Department of Radiation Oncology, Cedars-Sinai Medical Center, Los Angeles, CA, USA

Article info

Abstract

Article history: Accepted December 24, 2018

Background: Current pathological nodal staging for bladder cancer is based on lymph node (LN) location but not on the number of positive LNs. Objective: We sought to improve prognostic classification by creating a novel staging system incorporating positive LN burden. Design, setting, and participants: We sampled 12 515 patients with muscleinvasive bladder cancer (MIBC) from the National Cancer Database (NCDB) and 5928 MIBC patients from the Surveillance, Epidemiology, and End Results (SEER) database for our development and validation cohorts, respectively. Outcome measurements and statistical analysis: Multivariable Cox proportional hazards analysis with restricted cubic splines was used to assess the association between the number of metastatic LNs and overall mortality (OM). A novel staging system was derived by recursive partitioning analysis (RPA) in NCDB and was validated in SEER by assessing discrimination (Harrel’s c-index) and calibration (mean absolute prediction error). Results and limitations: Mortality risk increased continuously with more metastatic LNs; the effect was most pronounced up to four LNs (hazard ratio [HR] 1.17, 95% confidence interval [CI] 1.12–1.22) and attenuated beyond four nodes (HR 1.03, 95% CI 1.02–1.05). RPA generated a novel staging system predicting mortality by metastatic nodal number with cutpoints at zero (reference), one (HR 1.57, 95% CI 1.46–1.69), two to three (HR 2.03, 95% CI 1.88–2.19), four to seven (HR 2.46, 95% CI 2.25–2.70), and more than seven (HR 3.83, 95% CI 3.38–4.33) positive LNs. Location of LN involvement was not a significant predictor of OM. In external validation, the novel staging system showed good risk discrimination (optimism corrected c-index 0.677, 95% CI 0.672– 0.682) and calibration (mean absolute prediction error 0.011 for 5-yr OM). Results are limited by development and validation using secondary data. Conclusions: The number of metastatic LNs predicts mortality better than LN location and may improve pathological nodal staging in MIBC. Patient summary: This retrospective study found that the number of metastatic lymph nodes more accurately predicts survival than the location of metastatic lymph nodes in patients with muscle-invasive bladder cancer. This finding argues for change to the current bladder cancer staging system.

Associate Editor: Ashish Kamat Keywords: Bladder cancer Nodal staging Prognosis

© 2019 European Association of Urology. Published by Elsevier B.V. All rights reserved. * Corresponding author. Cedars-Sinai Medical Center, 863 West 3rd Street Suite 1070W, Los Angeles, CA 90048, USA. Tel.: +1 310 423 3497; Fax: +1 310 423 4711. E-mail address: [email protected] (T.J. Daskivich).

https://doi.org/10.1016/j.euo.2018.12.012 2588-9311/© 2019 European Association of Urology. Published by Elsevier B.V. All rights reserved.

Please cite this article in press as: Patel DN, et al. Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder. Eur Urol Oncol (2019), https://doi.org/10.1016/j.euo.2018.12.012

EUO-165; No. of Pages 8 2

E U R O P E A N U R O L O GY O N C O L O GY X X X ( 2 018 ) X X X – X X X

1.

Introduction

Individuals with unknown or less than five LNs sampled were excluded. The final analytic sample comprised 5928 individuals.

Muscle-invasive bladder cancer (MIBC) found to be metastatic to the lymph nodes (LNs) at the time of radical cystectomy is a common but understudied disease state [1]. While this finding is ominous with 5-yr overall survival (OS) of 25–40%, little information is available regarding prognostic risk stratification [2]. Discernment of mortality risk among these patients is critical for appropriate patient counseling regarding prognosis and can help clinicians identify patients at the highest risk of cancer mortality who would be best suited for clinical trial enrollment. The current American Joint Committee on Cancer (AJCC) bladder cancer pathological nodal staging system relies on the location of nodal involvement but not on the total LN burden for assigning nodal (N) stage. Yet, previous studies have suggested that the total metastatic nodal burden is a strong independent predictor of survival [2–5]. Despite this, no clear stratification paradigm incorporating the number of positive LNs has emerged, and it is unclear whether nodal location, metastatic nodal burden, or both should be used to optimize staging. To address this need, we investigated the relative impact of quantitative metastatic LN burden and nodal location on OS in a large population of patients with MIBC, with the intent of developing and validating an integrated pathological nodal staging system. We used the National Cancer Database (NCDB) to develop this system and the Surveillance, Epidemiology, and End Results (SEER) dataset for validation. We hypothesized that the total positive LN count would strongly predict OS, potentially obviating the need for including nodal location in staging and improving prognostic risk stratification of patients with node-positive MIBC.

2.3.

Statistical analysis

Baseline characteristics in patients with N0 and N-positive classification were compared using Student t test and Wilcoxon-Mann-Whitney’s test for continuous variables and using Pearson’s chi-square test for categorical variables. Median follow-up was determined by reverse Kaplan-Meier method. Survival functions were estimated using Kaplan-Meier method and compared using log-rank test. Univariate and multivariable survival analyses were performed using Cox proportional hazards models. The primary predictor in our models was the number of positive metastatic LNs. Covariates included the number of LNs examined, age, gender, race, facility type, education, comorbidity, income, diagnosis year, margins, and insurance status. Multivariable analysis variable selection was performed using a stepwise procedure optimized on the Akaike information criterion (AIC) [7]. The proportional hazards assumption was assessed by Schoenfeld residuals and the goodness-of-fit test [8]. Upon multivariable analysis, postoperative chemotherapy and tumor stage were found to violate the assumption of proportional hazards and were adjusted for using the stratified Cox model [9]. The number of positive metastatic LNs was modeled as a restricted cubic spline function to allow for nonlinear association with OS, with three knots at the 82nd, 88th, and 97th quantiles corresponding to one, two, and seven nodes, respectively. The optimal number of knots was determined on the basis of AIC, and knot location was placed on locations as recommended by Harrell [10]. The estimated association of the number of positive LNs was presented as a smoothed restricted cubic spline plot, where the log relative hazard is presented as a function of increasing positive LNs with zero positive LNs as a referent. Change point analysis was conducted with a piecewise linear regression model to find the optimal change point between the two linear segments [11]. The optimal change point location was used in a stratified analysis to obtain the estimated association within each segment. We used recursive partitioning analysis (RPA) to devise a new N classification system with independent nodal predictors of mortality from our multivariable model, to identify optimal risk stratification for

2.

Methods

the number of positive nodes in predicting OS. RPA was performed with a conditional inference tree that estimates a regression relationship

2.1.

Study population

by binary recursive partitioning in a conditional inference framework [12]. The performance of our model was compared with the AJCC N

Individuals for the development cohort were drawn from the NCDB, a

classification in multivariable analysis and assessed with c-indices.

hospital-based nationwide oncology outcomes database containing

Interval validation was performed with 1000 bootstrap replicates

information on approximately 70% of all new cancer diagnoses in the

correcting for possible optimism in c-indices [10].

USA [6]. Patients >18 yr old diagnosed with bladder carcinoma between

We externally validated our novel N classification system with the

January 1, 2004, and December 31, 2013, were eligible. Individuals with

SEER validation cohort. Kaplan-Meier survival curves were replicated

nonurothelial, non-MIBC, and metastatic disease were excluded

with our devised N classification system from our RPA model and

(Supplementary Fig. 1A). Patients with unknown or less than five LNs

compared with the development cohort. The calibration curve was

examined were omitted to eliminate cases of “LN sampling.” To reduce

constructed using an adaptive linear spline hazard regression with

the independent impact of margin status on survival, patients with

200 bootstrap resampling replicates to obtain unbiased calibration

positive margins were excluded in the development and validation of our

accuracy for 60-mo survival [13]. c-Indices were further estimated

staging system. The final analytic sample comprised 12 515 individuals. A

using bootstrap resampling with 1000 replicates to get optimism-

separate sensitivity analysis was carried out in patients receiving

corrected c-indices [10,14]. Statistical analyses were performed using R (version 3.5.1; R

neoadjuvant chemotherapy within 180 d of surgery and in patients with positive margins.

Foundation for Statistical Computing, Vienna, Austria) with a two-sided test and significance level of 0.05.

2.2.

Validation cohort

Individuals for the validation cohort were drawn from the SEER 18 registry. Patients >18 yr old with primary bladder urothelial carcinoma between 1973 and 2014 were eligible. Those with non-MIBC and metastatic disease were excluded (Supplementary Fig. 1B).

3.

Results

Among 12 515 patients meeting the inclusion criteria within the NCDB, median follow-up was 63 mo (95% confidence

Please cite this article in press as: Patel DN, et al. Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder. Eur Urol Oncol (2019), https://doi.org/10.1016/j.euo.2018.12.012

EUO-165; No. of Pages 8 3

E U R O P E A N U R O L O G Y O N C O L O GY X X X ( 2 0 18 ) X X X – X X X

interval [CI] 62–64). Overall median number of LNs examined among all patients was 14 (interquartile range [IQR] 9–21). Among patients with node-positive disease, the median number of identified positive LNs was 2 (IQR 1–4); of these, 94.4% had pelvic LN involvement and 5.6% had common iliac LN involvement (Table 1).

In univariate and multivariable Cox proportional hazards analysis, a higher number of metastatic LNs, but not nodal location, were associated with worse OS. Using a three-knot restricted cubic spline function, mortality risk escalated continuously with an increasing number of metastatic nodes without plateau (Fig. 1A). This relationship was

Table 1 – Baseline characteristics Characteristics

Age Mean (SD) Median (IQR) Age category (%) 65 >65 Sex (%) Male Female Race (%) White Black Other Charlson-Deyo comorbidity index (%) 0 1 2 Academic center (%) Nonacademic center Academic center Insurance (%) Uninsured Private Medicaid Medicare Other/unknown Lymph nodes examined Mean (SD) Median (IQR) Number of positive metastatic lymph nodes Mean (SD) Median (IQR) Pathological T stage (%) T2 T3 T4 Location (%) Pelvic Common iliac None Postoperative radiation (%) No Yes Postoperative chemotherapy (%) No Yes Income (%) Low Med High Education (%) Low Medium High Year (%) 2004–2006 2007–2009 2010–2013

Overall

N0

N+

n = 12515

n = 8657

n = 3858

68.23 (10.08) 69.00 (61.00, 76.00)

68.55 (10.01) 69.00 (62.00, 76.00)

67.52 (10.19) 68.00 (60.00, 75.00)

p value

<0.001 <0.001

4763 (38.1) 7752 (61.9)

3180 (36.7) 5477 (63.3)

1583 (41.0) 2275 (59.0)

<0.001

9302 (74.3) 3213 (25.7)

6453 (74.5) 2204 (25.5)

2849 (73.8) 1009 (26.2)

0.42

11 459 (91.6) 689 (5.5) 367 (2.9)

7942 (91.7) 455 (5.3) 260 (3.0)

3517 (91.2) 234 (6.1) 107 (2.8)

0.15

8698 (69.5) 2940 (23.5) 877 (7.0)

6014 (69.5) 2037 (23.5) 606 (7.0)

2684 (69.6) 903 (23.4) 271 (7.0)

0.99

5696 (45.5) 6819 (54.5)

3908 (45.1) 4749 (54.9)

1788 (46.3) 2070 (53.7)

0.219

333 4013 471 7383 315

215 2701 312 5216 213

118 1312 159 2167 102

0.001

(2.7) (32.1) (3.8) (59.0) (2.5)

(2.5) (31.2) (3.6) (60.3) (2.5)

(3.1) (34.0) (4.1) (56.2) (2.6)

16.61 (11.49) 14.00 (9.00, 21.00)

16.29 (11.37) 13.00 (8.00, 20.00)

17.32 (11.73) 14.00 (9.00, 22.00)

<0.001 <0.001

1.03 (2.76) 0.00 (0.00, 1.00)

0.00 (0.00) 0.00 (0.00, 0.00)

3.34 (4.11) 2.00 (1.00, 4.00)

<0.001 <0.001

4893 (39.1) 5978 (47.8) 1644 (13.1)

4114 (47.5) 3709 (42.8) 834 (9.6)

779 (20.2) 2269 (58.8) 810 (21.0)

<0.001

3642 (29.1) 216 (1.7) 8657 (69.2)

0 (0.0) 0 (0.0) 8657 (100.0)

3642 (94.4) 216 (5.6) 0 (0.0)

<0.001

12325 (98.5) 190 (1.5)

8587 (99.2) 70 (0.8)

3738 (96.9) 120 (3.1)

<0.001

8218 (65.7) 4297 (34.3)

6451 (74.5) 2206 (25.5)

1767 (45.8) 2091 (54.2)

<0.001

1977 (15.8) 6807 (54.4) 3731 (29.8)

1386 (16.0) 4696 (54.2) 2575 (29.7)

591 (15.3) 2111 (54.7) 1156 (30.0)

0.619

1834 (14.7) 7668 (61.3) 3013 (24.1)

1298 (15.0) 5285 (61.0) 2074 (24.0)

536 (13.9) 2383 (61.8) 939 (24.3)

0.274

3009(24.0) 3865(30.9) 5641(45.1)

2119 (24.5) 2620 (30.3) 3918 (45.3)

890 (23.1) 1245 (32.3) 1723 (44.7)

0.053

IQR = interquartile range; N = nodal; SD = standard deviation.

Please cite this article in press as: Patel DN, et al. Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder. Eur Urol Oncol (2019), https://doi.org/10.1016/j.euo.2018.12.012

EUO-165; No. of Pages 8 4

E U R O P E A N U R O L O GY O N C O L O GY X X X ( 2 018 ) X X X – X X X

Fig. 1 – Restricted cubic spline Cox model comparing hazard of overall mortality by (A) metastatic nodal count and (B) regional nodes examined. Adjusted hazard ratios (HRs) with an increasing number of positive lymph nodes (LNs) and LNs were examined. (A) The number of positive nodes was modeled as a restricted cubic spline with three nodes at 82nd, 88th, and 99th quantiles with an estimated change point found at four positive nodes. The predicted log relative hazard was plotted with a referent value of zero positive nodes. (B) The number of nodes examined was modeled as a linear predictor with an adjusted HR of 0.991 for every increasing node examined. The predicted log relative hazard was plotted with a referent value of five nodes examined.

nonlinear, with a change point identified at four metastatic LNs. Hazard of overall mortality per metastatic LN increased sharply up to four metastatic LNs (hazard ratio [HR] 1.168, 95% CI 1.120–1.218, p < 0.001). Beyond four metastatic LNs, the increase in overall mortality continued with each additional metastatic LN, albeit more slowly (HR 1.035, 95% CI 1.020–1.049, p < 0.001). Notably, there was no difference in survival for patients with pelvic versus common iliac nodal metastases in multivariable analysis (HR 1.027, 95% CI 0.864–1.221, p = 0.76). An increasing number of LNs examined were also associated with improved OS on univariable (HR 0.993, 95% CI 0.991–0.996, p < 0.001) and multivariable (HR 0.992; 95% CI 0.989–0.994, p < 0.001) analysis. The number of LNs examined exhibited a linear relationship with mortality with no change point identified (Fig. 1B).

RPA using all covariables independently associated with OS empirically generated a novel nodal pathological staging schema entirely driven by the number of positive LNs (Fig. 2). Multivariable Cox proportional hazards models showed excellent mortality risk stratification, with an incrementally increased risk of mortality in patients with one (N1; (HR 1.569, 95% CI 1.461–1.686, p < 0.001), two to three (N2a; HR 2.028, 95% CI 1.881–2.188, p < 0.001), four to seven (N2b; HR 2.461, 95% CI 2.247–2.695, p < 0.001), and eight or more (N3; HR 3.825, 95% CI 3.383–4.325, p < 0.001) positive LNs compared with those with no positive LNs (N0). Kaplan-Meier estimates of this model and the AJCC (8th edition) system are shown (Fig. 3). Estimated 5-yr OS rates were 52.1% for patients with zero metastatic LNs (N0), 32.2% for patients with one LN (N1), 21.2% for patients with two to three LNs (N2a), 18.4% for patients with four to seven LNs

Fig. 2 – Novel proposed nodal staging system developed by recursive portioning analysis in bladder cancer patients who did not receive neoadjuvant chemotherapy. Bonferroni-adjusted p values are given in the inner nodes, and Kaplan-Meier estimates for 5-yr overall survival are displayed in the terminal nodes. LN = lymph node; OS = overall survival.

Please cite this article in press as: Patel DN, et al. Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder. Eur Urol Oncol (2019), https://doi.org/10.1016/j.euo.2018.12.012

EUO-165; No. of Pages 8 E U R O P E A N U R O L O G Y O N C O L O GY X X X ( 2 0 18 ) X X X – X X X

5

Fig. 3 – Kaplan-Meier curves depicting overall survival by (A) current AJCC pathological nodal classification and (B) proposed pathological nodal classification in the NCDB. AJCC = American Joint Committee on Cancer; LN = lymph node; NCDB = National Cancer Database.

(N2b), and 9.3% for patients with eight or more metastatic LNs (N3). The optimism-corrected c-index for the proposed system showed slight but significant improvement in predictive ability (0.677, 95% CI 0.672–0.682) over the AJCC (8th edition) nodal classification system (0.674, 95% CI 0.673–0.689; Table 2). The novel staging system showed excellent discrimination and calibration in the SEER validation sample (Fig. 4A and B). Multivariable Cox proportional hazards showed an incrementally increased risk of mortality in patients with one (N1; HR 1.828, 95% CI 1.639–2.040, p < 0.001), two to three (N2a; HR 2.367, 95% CI 2.271–2.833, p < 0.001), four to seven (N2b; HR 2.897, 95% CI 2.545–3.298, p < 0.001), and eight or more (N3; HR 4.584, 95% CI 3.851–5.457, p < 0.001) positive LNs compared with those with no positive LNs (N0). Kaplan-Meier survival plots by novel nodal stage are shown in Fig. 4A. The optimism-corrected c-index for the prediction of 5-yr overall mortality was 0.651 (95% CI 0.648– 0.661) in our validation cohort and 0.677 (95% CI 0.672– 0.682) in our development cohort. Plots depicting observed to expected 5-yr overall mortality revealed superb calibration with an optimism-corrected slope of 0.987 and the mean absolute prediction error of 0.011 (Fig. 4B). In a sensitivity analysis analyzing only patients who received neoadjuvant chemotherapy, a similar relationship was seen between metastatic nodal count and overall mortality risk. The same was true in patients with positive surgical margins. Mortality risk again escalated continuously with increasing number of metastatic nodes (Supplementary Fig. 2A), with a change point at four metastatic LNs. The number of positive LNs up to 4 (HR 1.371, 95% CI 1.298–1.448, p < 0.001) was associated with a higher hazard of mortality. Beyond four nodes, each additional metastatic node showed a nonsignificant, smaller increase in risk (HR 1.017, 95% CI 0.983–1.053, p = 0.315). RPA analysis again identified strong risk

discrimination by increasing metastatic nodal burden, with slightly different cutpoints (Supplementary Fig. 2B). 4.

Discussion

MIBC is a highly aggressive malignancy, and among patients with LN metastasis found at the time of cystectomy, there is a particularly poor prognosis [15]. For such patients, it is critical to have detailed information regarding survival to inform prognosis and identify patients best suited for early enrollment into clinical trials. We demonstrate that the absolute number of metastatic LNs is a critical predictor of bladder cancer mortality, improving risk stratification over nodal location. We found that each positive LN confers an additional 17% increased risk of mortality through four positive nodes, whereas each successive positive node beyond this increases relative mortality by 1%. This may indirectly suggest that a modest positive nodal burden represents a potentially more treatable disease state than a larger burden, translating to larger observed differences in survival on a per-node basis among lower numbers of positive nodes compared with higher numbers of positive nodes. Furthermore, when nodal count was included in models predicting OS, nodal location was no longer a significant predictor of survival. These results suggest that pathological nodal staging for bladder cancer should prioritize nodal count over nodal location. Current bladder cancer staging systems are based on typical nodal spread pattern. However, data assessing the independent impact of nodal disease location on survival is limited [16], and the prognostic effect of positive LN location versus the number of positive LNs has not been studied. We performed such a head-to-head comparison in the NCDB and found that nodal location had no impact on survival when adjusting for the number of positive LNs in multivariable analysis. The predictive utility of the number

Please cite this article in press as: Patel DN, et al. Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder. Eur Urol Oncol (2019), https://doi.org/10.1016/j.euo.2018.12.012

EUO-165; No. of Pages 8 6

E U R O P E A N U R O L O GY O N C O L O GY X X X ( 2 018 ) X X X – X X X

Table 2 – Multivariable Cox proportional hazards model predicting overall survival by proposed pathological nodal classification and current AJCC pathological nodal classification Univariate survival analysis

Number of positive metastatic lymph nodes Number of lymph nodes examined Age Sex Male Female Race White Black Other Charlson-Deyo comorbidity index 0 1 2 Facility type Nonacademic center Academic center Insurance Uninsured Private Medicaid Medicare Other/unknown Year of diagnosis T stage T2 T3 T4 Location Pelvic Common iliac None Postoperative radiation No Yes Postoperative chemotherapy No Yes Income (%) Low Medium High Education (%) Low Medium High

a

Multivariable survival analysis

HR (95% CI)

p value

– 0.993 (0.991–0.996) 1.024 (1.021–1.026)

<0.001 <0.001 <0.001

1.000 1.000 (0.947–1.056)

– 0.995

1.000 1.195 (1.080–1.323) 0.915 (0.791–1.059)

– <0.001 0.235

1.000 1.184 (1.068–1.313) 0.912 (0.787–1.056)

– 0.001 0.219

1.000 1.246 (1.179–1.318) 1.582 (1.450–1.727)

– <0.001 <0.001

1.000 1.181 (1.117–1.249) 1.465 (1.341–1.599)

– <0.001 <0.001

1.000 0.943 (0.899–0.989)

– 0.015

1.000 1.019 (0.863–1.204) 1.291 (1.053–1.583) 1.479 (1.256–1.741) 1.119 (0.895–1.398) 1.006 (0.997–1.016)

– 0.823 0.014 <0.001 0.325 0.184

1.000 1.006 (0.851–1.190) 1.267 (1.033–1.553) 1.122 (0.946–1.330) 1.023 (0.817–1.281)

1.000 2.102 (1.989–2.222) 3.067 (2.853–3.297)

– <0.001 <0.001

c

1.000 1.344 (1.138–1.587) 0.482 (0.458–0.506)

– <0.001 <0.001

1.000 1.027 (0.864–1.221) 0.814 (0.737–0.898)

– 0.763 <0.001

1.000 2.028 (1.729–2.379)

– <0.001

1.000 1.435 (1.220–1.689)

– <0.001

1.000 0.989 (0.941–1.040)

– 0.664

1.000 0.937 (0.877–1.002) 0.844 (0.784–0.908)

– 0.058 <0.001

1.000 0.929 (0.868–0.994) 0.860 (0.798–0.927)

1.000 0.951 (0.889–1.018) 0.837 (0.773–0.905)

– 0.150 <0.001

b

HR (95% CI)

0.992 (0.989–0.994) 1.019 (1.016–1.023)

p value

<0.001 <0.001

b

b

– 0.941 0.023 0.188 0.845

b

c

– 0.034 <0.001

CI = confidence interval; HR = hazard ratio; LN = lymph node. The number of positive metastatic LNs was modeled using a restricted cubic spline function with three knots at the 82nd, 88th, and 97th quantiles corresponding to one, two, and seven nodes, respectively. b Variables dropped out of the model. c Multivariable model adjusted for postoperative chemotherapy and stage by stratification due to nonproportional hazards. a

of positive LNs persisted in a validation dataset from SEER. Our findings suggest that metastatic nodal count is a stronger predictor of survival than nodal location. These results argue for basing staging on the number of LNs rather than on their location [2]. Recognizing that metastatic nodal count is strongly predictive of OS, we developed a novel nodal staging schema using RPA, which empirically derives the best fit for risk stratification. As a testament to the strength of nodal count in predicting survival, the model eschewed all other covariates when identifying the ideal risk stratification

variables, including nodal location. The novel staging system also showed excellent risk discrimination and calibration in external validation using the SEER dataset. Our proposed system has several advantages. First, the system represents a concise stratification relying on a single variable. Furthermore, using the current AJCC system, 5-yr OS rates did not differ significantly between patients with pathological N2 (19.0%) and N3 (17.8%) disease [17]. In our proposed system, all N categories had distinct, nonoverlapping prognoses. Lastly, the proposed system stratifies risk over a greater spectrum, specifically for those at the highest

Please cite this article in press as: Patel DN, et al. Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder. Eur Urol Oncol (2019), https://doi.org/10.1016/j.euo.2018.12.012

EUO-165; No. of Pages 8 E U R O P E A N U R O L O G Y O N C O L O GY X X X ( 2 0 18 ) X X X – X X X

7

Fig. 4 – (A) Kaplan-Meier curves depicting overall survival by the proposed pathological nodal classification in SEER dataset. (B) Calibration curve showing observed versus predicted 5-yr overall survival based on the novel staging system in SEER dataset. The calibration curve depicts observed (black), ideal (gray), and optimism-corrected (blue) calibration for survival at 60 mo. Perfect calibration can be characterized by the gray ideal curve along the 45 line with a slope of 1. Predicted 60-mo survival corresponds to the predicted survival probability at 60 mo, and the fraction surviving 60 mo corresponds to the Kaplan-Meier survival estimates at 60 mo. Overfitting can be characterized by the area below the ideal curve and underfitting by the area above the ideal curve.

risk of poor prognosis. For example, patients with eight or more positive LNs were approximately half as likely to be alive at 5 and 10 yr from diagnosis as patients with four to seven positive LNs. Importantly, our results support the use of extended pelvic lymph node dissection (LND). Specifically, the relative risk of mortality was reduced by 1% for each additional benign or malignant LN examined. Evidence in favor of an extended LND has previously come from single-center, multicenter, and national database studies [18–20]. Although the exact reason for improved survival is unclear, several plausible explanations exist. An extended LND has been shown to improve outcomes for patients with node-positive and node-negative disease, suggesting a possible benefit of removal of micrometastatic disease [20,21]. Second, the degree of LND may be a surrogate for the quality of surgery and pathological examination [19,22]. Furthermore, an extended LND may provide improved information for pathological staging. We also found that the observed associations between the number of nodes harvested and the number of positive LNs with survival were similar in a sample of patients who received neoadjuvant chemotherapy. Specifically, the change point at which further positive nodes did not strongly impact survival and the overall ability of node count to stratify the risk of OS were consistent between those treated with or without neoadjuvant chemotherapy. That this information can be applied regardless of pretreatment status is important, since neoadjuvant chemotherapy is the guidelines-endorsed standard of care for eligible patients undergoing radical cystectomy.

Our study has several limitations. First, since cutoff nodal counts for our novel staging system were developed and validated using secondary data, our findings should be externally validated within large institutional databases, specifically in the context of a standardized LND template. Second, our main outcome from this study was OS, and not disease-specific survival. Our results, therefore, assume that bladder cancer is causing the majority of deaths among patients with advanced non–organ-confined disease, which is supported by the literature [23]. Furthermore, since staging criteria are traditionally based on OS rather than on disease-specific survival, OS is arguably a better outcome for the purposes of this analysis. Third, due to dataset limitations, we were unable to analyze other potential prognostic factors, such as LN size and the presence of extranodal extension, and were unable to characterize the nature of disease recurrence as either local or distant. We were also unable to capture subsequent therapies that patients may have received in the metastatic setting. However, it is important to note that our results were based on pathological and not on clinical nodal stage. Given the discrepancy between clinical and radiographic LN disease and results from pathological specimens, it is unclear whether our results can be translated to clinical staging patient.[24,25]. 5.

Conclusions

Among patients treated with radical cystectomy for MIBC, we found that metastatic nodal count predicted OS better than nodal location. Using empiric RPA, we identified

Please cite this article in press as: Patel DN, et al. Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder. Eur Urol Oncol (2019), https://doi.org/10.1016/j.euo.2018.12.012

EUO-165; No. of Pages 8 8

E U R O P E A N U R O L O GY O N C O L O GY X X X ( 2 018 ) X X X – X X X

optimal cutpoints for risk stratification at zero, one, two to three, four to seven, and eight or more positive LNs. This schema improved c-index over the traditional AJCC nodal staging system in internal validation, and showed excellent discrimination and calibration in an external validation dataset. These findings suggest that nodal count rather than nodal location should be the primary determinant of pathological nodal staging. Author contributions: Timothy J. Daskivich 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.

[5] Tarin TV, Power NE, Ehdaie B, et al. Lymph node-positive bladder cancer treated with radical cystectomy and lymphadenectomy: effect of the level of node positivity. Eur Urol 2012;61:1025–30. [6] National Cancer Database. https://www.facs.org/quality-programs/ cancer/ncdb; 2017. [7] Yamashita T, Yamashita K, Kamimura R. A stepwise AIC method for variable selection in linear regression. Commun Stat Theory Methods 2007;36:2395–403. [8] Harrell FE, Lee KL. Verifying assumptions of the Cox proportional hazards modelIn: Proceedings of the eleventh annual SAS Users Group International Conference. 1986. [9] Kleinbaum DG. Survival analysis, a self-learning text. Biom J 1998;40:107–8.

Study concept and design: Daskivich, Zumsteg. Acquisition of data: Luu, Patel. Analysis and interpretation of data: Luu, Daskivich, Patel, Zumsteg. Drafting of the manuscript: Patel, Luu. Critical revision of the manuscript for important intellectual content: Daskivich, Zumsteg. Statistical analysis: Luu, Daskivich. Obtaining funding: Daskivich. Administrative, technical, or material support: Daskivich. Supervision: Daskivich, Zumsteg. Other: None.

[10] Harrell Jr FE. Regression modeling strategies: with applications to linear models, logistic and ordinal regression, and survival analysis. Springer; 2015. [11] Toms JD, Lesperance ML. Piecewise regression: a tool for identifying ecological thresholds. Ecology 2003;84:2034–41. [12] Strasser H, Weber C. On the asymptotic theory of permutation statistics. 1999. [13] Kooperberg C, Stone CJ, Truong YK. Hazard regression. J Am Stat Assoc 1995;90:78–94. [14] Moons KG, Kengne AP, Grobbee DE, et al. Risk prediction models: II. External validation, model updating, and impact assessment. Heart 2012;98:691–8. [15] von der Maase H, Sengelov L, Roberts JT, et al. Long-term survival results of a randomized trial comparing gemcitabine plus cisplatin, with methotrexate, vinblastine, doxorubicin, plus cisplatin in

Financial disclosures: Timothy J. Daskivich certifies 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.

patients with bladder cancer. J Clin Oncol 2005;23:4602–8. [16] Vieweg J, Gschwend JE, Herr HW, Fair WR. The impact of primary stage on survival in patients with lymph node positive bladder cancer. J Urol 1999;161:72–6. [17] Amin MB, Greene FL, Edge SB, et al. The eighth edition AJCC cancer staging manual: continuing to build a bridge from a populationbased to a more “personalized” approach to cancer staging. CA Cancer J Clin 2017;67:93–9. [18] Konety BR, Joslyn SA, O’Donnell MA. Extent of pelvic lymphadenectomy and its impact on outcome in patients diagnosed with bladder cancer: analysis of data from the Surveillance Epidemiology and

Funding/Support and role of the sponsor: None.

End Results Program data base. J Urol 2003;169:946–50. [19] Herr HW, Bochner BH, Dalbagni G, Donat SM, Reuter VE, Bajorin DF. Impact of the numberof lymph nodes retrieved on outcome in patients

Appendix A. Supplementary data

with muscle invasive bladder cancer. J Urol 2002;167:1295–8. [20] Dhar NB, Klein EA, Reuther AM, Thalmann GN, Madersbacher S, Studer UE. Outcome after radical cystectomy with limited or

Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.euo.2018.12.012.

extended pelvic lymph node dissection. J Urol 2008;179:873–8, [discussion 878]. [21] Bruins HM, Veskimae E, Hernandez V, et al. The impact of the extent

References [1] Abdel-Latif M, Abol-Enein H, El-Baz M, Ghoneim MA. Nodal involvement in bladder cancer cases treated with radical cystectomy: incidence and prognosis. J Urol 2004;172:85–9. [2] Stein JP, Lieskovsky G, Cote R, et al. Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1054 patients. J Clin Oncol 2001;19:666–75. [3] Lerner SP, Skinner DG, Lieskovsky G, et al. The rationale for en bloc

of lymphadenectomy on oncologic outcomes in patients undergoing radical cystectomy for bladder cancer: a systematic review. Eur Urol 2014;66:1065–77. [22] Larcher A, Sun M, Schiffmann J, et al. Differential effect on survival of pelvic lymph node dissection at radical cystectomy for muscle invasive bladder cancer. Eur J Surg Oncol 2015;41:353–60. [23] Gschwend JE, Dahm P, Fair WR. Disease specific survival as endpoint of outcome for bladder cancer patients following radical cystectomy. Eur Urol 2002;41:440–8.

pelvic lymph node dissection for bladder cancer patients with nodal

[24] Svatek RS, Shariat SF, Novara G, et al. Discrepancy between clinical

metastases: long-term results. J Urol 1993;149:758–64, 64-5, dis-

and pathological stage: external validation of the impact on prog-

cussion. [4] Stein JP, Cai J, Groshen S, Skinner DG. Risk factors for patients with pelvic lymph node metastases following radical cystectomy with en

nosis in an international radical cystectomy cohort. BJU Int 2011;107:898–904. [25] Shariat SF, Palapattu GS, Karakiewicz PI, et al. Discrepancy between

bloc pelvic lymphadenectomy: concept of lymph node density. J

clinical and pathologic stage: impact on prognosis after radical

Urol 2003;170:35–41.

cystectomy. Eur Urol 2007;51:137–49, [discussion 149–51].

Please cite this article in press as: Patel DN, et al. Development and Validation of an Improved Pathological Nodal Staging System for Urothelial Carcinoma of the Bladder. Eur Urol Oncol (2019), https://doi.org/10.1016/j.euo.2018.12.012