Int J Clin Oncol DOI 10.1007/s10147-014-0693-3

ORIGINAL ARTICLE

External validation of existing nomograms predicting lymph node metastases in cystectomized patients Miroslav M. Stojadinovic • Rade I. Prelevic

Received: 6 January 2014 / Accepted: 28 March 2014 Ó Japan Society of Clinical Oncology 2014

Abstract Objectives Karakiewicz et al. and Green et al. created pre-cystectomy nomograms to predict lymph node involvement. The aim of the study was to externally validate these two nomograms in intermediate-volume institutions in Europe. Patients and methods Data from a Serbian single-centre cystectomy series comprising 183 patients with bladder cancer were used for the validation of two US nomograms, which were originally based on data from 726 and 201 patients, respectively. A multivariate regression model assessed the value of the clinical parameters integrated in the two nomograms. The expected predictive accuracy, calibration and clinical utility according to the nomograms were calculated. Results Comparison of our dataset with the previously published data shows differences in nearly all underlying risk variables. Overall, 109 (59.6 %) patients had lymph node metastases. The analysis demonstrated that hydronephrosis and status of lymph nodes on computed tomography have independent prognostic value. The performance of the nomograms deteriorated from the development set, and the predictive accuracies for the two models showed

M. M. Stojadinovic (&) Deparment of Urology, Clinic of Urology and Nephrology, Clinical Centre Kragujevac, Zmaj Jovina 30, 34 000 Kragujevac, Serbia e-mail: [email protected] R. I. Prelevic Military Medical Academy, Clinic of Urology, Belgrade, Serbia

moderate discriminatory ability (61.2–69.1 %). In the decision curve analysis, only the Green et al. model predicting lymph node positivity provided net benefit. Conclusions The Green et al. nomogram seems applicable to patients from Europe, despite varying risk factors in the validation dataset. Acceptance of such a tool into daily clinical management may lead to more appropriate decision-making. Nevertheless, further improvement and implementation of novel statistical models with enhanced predictive accuracy is needed. Keywords Bladder cancer  Radical cystectomy  Lymph node metastasis  Validation study Abbreviations AUC Area under the receiver operating characteristic curve BC Bladder cancer CI Confidential interval CIS Carcinoma in situ CT Computed tomography LN Lymph node ? LN Positive lymph node LR Logistic regression LVI Lymphovascular invasion MRI Magnetic resonance imaging NC Neoadjuvant chemotherapy NOC Non organ-confined NPV Negative predictive value OR Odds ratio PLND Pelvic lymph node dissection PPV Positive predictive value RC Radical cystectomy ROC Receiver operating characteristic curve TUR Transurethral resection

123

Int J Clin Oncol

Introduction Bladder cancer (BC) represents the most common urological cancer and the fifth most common tumour worldwide. Although many BCs are superficial in nature, muscleinvasive disease is present at about 30 % of patients at the time of diagnosis [1]. BC spreads from the bladder in a predictable stepwise manner to the lymph nodes (LNs) and then to visceral organs. The rate of positive LNs (LN?) in radical cystectomy (RC) specimens is between 5 and 10 % up to [40 % depending on tumour stage [2, 3]. Lymph node staging is a prerequisite for clinical decision-making and the oncological outcome after radical surgery. The extent of pelvic lymph node dissection (PLND), patient selection for bladder-preserving approach, follow-up regimens and the neoadjuvant and adjuvant chemotherapy (NC) that is associated with a potential benefit for this group of patients depends greatly on the extent of the disease [1]. Today, LN staging is mainly based on imaging techniques [1]. The major shortcoming of this standard practice is the lack of accuracy and the evident discrepancy between clinical and pathological staging. The assessment of metastases to LNs based simply on size is limited by the inability of both computed tomography (CT) and magnetic resonance imaging (MRI) to identify metastases in normalsized or minimally enlarged nodes. To overcome these deficiencies, several new techniques are currently being studied. Although early reports on new techniques are encouraging, it will take time to confirm the results before these new approaches can be incorporated into clinical routine [1]. Consequently, clinical staging has evolved from physician judgment alone to risk group stratification, prediction models, nomograms and decision tree models [4–9]. Numerous recent studies have confirmed that predictive models are more accurate than most informative single predictors [8]. A nomogram is a graphical illustration of a mathematical formula or algorithm that incorporates numerous predictors modelled as continuous variables to predict a particular endpoint based on traditional statistical methods [10]. In uro-oncology, nomograms have been established for prostate, non-muscle invasive BC [11, 12], invasive BC to provide individualized risk assessment [4–6], and to predict prognosis after RC [10, 12–15]. The value of nomograms is increasingly accepted in routine use [16, 17]. Nomograms are created using single highly predictable factors, but as risk factors vary among institutions, the performance of the nomograms can deteriorate from the development set to an external validation set. To date, some currently existing nomograms for prediction of

123

tumour stages pT3–4 or LN involvement have been externally validated, but some have not. In the study that externally validated two US nomograms from a German multicentre cystectomy series, both underestimated the real incidence of locally advanced tumours [18]. Based on these considerations, the objective of the study was to externally validate two existing nomograms and compare their accuracy for predicting LN involvement in patients with BC treated with radical surgery in intermediate-volume institutions in Europe.

Patient and methods Study population We validated two nomograms published in the USA using patients treated with radical surgery and bilateral lymphadenectomy for urothelial carcinoma of the bladder. The study was approved by the institutional review boards. The study comprised 183 consecutive patients from the Military Medical Academy, over an 11-year study period between 2002 and 2012. Detailed information on patient pre-cystectomy assessment and pathological review was collected. Patients with non-urothelial BC, salvage RC after failed radiotherapy or NC, distant metastatic disease, or incomplete data were excluded. All patients underwent surgery according to criteria consistent with current guideline recommendations [19]. Bilateral lymphadenectomy was performed at least in the obturator fossa and along the internal and external iliac artery. The study included only patients with at least 9 LNs removed, as suggested previously [20], or fewer if they were positive. On average, 14 LNs were removed per patient. In 31 (28.4 %) of all positive LN patients fewer than 9 LNs were removed. Tumour size and number were defined according to preoperative radiological examinations. Evaluation for the presence of hydronephrosis, if any, was performed in all patients, as described previously [21]. Data on transurethral resection (TUR) stage and grade and surgical specimens were classified according to the 7th edition TNM classification system and the 1973 WHO grading system, respectively [22, 23]. Lymphovascular invasion (LVI) in TUR of bladder tumour or biopsy specimens was defined as the unequivocal presence of tumour cells within an endothelium-lined space, with no underlying muscular walls [24]. The status of LNs on CT was categorized into three main groups. The first group represented patients with no LN involvement or the presence of LNs with a short axis of \10 mm. The second group contained those with a nodal enlargement C10–20 mm in the long axis and the third group comprised

Int J Clin Oncol

patients in whom LNs C20 mm in the long axis were found [25]. The Karakiewicz et al. [6] nomogram used a multiinstitutional cohort of 726 evaluable patients from three high-volume centres in the USA between 1984 and 2003. The primary outcomes of interest were probabilities of pT3–4 stage and presence of pN1–3 stage. Two separate nomograms were developed addressing the different outcomes. Internal validation was performed using 200 bootstrap resamples. The nomogram includes age, TUR stage and grade, preoperative carcinoma in situ (CIS) for prediction of pathological stage, and TUR stage (ordinal factor) and grade (nominal factor) for prediction of pathological stage N1–3. The multivariate pT3–4 and pN1–3 nomograms were 75.7 and 63.1 % accurate, respectively. The authors concluded that the multivariate nomograms are not perfect, but they do predict more accurately than TUR T stage alone. The Green et al. [5] nomogram used a cohort of 201 consecutive patients with BC who were treated with RC or partial cystectomy and bilateral PLND from one highvolume centre in the USA. All had clinically localized disease. Regression coefficients were used to develop a nomogram without internal validation. The nomogram includes TUR stage (ordinal factor), LVI (nominal factor) and abnormal imaging (nominal factor) for prediction of CpT3/Nany or pTany/N?. Abnormal imaging was defined by the presence of hydronephrosis and/or suggestion of non organ-confined (NOC) BC. A nomogram to predict CpT3/ Nany or pTany/N? based on all three variables was highly accurate (area under the curve 0.828). The authors concluded that NOC BC can be predicted with high accuracy by integrating standard clinicopathological factors with imaging information. As an assessment of the level of complexity of the Karakiewicz et al. and Green et al. nomograms, the calculation of the individual prognosis of each patient took \2 min in all cases using both models. Statistical analyses We used univariate and multivariate logistic regression (LR) to identify and quantify the potential and independent predictors for LN metastases with Backward–Wald stepwise. The results of regressions were expressed in odds ratios (ORs) with 95 % confidence intervals (CIs). The patients’ LN status according to the nomograms and our model was calculated and compared with their actual stage. For each model we calculated area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, calibration plots, Hosmer– Lemeshow statistic, and the Brier score (mean square error). The comparisons of AUC were performed using the

method proposed by DeLong et al. [26]. Clinical usefulness was assessed by using decision curve analyses [27]. These analyses estimate a ‘net benefit’ for prediction models by summing the benefits (true positives) and subtracting the harms (false positives). The assumption is made that the identification of advanced BC would lead to treatment with NC. Net benefit is plotted against threshold probabilities compared with ‘NC for all’ strategy and ‘NC for none’. The interpretation of a decision curve is that the model with the highest net benefit at a particular threshold probability should be chosen. We calculated and graphed net benefit in Excel using the recommended formula from trueand false-positive counts of patients [27]. All other analyses were performed using SPSS version 20.0 (SPSS Inc., Chicago, IL, USA). Statistical significance was set at P \ 0.05.

Results Table 1 shows the characteristics of the patients used for each model. Comparing our dataset with the previously published data shows some differences in nearly all underlying risk variables. In our cohort NOC BC dominated, often presenting with hydronephrosis and lymphovascular invasion, with common multifocality of tumours, frequently TUR T2 stage, resulting in a number of positive LNs (in about 60 % of patients). On the other hand, in the existing nomogram cohort organ-confined BC dominated, with positive LNs in less than one quarter of patients, but with frequently present concomitant CIS at TUR. Furthermore, our and Green et al.’s study excluded patients who received NC, while Karakiewics et al.’s included 38 (5.2 %) patients with NC. Finally, the Karakiewicz et al. study did not publish the characteristics of the underlying dataset regarding size, multifocality of tumours, presence of lymphovascular invasion, hydronephrosis or LN staging on CT (Table 1), and therefore a direct comparison cannot be made on the basis of the patients included. Overall, 109 (59.6 %) patients had LN metastases in our dataset. In a univariate analysis 6 risk factors displayed significant correlation with LN positivity (Table 2). During multivariate analysis, two patients maintained their prognostic significance (Table 2). The analysis demonstrated that hydronephrosis and status of LNs on CT have strong prognostic value of LN involvement (Table 2). AUC for all the models showed they had moderate discriminatory ability (61.2–69.1 %) (Fig. 1), and in pairwise comparison of ROC curves the difference between the areas of Green et al.’s and Karakiewicz et al.’s nomograms (7.9 %) were significant (P = 0.0004). Calibration evaluates the accuracy of the model in predicting risk in a group of patients. However, neither model showed satisfactory

123

Int J Clin Oncol Table 1 The characteristics of the patients used for each model Variables Study period a

Study dataset (n = 183)

Karakiewicz et al. cohort (n = 726)

Green et al. cohort (n = 201)

2002–2012

1984–2003

NP

63.4 (42–86)

64.6 (33.8 – 89.2)

72.9 (41 – 92)

Genderb

17/166 (9.3/90.7)

133/593 (18.3/81.7)

36/165 (17.9/82.1)

Size of tumoursc

31/65/87 (16.9/35.5/47.5)

NP

20/42/85 (13.6/28.6/57.8)

TUR graded

0/22/161

7/61/658

0/3/198

(0/12/88)

(1/8.4/90.6)

(0/1.5/98.5)

TUR T stage Multifocalityf

6/39/138 (3.3/21.3/75.4) 28/155 (15.3/84.7)

96/173/457 (13.2/23.8/63) NP

34/67/100 (17/33.3/49.8) 115/66 (63.5/36.5)*

TUR lympho-vascular invasionf

51/132 (27.9/72.1)

NP

147/46 (76.2/22.9)

Hydronephrosisf

77/106 (42.1/57.9)

NP

112/43 (72.3/27.7)

Concomitant CIS at TURg

4 (2.2)

294 (40.5)

46 (22.9)

Status LN on CTh

73/81/29 (39.9/44.3/15.8)

NP

NP

Pathological stagei

9/52/73/49 (4.9/28.4/39.9/26.8)

256/166/220/84 (32.2/22.9/30.3/11.6)

94/36/52/19 (46.8/17.9/25.9/9.5)

Pathological LN statusf

74/109 (40.4/59.6)

553/173 (76.2/23.8)

163/38 (81.1/18.9)

Age (years)

e

NP not published * Evaluable n = 181 a

Mean (range)

b

Female/male

c

B2 / [2, \5 / C5 cm 1/2/3

d e

Tis or Ta/T1/CT2

f

No/yes

g

Yes

h

LN size (\10 / 10–20 / [20 mm)

i

T0/CIS/Ta/T1, T2, T3, T4

b–i

The values represent number of patients (percent)

Table 2 Logistic regression analysis of predictors for lymph node-positive BC patients in our dataset

LN lymph node, CT computed tomography

Factor

Univariate analysis OR (95 % CI)

P value

TUR tumour grade

2.946 (1.167–7.433)

0.022

TUR tumour stage

2.033 (1.134–3.644)

0.017

Lymphovascular invasion

2.548 (1.316–4.935)

0.006

Size of tumours

1.393 (1.176–1.651)

0.000

Hydronephrosis

3.138 (2.006–4.909)

LN status on CT

2.460 (1.539–3.932)

calibration (Table 3). For the given models, the expected squared difference between patient status and predicted probability, the Brier score, was estimated (Table 3). It has an interpretation for a single patient. The Brier scores ranged from a low of 0.142 for the Green et al. model, the best predictive performance, to a high of 0.36.1 (noninformative) for the Karakiewicz et al. nomogram (Table 3).

123

Multivariable analysis OR (95 % CI)

P value

0.000

2.674 (1.675–4.269)

0.000

0.000

1.686 (1.042–2.729)

0.034

In the decision curve analysis (Fig. 2), Green et al.’s nomogram predicting LN positivity provided net benefit throughout the entire range of threshold probabilities, as compared with the strategy of treating all patients with NC or, alternatively, treating no-one. On the other hand, Karakiewicz et al.’s nomogram (red line) did not provide net benefit throughout the entire range of threshold probabilities, as compared with the strategy of treating all patients

Int J Clin Oncol

Fig. 1 ROC curve analysis of Karakiewicz et al.’s and Green et al.’s nomograms for predicted lymph node-positive bladder cancer

Fig. 2 Decision curve analysis of the effect of prediction models on the detection of lymph node positivity. Net benefit is compared with ‘neoadjuvant chemotherapy for all’ strategy and ‘neoadjuvant chemotherapy for none’

Table 3 Predictive performance of different methods Efficacy measure

Classification method Karakiewicz et al. nomogram

Green et al. nomogram

AUC (95 % CI)

61.2 (52.8–69.7)

69.1 (61.3–77)

Sensitivity (95 % CI)

79.8 (71–86.9)

61.5 (51.7–70.6)

Specificity (95 % CI)

41.9 (30.5–53.9)

75.7 (64.3–84.9)

PPV (95 % CI)

66.9 (58.1–74.9)

78.8 (68.6–86.9)

NPV (95 % CI)

58.5 (44.1–71.9)

57.1 (46.7–67.1)

Accuracy (95 % CI) HL test v2, P value

64.5 (57.1–71.4)

67.2 (59.9–73.9)

122.641, 0.000

77.520, 0.268

0.361

0.142

Brier score

2

HL Hosmer–Lemeshow, v chi-squared

with NC. Green et al.’s nomogram (blue line) was the optimal one leading to net benefit in various threshold probabilities of approximately 5–77 % (Fig. 2).

Discussion The aim of the present study was to externally validate the two previously developed US nomograms predicting LN metastases in a European cohort of patients who had undergone RC for BC. Considering that prediction of nodal metastases prior to cystectomy is fraught with even greater difficulties than predicting pathological stage, our findings showed that only Green et al.’s nomogram performed moderately well within this external cohort of patients. It has been shown to have predictive accuracy, as well as

clinical utility. Taken together, this study showed that the application of the Green et al. nomogram in different patient settings with different, most informative, single predictors for European patients is feasible. To date, numerous preoperative variables have been correlated with the pathological stage and nodal metastases for surgically treated BC patients and they can help to determine further treatment or to better inform patients about the options and potential consequences of therapies. The existing models included TUR parameters of stage and grade [4–6, 8, 9], LVI [4, 5, 7], hydronephrosis [5, 7–9], age [5, 6, 9], female gender [6, 28], CIS [6], histological variants [28], tumour size [8], tumour growth pattern [9], multiplicity of tumours [7, 9], palpable mass [7], number of intravesical treatments [7], NC [6], primary versus secondary RC [29], oncofetal markers [8], preoperative plasma soluble E-cadherin [30], expression of vascular endothelial growth factor-C [31], or gene expression model on primary tumour tissue [32]. In line with previous studies, several of those factors reached statistical significance in the univariate or multivariate analysis in our cohort. However, many of these parameters did not retain their independent predictive value. Nevertheless, we found that unilateral and bilateral hydronephrosis was a strong independent predictor of nodal metastases. These findings support those of previous investigators such as Stimson et al. [33], who reported that preoperative hydronephrosis was independently associated with extravesical and nodepositive disease at the time of cystectomy, directly affecting cancer-specific survival and predicting the side of nodal involvement. Although Green et al.’s nomogram included abnormal imaging and presence of hydronephrosis, the power of their variables is four times less than TUR T stage and less than half that of LVI, in contrast to our results

123

Int J Clin Oncol

where it represented a major variable. Furthermore, LVI was a highly predictable factor in univariate analysis, but was not an independent predictor in multivariate analysis. The explanation is the highly significant correlation between LVI and hydronephrosis and status LN on CT (data not presented). Also, we consider that our independent predictors reflect a more realistic clinical setting, because they include LN staging on CT, without which it is difficult to achieve a clinical staging. On the other hand, LN staging has not been included in the existing nomograms. Consequently, our predictors would make the findings more likely to be generally applicable to patients treated with RC at other centres. We are not the first researchers to perform a validation between different preoperative models predicting LN metastases in BC patients. May et al. [18] performed a validation study of the Karakiewicz et al. models in European patients using data from a German multicentre cystectomy series comprising 2,477 patients with urothelial BC, and demonstrated a notable decrease in model performance (the AUC was 54.5 % for pN? disease) that remained similar after refitting the model (64.7 % for pN? disease). Our study confirmed these findings and the clinical ineffectiveness of this model in decision curve analysis. As expected, Green et al.’s nomogram deteriorated in predictive performance from the development set to 69.1 % in the current dataset, but was similar to others (67–85 %) [4, 5, 7, 8, 32]. Validation on heterogeneous external data sets allows for evaluation of the generalizability of the risk prediction tool to a wider population than originally reported [34]. Both nomograms are based on multicentre data sets from mainly high-volume centres in the USA, but most RCs are performed in mid- to small-volume institutions. In the current study there were many differences when comparing the underlying data sets and the most important were frequency of pathological stage and outcome variable. Thus, in our study, [59 % of patients harboured LN metastasis because in our cohort locally advanced BC dominated (66.7 %), contrary to other studies where organ-confined BC dominated. Some of these might be explained by surgical strategy, sampling error due to incompleteness of the TUR, the extent of lymph node sampling, which depends on the surgeon’s and pathologist’s expertise and effort, accuracy of nodal status by performing extended LND, and great interobserver inconsistency in the histopathological evaluation of this tumour entity [3]. Therefore, the predictive performance of Green et al.’s nomogram in this dataset suggests on its validity even further. An important consideration in our study is the level of complexity, since all the models addressed incorporated easily obtainable variables and did not require predictors that are not routinely recorded, so the calculation is simple.

123

However, the existing models do not include established risk factors that have been shown to correlate significantly with the clinical outcome of patients, including LN staging by CT. The current study has several limitations worth noting. First, enrolled patients were retrospectively collected in a single tertiary centre with a relatively small patient cohort. Second, LN positivity is a useful intermediate endpoint whereas predicting disease outcome or response to therapy is more clinically significant. In our study, the extent of LND was at the surgeon’s discretion and extended PLND was not routinely performed. In addition, the number of LNs removed required for accurate nodal staging was dependent on the inherent differences among patients, the location of LNs from an area with a high likelihood of malignancy [20], the recently established clinical nodal staging score [3], and the pathological nodal staging score [35]. In addition, there were some differences in patient population regarding NC. In a recent study which addressed the effect of NC on the incidence of LN metastases in clinically node-negative (cN0) patients with muscle-invasive bladder cancer, it was shown that NC was independently associated with a lower incidence of occult LN metastases (OR 0.41) at the time of RC. Furthermore, patients with cT3–4 disease who recived NC had a median overall survival significantly longer than in the non-NC group [36]. Nevertheless, to our knowledge, we report the first external evaluation of the Green nomogram. The above findings indicate that the two proposed nomograms are limited by suboptimal accuracy. However, in assessing the status of LNs it seems that the clinical and pathological prognosticators appear to have limited prognostic ability. In addition, there are no absolute rules on how large the AUC must be for the predictive model to be useful [34]. Although the nomograms underestimated the real incidence of LN involvement, these results support the use of prediction models derived from a non-European patient cohort for predicting clinical outcomes of European patients.

Conclusion The present data are the first to show the validity of the Green et al. prognostic nomogram. The nomogram seems applicable to patients from Europe despite varying risk factors in validation dataset and decrease in model performance. Adoption of such a tool into daily clinical management may lead to more appropriate decision-making, thereby potentially improving survival in patients with BC. Nevertheless, further improvement and implementation of novel statistical models with enhanced predictive accuracy is needed.

Int J Clin Oncol Acknowledgments The authors were financially supported through a research Grant N0175014 of the Ministry of Science and Technological Development of Serbia. The authors thank the Ministry for this support. Conflict of interest

None declared.

References 1. Karl A, Carroll PR, Gschwend JE et al (2009) The impact of lymphadenectomy and lymph node metastasis on the outcomes of radical cystectomy for bladder cancer. Eur Urol 55:826–835 2. Youssef RF, Raj GV (2011) Lymphadenectomy in management of invasive bladder cancer. Int J Surg Oncol 2011:758189. doi:10. 1155/2011/758189 3. Shariat SF, Ehdaie B, Rink M et al (2012) Clinical nodal staging scores for bladder cancer: a proposal for preoperative risk assessment. Eur Urol 61:237–242 4. Xie HY, Zhu Y, Yao XD et al (2012) Development of a nomogram to predict non-organ-confined bladder urothelial cancer before radical cystectomy. Int Urol Nephrol 44:1711–1719 5. Green DA, Rink M, Hansen J et al (2012) Accurate preoperative prediction of non-organ-confined bladder urothelial carcinoma at cystectomy. BJU Int 111(3):404–411 6. Karakiewicz PI, Shariat SF, Palapattu GS et al (2006) Precystectomy nomogram for prediction of advanced bladder cancer stage. Eur Urol 50:1254–1262 7. Ahmadi H, Mitra AP, Abdelsayed GA et al (2012) Principal component analysis based pre-cystectomy model to predict pathological stage in patients with clinical organ-confined bladder cancer. BJU Int 111(4 Pt B):E167–E172 8. Margel D, Harel A, Yossepowitch O, Baniel J (2009) A novel algorithm to improve pathologic stage prediction of clinically organ-confined muscle-invasive bladder cancer. Cancer 115:1459–1464 9. Mitra AP, Skinner EC, Miranda G, Daneshmand S (2013) A precystectomy decision model to predict pathological upstaging and oncological outcomes in clinical stage T2 bladder cancer. BJU Int 111:240–248 10. Shariat SF, Karakiewicz PI, Suardi N et al (2008) Comparison of nomograms with other methods for predicting outcomes in prostate cancer: a critical analysis of the literature. Clin Cancer Res 15(14):4400–4407 11. Sylvester RJ, van der Meijden AP, Oosterlinck W et al (2006) Predicting recurrence and progression in individual patients with stage Ta T1 bladder cancer using EORTC risk tables: a combined analysis of 2596 patients from seven EORTC trials. Eur Urol 49:466–477 12. Margulis V, Lotan Y, Montorsi F et al (2008) Predicting survival after radical cystectomy for bladder cancer. BJU Int 102:15–22 13. El-Mekresh M, Akl A, Mosbah A, Abdel-Latif M et al (2009) Prediction of survival after radical cystectomy for invasive bladder carcinoma: risk group stratification, nomograms or artificial neural networks? J Urol 182:466–472 14. International Bladder Cancer Nomogram Consortium (2006) Postoperative nomogram predicting risk of recurrence after radical cystectomy for bladder cancer. J Clin Oncol 24:3967–3972 15. Karakiewicz PI, Shariat SF, Palapattu GS et al (2006) Nomogram for predicting disease recurrence after radical cystectomy for transitional cell carcinoma of the bladder. J Urol 176:1354–1361 16. Nuhn P, May M, Sun M et al (2012) External validation of postoperative nomograms for prediction of all-cause mortality, cancer-specific mortality, and recurrence in patients with urothelial carcinoma of the bladder. Eur Urol 61:58–64

17. Zaak D, Burger M, Otto W et al (2010) Predicting individual outcomes after radical cystectomy: an external validation of current nomograms. BJU Int 106:342–348 18. May M, Burger M, Brookman-May S et al (2011) Validation of pre-cystectomy nomograms for the prediction of locally advanced urothelial bladder cancer in a multicentre study: are we able to adequately predict locally advanced tumour stages before surgery? Der Urologe Ausg A 50:706–713 19. Stenzl A, Cowan NC, De Santis M et al (2009) The updated EAU guidelines on muscle-invasive and metastatic bladder cancer. Eur Urol 55:815–825 20. Rink M, Hansen J, Cha EK et al (2013) Outcomes and prognostic factors in patients with a single lymph node metastasis at time of radical cystectomy. BJU Int 111:74–84 21. Haleblian GE, Skinner EC, Dickinson MG et al (1998) Hydronephrosis as a prognostic indicator in bladder cancer patients. J Urol 160:2011–2014 22. Sobin LH, Gospodariwicz M, Wittekind C (2009) TNM classification of malignant tumors. UICC International Union Against Cancer, 7th edn. Wiley, London 23. Mostofi FK, Sobin LH, Torloni H (1973) Histological typing of urinary bladder tumours. World Health Organization, Geneva 24. 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 25. Eisenhauer EA, Therasse P, Bogaerts J et al (2009) New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 45:228–247 26. DeLong ER, DeLong DM, Clarke-Pearson DL (1988) Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 44:837–845 27. Vickers AJ, Elkin EB (2006) Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making 26(6):565–574 28. Turker P, Bostrom PJ, Wroclawski ML et al (2012) Upstaging of urothelial cancer at the time of radical cystectomy: factors associated with upstaging and its effect on outcome. BJU Int 110(6):804–811 29. McLaughlin S, Shephard J, Wallen E et al (2007) Comparison of the clinical and pathologic staging in patients undergoing radical cystectomy for bladder cancer. Int Braz J Urol 33:25–31 30. Matsumoto K, Shariat SF, Casella R et al (2003) Preoperative plasma soluble E-cadherin predicts metastases to lymph nodes and prognosis in patients undergoing radical cystectomy. J Urol 170:2248–2252 31. Suzuki K, Morita T, Tokue A (2005) Vascular endothelial growth factor-C (VEGF-C) expression predicts lymph node metastasis of transitional cell carcinoma of the bladder. Int J Urol 12:152–158 32. Smith SC, Baras AS, Dancik G et al (2011) A 20-gene model for molecular nodal staging of bladder cancer: development and prospective assessment. Lancet Oncol 12:137–143 33. Stimson CJ, Cookson MS, Barocas DA et al (2010) Preoperative hydronephrosis predicts extravesical and node positive disease in patients undergoing cystectomy for bladder cancer. J Urol 183:1732–1737 34. Taylor JM, Ankerst DP, Andridge RR (2008) Validation of biomarker-based risk prediction models. Clin Cancer Res 14(19):5977–5983 35. Shariat SF, Rink M, Ehdaie B et al (2013) Pathologic nodal staging score for bladder cancer: a decision tool for adjuvant therapy after radical cystectomy. Eur Urol 63:371–378 36. Mertens LS, Meijer RP, Meinhardt W et al (2013) Occult lymph node metastases in patients with carcinoma invading bladder muscle: incidence after neoadjuvant chemotherapy and cystectomy vs after cystectomy alone. BJU Int. doi:10.1111/bju.12447

123

External validation of existing nomograms predicting lymph node metastases in cystectomized patients.

Karakiewicz et al. and Green et al. created pre-cystectomy nomograms to predict lymph node involvement. The aim of the study was to externally validat...
364KB Sizes 0 Downloads 3 Views