World J Urol DOI 10.1007/s00345-015-1502-y

ORIGINAL ARTICLE

Risk stratification for locoregional recurrence after radical cystectomy for urothelial carcinoma of the bladder Vladimir Novotny · Michael Froehner · Matthias May · Chris Protzel · Katrin Hergenröther · Michael Rink · Felix K. Chun · Margit Fisch · Florian Roghmann · Rein‑Jüri Palisaar · Joachim Noldus · Michael Gierth · Hans‑Martin Fritsche · Maximilian Burger · Danijel Sikic · Bastian Keck · Bernd Wullich · Philipp Nuhn · Alexander Buchner · Christian G. Stief · Stefan Vallo · Georg Bartsch · Axel Haferkamp · Patrick J. Bastian · Oliver W. Hakenberg · Stefan Propping · Atiqullah Aziz  Received: 2 December 2014 / Accepted: 25 January 2015 © Springer-Verlag Berlin Heidelberg 2015

Abstract  Purpose  To externally validate the Christodouleas risk model incorporating pathological tumor stage, lymph node (LN) count and soft tissue surgical margin (STSM) and stratifying patients who develop locoregional recurrence (LR) after radical cystectomy (RC) for urothelial carcinoma of the bladder (UCB). In addition, we aimed to generate a new model including established clinicopathological features that were absent in the Christodouleas risk model. Methods  Prospectively assessed multicenter data from 565 patients undergoing RC for UCB in 2011 qualified for final analysis. For the purpose of external validation, risk group stratification according to Christodouleas was performed. Competing-risk models were calculated to compare the cumulative incidences of LR after RC. Results  After a median follow-up of 25 months (interquartile range 19–29), the LR-rate was 11.5 %. The

Christodouleas model showed a predictive accuracy of 83.2 % in our cohort. In multivariable competing-risk analysis, tumor stage ≥pT3 (HR 4.32, p < 0.001), positive STSM (HR 2.93, p = 0.005), lymphovascular invasion (HR 3.41, p < 0.001), the number of removed LNs <10 (HR 2.62, p < 0.001) and the administration of adjuvant chemotherapy (HR 0.40, p = 0.008) independently predicted the LR-rate. The resulting risk groups revealed significant differences in LR-rates after 24 months with 4.8 % for lowrisk patients, 14.7 % for intermediate-risk patients and 38.9 % for high-risk patients (p < 0.001 for all), with a predictive accuracy of 85.6 %, respectively. Conclusions  The Christodouleas risk model has been successfully externally validated in the present prospective series. However, this analysis finds that overall model performance may be improved by incorporating lymphovascular invasion. After external validation of the newly

V. Novotny · M. Froehner · S. Propping  Department of Urology, University Hospital “Carl Gustav Carus”, Dresden, Germany

M. Gierth · H.‑M. Fritsche · M. Burger  Department of Urology, Caritas St. Josef Medical Center, University of Regensburg, Regensburg, Germany

M. May  Department of Urology, St. Elisabeth Hospital, Straubing, Germany

D. Sikic · B. Keck · B. Wullich  Department of Urology, University Hospital Erlangen, Erlangen, Germany

C. Protzel · K. Hergenröther · O. W. Hakenberg  Department of Urology, University Medical Center Rostock, Rostock, Germany

P. Nuhn · A. Buchner · C. G. Stief  Department of Urology, Ludwig-Maximilians-University Munich, Munich, Germany

M. Rink · F. K. Chun · M. Fisch · A. Aziz (*)  Department of Urology, University Medical Center HamburgEppendorf, Martinistr. 52, 20246 Hamburg, Germany e-mail: [email protected]

S. Vallo · G. Bartsch · A. Haferkamp  Department of Urology, Goethe-University Frankfurt, Frankfurt am Main, Germany

F. Roghmann · R.‑J. Palisaar · J. Noldus  Department of Urology, Marienhospital Herne, Ruhr-University Bochum, Herne, Germany

P. J. Bastian  Department of Urology, Paracelsus Medical Center Golzheim, Düsseldorf, Germany

13



proposed risk model, it may be used to identify patients who benefit from an adjuvant therapy and suit for inclusion in clinical trials. Keywords  Bladder cancer · Radical cystectomy · Recurrence · Outcome

World J Urol

excluded due to histology other than UCB, evidence of distant metastases and salvage-RC. All patients with mixed histology had predominantly UCB combined with a variant histology in the specimen. In total, 565 patients with RC for curative intent were eligible for final analyses in the current study. Data assessment

Introduction Despite better knowledge of the disease’s history, advances in surgical techniques and improvement of systemic chemotherapy, data from the Southwest Oncology Group (SWOG) Intergroup trial report locoregional recurrence (LR) rates after radical cystectomy (RC) for urothelial carcinoma of the bladder (UCB) from 32 % in patients with tumor stage ≥pT3, 29 % in patients with nodal-positive disease and 68 % in patients displaying positive soft tissue surgical margin [1]. In addition, once disease recurs, roughly 83 % succumb to UCB within 2 years emphasizing the aggressive systemic potential of the disease resulting to an impaired cancer-specific survival after RC [2]. In the face of the lethal potential of LR after UCB, data on this issue and particularly risk stratification models for the development of LR after RC are scarce. Here, an accurate stratification of patients potentially experiencing LR could help in patient counseling regarding adjuvant therapy and in clinical trial design setting and analysis. Therefore, the attempt of the current study was to externally validate the recent risk model of Christodouleas et al. based on the variables pathological tumor stage, STSM and removed number of lymph nodes (LNs) [3, 4]. In addition, we hypothesized that established clinicopathological features for outcome prediction after RC for UCB that were not included in Christodouleas et al.’s model might have an impact on the occurrence of LR as well. Thus, we aimed to generate a model stratifying patients in risk groups for developing LR after RC in a large prospective European multicenter series.

Patient characteristics were evaluated and documented at admission for RC and comprised clinical variables such as age, gender, American Society of Anaesthesiologists (ASA)-score and treatment characteristics such as urinary diversion, administration of neoadjuvant chemotherapy (NACT) and adjuvant chemotherapy (ACT). ACT and NACT were only recorded when patients received at least one complete cycle of chemotherapy. Pathological variables included pathological tumor and nodal stage according to the 2009 TNM-classification [6], LN count, lymphovascular invasion (LVI) and STSM. A positive STSM was defined when tumor was present at inked areas of soft tissue on the RC specimen but not at the urethral and/or ureteral margins. LVI was defined as the unequivocal presence of tumor cells in an endothelium-lined space underlying muscular walls [7]. Follow‑up

Patients and methods

Follow-up was conducted prospectively according to current guidelines for at least 12 months following RC [8] including imaging such as ultrasound, chest radiography and radiographic evaluation of the upper urinary tract. Computed tomography (CT) or magnetic resonance imaging (MRI) was performed semiannually or earlier when clinically indicated. Bone scan was arranged in case of suspicious findings in CT and X-ray. Cause of death was determined by the treating physician, by chart review corroborated by death certificates, or by death certificates alone [9]. LR was defined as imaging-confirmed (via CT or MRI) or histological-confirmed tumor relapse in the pelvic soft tissues or LNs below the aortic bifurcation after RC, respectively.

Study population

Statistical analysis

Data for the current study were extracted from the Prospective Multicenter Radical Cystectomy Series 2011 (PROMETRICS 2011), which was an institutional review board—approved study that included prospectively assessed data from 18 European centers resulting in a database of 679 consecutive patients undergoing RC for muscle-invasive or high-risk UCB from January 1 to December 31, 2011 [5]. A total of 114 patients (16.8 %) were

Medians and interquartile ranges (IQR) were generated for continuously coded variables; frequencies and proportions were generated for categorical variables. The Mann–Whitney and Chi-square tests were used to assess differences in medians and proportions, respectively. For the purpose of external validation, risk group stratification for the development of LR was performed according to Christodouleas: patients with ≤pT2, ≥pT3 with negative

13

World J Urol

STSM and ≥10 LNs identified, ≥pT3 with positive STSM or <10 LNs identified, respectively, were stratified in lowrisk, intermediate-risk and high-risk, respectively [3]. LR was evaluated applying a cumulative incidence function in which death was classified as a competing event. Univariable and multivariable competing-risk regression models according to Fine and Gray were calculated to compare the cumulative incidences of LR or death between subgroups and to identify whether the predictive accuracy (PA) of our newly developed model could be improved with established clinicopathological features that were represented in the Christodouleas model [3, 4, 10]. In accordance with the latter, we included the variables age (<65 vs. ≥65 years), gender (male vs. female), pathological tumor stage (≤pT2 vs. ≥pT3), LN status (pN+ vs. pN0/pNX), LN count (removal of ≥10 LNs vs. <10 LNs) and STSM status (positive vs. negative), administration of NACT and ACT (administered vs. not administered) in our models. In addition, we integrated ASA-score (ASA 1–2 vs. ASA 3–4), urinary diversion (neobladder vs. other than neobladder) and LVI (presence vs. absence) to identify whether these variables could contribute to a higher PA compared to the original model by Christodouleas et al. [3]. In order to avoid a potential negative selection bias and to prove whether the variables of our multivariable competing-risk analysis still were robust, we conducted a subgroup analysis excluding all patients with ACT. Subsequently, a point value was calculated from the β-regression coefficients of all model variables exhibiting a significant influence [11]. This value was calculated based on the relationship between the β-regression coefficient and the smallest β-regression coefficient and is rounded to the nearest integer. Afterwards, three risk groups were generated using the factors that showed an independent effect in the multivariable model (low-risk, intermediate-risk and high-risk). The internal validity of the prediction model was confirmed by 1,000 bootstrap samples, in which the coefficients of the final regression model were estimated and tested in the original sample. The difference between the coefficients in the original sample and bootstrap samples as reflected by the slope index is the measure for the amount of “optimism.” Normally, slope values vary from 0 and 1. A slope value of 1 indicates no optimism. The slope index was used as a shrinkage method by multiplying coefficients with this slope index to correct for optimism. The PA of the Christodouleas model and our model was evaluated using the area under the receiver operating curves according to Harrell and compared using the Mantel–Haenszel test [12]. All analyses were performed using the R statistical package (v.2.12.2), the Statistical Package for Social Science 20.0 and STATA 12.0. All tests were two-sided with the statistical significance level set at p ≤ 0.05.

Results The clinicopathological characteristics of the entire cohort are summarized in Tables 1, 2 and 3. A total of 181 (32 %) patients died during follow-up with a median overall survival of 9 months (IQR 3–16). The median follow-up of patients alive was 25 months (IQR 19–29). A total of 11.5 % (n = 65) of the patients experienced LR with a median time from RC to diagnosis of LR of 8 months (IQR 2.5–14). Tumor stage ≥pT3 (p < 0.001), presence of LVI (p < 0.001) and removal of <10 LNs (p  = 0.015) were significantly associated with the development of LR after RC (Table 1). According to the risk stratification by Christodouleas, 55 % (n = 311) were classified low-risk, 31 % (n = 175) intermediate-risk and 14 % (n = 79) high-risk of developing LR in our entire cohort (p < 0.001, Table 1) [3, 4]. The application of the Christodouleas model in our cohort demonstrated 12- and 24-month estimates for LR of 2.5 and 5.8 % in the low-risk group (p < 0.001), 12.6 and 18.1 % in the intermediate-risk group (p < 0.001) and 31.8 and 45.4 % in the high-risk group (p < 0.001), respectively [3, 4]. The calculated risk of LR adjusted to competing risks using the low-risk group as referent showed significant differences in the intermediate-risk group [hazard ratio (HR) 4.42, 95 % confidence interval (CI) 2.35–8.30, p < 0.001] and the high-risk group (HR 13.33, 95 % CI 6.93–25.65, p < 0.001), respectively, with a PA of 0.832 (95 % CI 0.78– 0.88, p < 0.001; Fig. 1). In multivariable Fine and Gray regression models that were adjusted for competing risks, tumor stage ≥pT3 (HR 4.32, p < 0.001), positive STSM (HR 2.93, p  = 0.005), presence of LVI (HR 3.41, p < 0.001), number of removed LNs < 10 (HR 2.62, p < 0.001) and administration of ACT (HR 0.40, p = 0.008) significantly impacted on LR (Table 2). The regression coefficients of this model are also corrected for “optimism” by multiplying the coefficients of the original model by the shrinkage factor (slope index for shrinkage 0.96–0.99). After a subgroup analysis with exclusion of patients receiving ACT, we observed that the same factors had an independent impact on developing LR (Table 3). Our newly developed risk score comprised the significant variables of our multivariable Fine and Gray regression model providing a sum score total from 0 to 6: Tumor stage ≥pT3 was given 2 points and positive STSM, presence of LVI, number of removed LNs <10 and lack of administration of ACT was given 1 point each, respectively. Subsequently, patients with 0–2 points were stratified low-risk (55.6 %, n  = 314), with 3 points intermediate-risk (18.6 %, n  = 105) and with 4–6 points high-risk (25.8 %, n = 146), respectively. The calculated rates for developing LR after 12 and 24 months after RC according to our newly developed model in the study population were 1.5 and 4.8 % for low-risk patients (p < 0.001), 9.3 and 14.7 % for intermediate-risk patients (p < 0.001) and

13



World J Urol

Table 1  Descriptive characteristics Variables

All patients (n = 565)

Age (years)  <65 192 (34 %)  ≥65 373 (66 %) Gender  Male 456 (80.7 %)  Female 109 (19.3 %) ASA-score  1–2 290 (51.3 %)  3–4 275 (48.7 %) Histology  Pure UCB 528 (93.5 %)  Mixed histology 37 (6.5 %) Urinary diversion  Neobladder 160 (28.3 %)  Other than neobladder 405 (71.7 %) Pathologic tumor stage  ≤pT2 311 (55 %)  ≥pT3 254 (45 %) Pathologic nodal stage  pN0 389 (68.8 %)  pN+ 143 (25.3 %)  pNx 33 (5.8 %) Soft tissue surgical margin  Negative 521 (92.2 %)  Positive 44 (7.8 %) Lymphovascular invasion  Absence 366 (64.8 %)  Presence 199 (35.2 %) Lymph node count  ≥10 LNs removed 421 (74.5 %)  <10 LNs removed 144 (25.5 %) Administration of NACT  Administered 12 (2.1 %)  Not administered 533 (97.9 %) Administration of ACT  Administered 88 (15.6 %)  Not administered 477 (84.4 %) Risk group stratification according to [3]  Low-risk 311 (55 %)  Intermediate-risk 175 (31 %)  High-risk

79 (14 %)

Absence of LR (n = 500, 88.5 %)

Presence of LR (n = 65, 11.5 %)

169 (33.8 %) 331 (66.2 %)

23 (35.4 %) 42 (64.6 %)

0.783

403 (80.6 %) 97 (19.4 %)

53 (81.5 %) 12 (18.5 %)

1.000

257 (51.4 %) 243 (48.6 %)

33 (50.8 %) 32 (49.2 %)

1.000

467 (93.4 %) 33 (6.6 %)

61 (93.8 %) 4 (6.2 %)

1.000

144 (28.8 %) 356 (71.2 %)

16 (24.6 %) 49 (75.4 %)

0.559

296 (59.2 %) 204 (40.8 %)

15 (23.1 %) 50 (76.9 %)

<0.001

350 (70 %) 120 (24 %) 30 (6 %)

39 (60 %) 23 (35.4 %) 3 (4.6 %)

0.137

465 (93 %) 35 (7 %)

56 (86.2 %) 9 (13.8 %)

0.079

345 (69.0 %) 155 (31.0 %)

21 (32.3 %) 44 (67.7 %)

<0.001

381 (76.2 %) 119 (23.8 %)

40 (61.5 %) 25 (38.5 %)

0.015

10 (2 %) 490 (98 %)

2 (3.1 %) 63 (96.9 %)

0.638

75 (15 %) 425 (85 %)

13 (20 %) 52 (80 %)

0.280

296 (59.2 %) 148 (29.6 %)

15 (23.1 %) 27 (41.5 %)

56 (11.2 %)

23 (35.4 %)

p

<0.001

Bold values indicate statistical significance LR local recurrence, ASA American Society of Anaesthesiologists, UCB urothelial carcinoma of the bladder, LN lymph node, NACT neoadjuvant chemotherapy, ACT adjuvant chemotherapy

28.9 and 38.9 % for high-risk patients (p < 0.001), respectively (Fig. 2). Applying our model, the calculated risk of LR adjusted to competing risks using the low-risk group as referent revealed significant differences in the intermediate-risk

13

group (HR 4.33, 95 % CI 2.03–9.25, p < 0.001) and the highrisk group (HR 14.50, 95 % CI 7.55–27.83, p < 0.001), respectively, with a PA of 0.856 (95 % CI 0.81–0.90, p < 0.001) (Fig.  3). Our newly developed model showed a significant

World J Urol Table 2  Univariable and multivariable Fine and Gray competing-risk regression analysis addressing local recurrence after radical cystectomy for bladder cancer Variables

LR-rate (%)

Univariable analysis (adjusted with CR)

Multivariable analysis (adjusted with CR)

(First value vs. referent)

HR (95 % CI), p

HR (95 % CI), p

Age ≥ 65 years (referent: age < 65 years) Female gender (referent: male) ASA 3–4 (referent 1–2) Mixed histology (referent: pure UCB) Urinary diversion other than neobladder (referent: neobladder) Stage ≥pT3 (referent: ≤pT2) Stage pN+ (referent: pN0/pNx) Positive STSM (referent: negative) Presence of LVI (referent: absence) Number of LNs removed <10 (referent: ≥10 LNs) Administration of NACT (referent: not administered)

11.3 versus 12.0 11.0 versus 11.6 11.6 versus 11.4 10.8 versus 11.6 12.1 versus 10.0

1.05 (0.63–1.75), 0.849 1.07 (0.57–1.97), 0.841 1.35 (0.83–2.20), 0.222 1.04 (0.38–2.87), 0.934 1.56 (0.88–2.74), 0.124

0.81 (0.46–1.42), 0.454 1.34 (0.68–2.62), 0.396 1.04 (0.62–1.74), 0.887 0.70 (0.25–1.96), 0.500 0.88 (0.47–1.63), 0.681

19.7 versus 4.8 16.1 versus 10.0 20.5 versus 10.7 22.1 versus 5.7 17.4 versus 9.5 16.7 versus 11.4

6.35 (3.56–11.31), <0.001 2.78 (1.67–4.62), <0.001 3.52 (1.74–7.12), <0.001 6.10 (3.62–10.26), <0.001 2.21 (1.34–3.65), 0.002 1.25 (0.31–5.13), 0.751

4.32 (2.23–8.38), <0.001 1.69 (0.94–3.02), 0.077 2.93 (1.38–6.23), 0.005 3.41 (1.81–6.42), <0.001 2.62 (1.54–4.44), <0.001 0.84 (0.19–3.69), 0.815

Administration of ACT (referent: not administered)

14.8 versus 10.9

1.54 (0.84–2.84), 0.161

0.40 (0.20–0.79), 0.008

Bold values indicate statistical significance LR local recurrence, CR competing risks, HR hazard ratio, CI confidence interval, ASA American Society of Anaesthesiologists, UCB urothelial carcinoma of the bladder, STSM soft tissue surgical margin, LVI lymphovascular invasion, LN lymph node, NACT neoadjuvant chemotherapy, ACT adjuvant chemotherapy Table 3  Multivariable Fine and Gray competing-risk regression analysis addressing local recurrence after radical cystectomy for bladder cancer excluding patients with administration of adjuvant chemotherapy Variables

Multivariable analysis (adjusted with CR) HR (95 % CI), p

Age ≥ 65 years (referent age < 65 years) Female gender (referent male) ASA 3–4 (referent 1–2) Mixed histology (referent pure UCB) Urinary diversion other than neobladder (referent neobladder) Stage ≥pT3 (referent ≤pT2) Stage pN+ (referent pN0/pNx) Positive STSM (referent negative) Presence of LVI (referent absence) Number of LNs removed <10 (referent ≥10 LNs)

0.71 (0.37–1.37), 0.309 1.04 (0.50–2.18), 0.912 1.35 (0.75–2.43), 0.315 0.71 (0.22–2.32), 0.572 1.08 (0.52–2.25), 0.827 4.07 (1.99–8.35), <0.001 1.69 (0.87–3.29), 0.122 4.73 (2.05–10.91), <0.001 3.52 (1.76–7.04), <0.001 2.25 (1.26–4.03), 0.006

Administration of NACT (referent not administered)

0.36 (0.05–2.88), 0.335

Bold values indicate statistical significance LR local recurrence, CR competing risks, HR hazard ratio, CI confidence interval, ASA American Society of Anaesthesiologists, UCB urothelial carcinoma of the bladder, STSM soft tissue surgical margin, LVI lymphovascular invasion, LN lymph node, NACT neoadjuvant chemotherapy

gain of PA of 2.4 % (p < 0.001) compared to the risk model by Christodouleas (Fig. 3) [3, 4].

Discussion Patients experiencing LR after RC for UCB generally suffer from an adverse clinical course [1, 2, 13–16]. Pollack et al. [17] demonstrated that LR is found in 63 % of the cases

without distant metastases or more than 3 months prior to the detection of distant metastases, which was supported by two further studies [16, 18]. This observation underlines the hypothesis that LR abets the occurrence of distant metastases stressing out the aggressive behavior of UCB. Therefore, accurate risk estimation models for LR are compulsory for the treating physician in terms of potential recommendations of a further therapy. Such models could help integrate patients in the decision-making process, further may

13



World J Urol

Fig. 1  Probability of local recurrence after radical cystectomy for urothelial carcinoma of the bladder according to risk stratification by Christodouleas et al.’s original risk model

No. of Patients at

0

6

12

24

months

months

months

months

High-risk

77/0

40/15

29/19

11/22

Intermediate-risk

171/0

139/10

108/9

50/25

Low-risk

307/0

271/4

247/7

142/15

Risk / No. of Events

improve compliance and mitigate the probability of treatment regret [4]. Nevertheless, only one study group developed a score-based risk model for the prediction of LR after RC for UCB to date. The authors recently presented a risk model incorporating tumor stage, STSM and the number of removed LNs with a PA of 69 % [19]. Furthermore, they externally validated this risk model on an external cohort of the SWOG 8710 trial with a PA of 73 % [3]. We could previously confirm the capability of Christodouleas et al.’s model with a PA of 83.2 % in our prospective European cohort [4]. In addition, we demonstrated in the current study that the explanatory power of this model could be improved with our newly developed model including the variables LVI and ACT with a superior PA of 85.6 % within a more contemporary cohort of patients and a larger sample size, respectively. Although the PAs of both the models were similar at first glance, our newly developed risk model could identify higher HRs for high-risk patients compared to the Christodouleas model (25.8 vs. 14 %), thus underscoring the superior discriminative ability of our newly developed model in terms of risk stratification.

13

Considering the variables of our model, advanced pathological tumor stage ≥pT3 was the strongest predictor for LR in our multivariable competing-risk analyses confirming recent reports that LR is more frequently observed in patients with extravesicular tumor extension [13–16]. Furthermore, we could underscore the findings of Christodouleas that patients with removal of <10 LNs harbor an increased risk of LR [3]. Hence, removal of ≥10 LNs seems to be justified also in clinically N0 patients in order to eradicate undetected microscopic nodal disease that potentially hampers survival after RC [20, 21]. Interestingly, pathologic nodal stage did not have a significant impact on LR in our analysis which is in line with the findings of Christodouleas [3]. A possible explanation might be the high competing risk of early distant metastases in nodal-positive patients. In line with Christodouleas et al. and other reports [3, 22–24], positive STSM was an independent predictor for developing LR in our study as well. Consequently, more awareness during en bloc resection at surgery is warranted to avoid positive STSM and thus poor cancer-specific survival due to higher recurrence rates [22–24].

World J Urol Fig. 2  Probability of local recurrence after radical cystectomy for urothelial carcinoma of the bladder according to risk stratification by our newly developed risk model

No. of Patients at

0

6

12

24

months

months

months

months

High-risk

145/0

81/22

60/32

20/37

Intermediate-risk

104/0

91/6

73/9

40/13

Low-risk

310/0

278/1

251/4

143/12

Risk / No. of Events

Our hypothesis that the inclusion of further established clinicopathological features might improve the PA of Christodouleas et al. model could be confirmed after adding LVI and ACT in our risk model. Besides the fact that LVI is related to an aggressive tumor biology [25, 26], it has been ascribed a significant association with the occurrence of LR in 62 % of the cases of patients developing disease recurrence (n  = 86) according to data from Hassan et al. [16]. Most probably, the infiltration of the vascular and/or lymphatic tissue by tumor cells as well as occult micrometastases enhances tumor dissemination in the surrounding tissue and distant organs [26]. In contrast to Christodouleas, we included ACT in our analysis and found that the lack of ACT administration had a significant impact on the occurrence of LR in our study population. In order to verify whether the variables of our model still persisted without taking ACT into consideration, we performed a subgroup analysis excluding all patients that have been administered ACT. Here, we detected that the variables of our newly developed risk model still persisted which underscores the

robustness of our model. Furthermore, since patients with advanced tumor stage and positive nodal stage are more likely to receive ACT, our subgroup analysis could show a potential selection bias here. Thus, our risk model can be used for guidance in postoperative shared decision making in terms of direction of ACT and potentially offering highrisk patients a better survival given by ACT [27–29]. Some limitations apply. Despite the strengths of a prospective multicenter study, other important limitations include the lack of a central pathology/radiology review and the heterogeneity of clinical pathways, postoperative monitoring and the shortage of follow-up. The low rate of neoadjuvant chemotherapy in the present series does not allow drawing any conclusions in terms of outcome prediction, which limits our findings, even though this low rate represents general practice [30]. Due to the small number of patients in our prospective series, we were not able to split our cohort for the purpose of an external validation. Albeit the slopes of our model after internal validation with 1,000 bootstrap samples were promising, an external

13



World J Urol

Fig. 3  Calibration plot comparing Christodouleas et al.’s original risk model (dashed green line) and our newly developed risk model (solid red line) for the prediction of local recurrence after radical cystectomy for urothelial carcinoma of the bladder. PA predictive accuracy

validation of our risk model preferably in a prospective multicenter fashion is warranted to confirm its utility in clinical practice.

Ethical standard  This study has been approved by the appropriate ethics committee and has therefore been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. All persons gave their informed consent prior to their inclusion in the study

Conclusions References The Christodouleas risk model for the prediction of LR after RC for UCB could be externally validated with a good discrimination within our cohort. Our newly developed model allows to stratify patients for their individual risk of LR with a good PA and can serve as a tool for identifying eligible patients who can be administered ACT and included in clinical trials setting. However, an external validation of our risk model is warranted. Acknowledgments  We would like to thank the following colleagues for their valuable effort in the acquisition of data within our collaborative research group PROMETRICS 2011: Georg Bartsch, Christian Bolenz, Alexander Buchner, Sabine Brookman-May, Melanie Durschnabel, Jörg Ellinger, Galia Georgieva, Christian Gilfrich, Murat Gördük, Marc-Oliver Grimm, Boris Hadaschik, Florian Hartmann, Edwin Herrmann, Lothar Hertle, Markus Hohenfellner, Georg Janetschek, Nicole Kraischits, Annerose Krausse, Lukas Lusuardi, Thomas Martini, Roman Mayr, Maurice Stephan Michel, Rudolf Moritz, Stefan C. Müller, Sascha Pahernik, Armin Pycha, Jan Roigas, Christian Seitz, Shahrokh F. Shariat, Isabella Syring, Lutz Trojan, Florian Wagenlehner, Wolfgang Weidner, Manfred P. Wirth. Conflict of interest  The authors declare that they have no conflict of interest.

13

1. Herr HW, Faulkner JR, Grossman HB et al (2004) Surgical factors influence bladder cancer outcomes: a cooperative group report. J Clin Oncol 22:2781–2789 2. Rink M, Lee DJ, Kent M et al (2013) Predictors of cancer-specific mortality after disease recurrence following radical cystectomy. BJU Int 111(3 Pt B):E30–E36 3. Christodouleas JP, Baumann BC, He J et al (2014) Optimizing bladder cancer locoregional failure risk stratification after radical cystectomy using SWOG 8710. Cancer 120(8):1272–1280 4. Froehner M, Novotny V, Wirth MP, et al (2014) External validation of a model to predict locoregional failure after radical cystectomy. Cancer. doi:10.1002/cncr.28876. [Epub ahead of print] 5. Aziz A, May M, Burger M et al (2014) Prediction of 90-day mortality after radical cystectomy for bladder cancer in a prospective European multicenter cohort. Eur Urol 66(1):156–163 6. Greene FL, Gospodarowicz M, Wittekend C et al (2009) American Joint Committee on Cancer (AJCC) staging manual, 7th edn. Springer, Philadelphia 7. Quek ML, Stein JP, Nichols PW et al (2005) Prognostic significance of lymphovascular invasion of bladder cancer treated with radical cystectomy. J Urol 174:103–106 8. Stenzl A, Cowan NC, De Santis M et al (2011) Treatment of muscleinvasive and metastatic bladder cancer: update of the EAU guidelines. Eur Urol 59:1009–1018

World J Urol 9. Rink M, Fajkovic H, Cha EK et al (2012) Death certificates are valid for the determination of cause of death in patients with upper and lower tract urothelial carcinoma. Eur Urol 61:854–855 10. Fine JP, Gray RJ (1999) A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 94:496–504 11. May M, Bastian PJ, Brookman-May S et al (2013) Gender-specific differences in cancer-specific survival after radical cystectomy for patients with urothelial carcinoma of the urinary bladder in pathologic tumor stage T4a. Urol Oncol 31:1141–1147 12. Harrell FE Jr, Lee KL, Mark DB (1996) Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 15:361–387 13. Visser O, Nieuwenhuijzen JA, Horenblas S (2005) Local recurrence after cystectomy and survival of patients with bladder cancer: a population based study in greater Amsterdam. J Urol 174:97–102 14. Volkmer BG, Kuefer R, Bartsch GC Jr et al (2009) Oncological followup after radical cystectomy for bladder cancer—is there any benefit? J Urol 181:1587–1593 15. Honma I, Masumori N, Sato E et al (2004) Local recurrence after radical cystectomy for invasive bladder cancer: an analysis of predictive factors. Urology 64(4):744–748 16. Hassan JM, Cookson MS, Smith JA Jr et al (2006) Patterns of initial transitional cell recurrence in patients after cystectomy. J Urol 175:2054–2057 17. Pollack A, Zagars GK, Cole CJ et al (1995) The relationship of local control to distant metastasis in muscle invasive bladder cancer. J Urol 154:2059–2063 18. Ide H, Kikuchi E, Miyajima A et al (2008) The predictors of local recurrence after radical cystectomy in patients with invasive bladder cancer. Jpn J Clin Oncol 38:360–364 19. Baumann BC, Guzzo TJ, He J et al (2013) A novel risk stratification to predict local-regional failures in urothelial carcinoma of the bladder after radical cystectomy. Int J Radiat Oncol Biol Phys 85:81–88 20. Skinner EC, Stein JP, Skinner DG (2007) Surgical benchmarks for the treatment of invasive bladder cancer. Urol Oncol 25:66–71

21. May M, Herrmann E, Bolenz C et al (2011) Association between the number of dissected lymph nodes during pelvic lymphadenectomy and cancer-specific survival in patients with lymph node-negative urothelial carcinoma of the bladder undergoing radical cystectomy. Ann Surg Oncol 18(7):2018–2025 22. Herr H, Lee C, Chang S et al (2004) Standardization of radical cystectomy and pelvic lymph node dissection for bladder cancer: a collaborative group report. J Urol 171:1823–1828 23. Novara G, Svatek RS, Karakiewicz PI et al (2010) Soft tissue surgical margin status is a powerful predictor of outcomes after radical cystectomy: a multicenter study of more than 4,400 patients. J Urol 183(6):2165–2170 24. Xylinas E, Rink M, Novara G et al (2013) Predictors of survival in patients with soft tissue surgical margin involvement at radical cystectomy. Ann Surg Oncol 20(3):1027–1034 25. Lotan Y, Gupta A, Shariat SF et al (2005) Lymphovascular invasion is independently associated with overall survival, causespecific survival, and local and distant recurrence in patients with negative lymph nodes at radical cystectomy. J Clin Oncol 23:6533–6539 26. Shariat SF, Svatek RS, Tilki D et al (2010) International validation of the prognostic value of lymphovascular invasion in patients treated with radical cystectomy. BJU Int 105(10):1402–1412 27. Nakagawa T, Hara T, Kawahara T et al (2013) Prognostic risk stratification of patients with urothelial carcinoma of the bladder with recurrence after radical cystectomy. J Urol 189:1275–1281 28. Meeks JJ, Bellmunt J, Bochner BH et al (2012) A systematic review of neoadjuvant and adjuvant chemotherapy for muscleinvasive bladder cancer. Eur Urol 62(3):523–533 29. Booth CM, Siemens DR, Li G et al (2014) Perioperative chemotherapy for muscle-invasive bladder cancer: a population-based outcomes study. Cancer 120(11):1630–1638 30. Burger M, Mulders P, Witjes W (2012) Use of neoadjuvant chemotherapy for muscle-invasive bladder cancer is low among major European centres: results of a feasibility questionnaire. Eur Urol 61(5):1070–1071

13