Preoperative prediction of malignant involvement of resected ureters in patients undergoing radical cystectomy for bladder cancer

Authors


Correspondence: Tatsuo Gondo M.D., Department of Urology, Tokyo Medical University, 6-7-1 Nishi-shinjuku, Shinjuku-ku, Tokyo 160-0023, Japan. Email: eswl3500@yahoo.co.jp

Abstract

Objective

To investigate preoperative predictors of ureteral involvement of bladder malignancy and to develop a novel preoperative model for the prediction of ureteral involvement in bladder cancer patients undergoing radical cystectomy.

Methods

This study included 197 consecutive bladder cancer patients treated with radical cystectomy. The correlations of preoperative factors with ureteral involvement were analyzed by univariate analysis with Pearson's χ2-test and multivariate logistic regression analysis with a stepwise selection procedure.

Results

Positive ureteral involvement was observed in 38 (19.3%) patients. Tumor location (involvement of the vesical trigone), clinical T stage (≥ cT3) and the number of tumors (≥3), but not sex, tumor grade and histological features determined by transurethral resection of bladder tumor, tumor size, shape of tumor, concomitant presence of carcinoma in situ, preoperative intravesical therapy, number of transurethral resection of bladder tumor procedures or the presence of hydronephrosis were significantly associated with ureteral involvement in the univariate analysis. Multivariate logistic regression analysis confirmed that the aforementioned three significant factors identified in the univariate analysis were significant independent predictors of ureteral involvement. The probability of ureteral involvement estimated by a combination of these three parameters was well correlated with the real incidence (R = 0.904, P = 0.0021).

Conclusions

Tumor location (involvement of vesical trigone), clinical T stage (≥cT3) and the number of tumors (≥3) are significant independent preoperative predictors of ureteral involvement of malignancy in bladder cancer patients undergoing radical cystectomy. Our predictive model might be useful for preoperative prediction of ureteral tumor involvement.

Abbreviations & Acronyms
AUC

area under the curve

BCG

bacillus Calmette–Guérin

CI

confidence interval

CIS

carcinoma in situ

CT

computed tomography

FSA

frozen section analysis

MRI

magnetic resonance imaging

Neg.

negative

Pos.

positive

RC

radical cystectomy

TURBT

transurethral resection of bladder tumor

UC

urothelial carcinoma

Introduction

Intraoperative FSA of ureteral margins at the time of RC before urinary tract reconstruction has been accepted as a standard practice.[1] Tumor involvement of the distal ureter is considered an adverse prognostic factor,[2-6] and urologists typically attempt to resect the tumors with negative margins to effect clearance of all documented cancer, as this is assumed to provide better long-term local disease control.[1] However, intraoperative FSA cannot be always carried out in all institutions because of the absence of full-time pathologists. In such situations, it would be valuable to know the risk of ureteral involvement of malignancy pre- and intraoperatively in order to determine the length of ureter that should be preserved to carry out the appropriate urinary diversion, as some types of urinary diversion require the ureters to be as long as possible in order to obtain tensionless urinary anastomosis. Furthermore, even in an institution where FSA is available, it might not routinely be carried out because of recent cost–benefit considerations.[7] Therefore, preoperative prediction of the risk of ureteral tumor involvement could also be useful for deciding whether to carry out FSA. However, there have been few studies to date that evaluate preoperative factors for predicting ureteral tumor involvement. In addition, to our knowledge, no preoperative model has been developed for predicting ureteral tumor involvement in bladder cancer patients undergoing RC.

The present retrospective study was undertaken to investigate the possible preoperative predictors of ureteral involvement of malignancy and to develop a preoperative model for the prediction of ureteral involvement in bladder cancer patients treated with RC.

Methods

We analyzed 197 consecutive bladder cancer patients who underwent RC at Tokyo Medical University Hospital, Tokyo, Japan, between January 2000 and October 2009. The diagnosis of bladder cancer was histologically confirmed in each patient by TURBT. The indications for RC included muscle-invasive tumors or high-grade non-muscle-invasive tumors with no evidence of distant metastasis or BCG-resistant CIS.

The present study included 165 male and 32 female patients. The mean age of the patients was 69 ± 0.7 (SE) years (range 38–89 years). All patients had undergone TURBT before RC, and no patients had received neoadjuvant chemotherapy and/or radiotherapy. The preoperative clinicopathological features analyzed were sex, tumor grade and histological features at the time of TURBT, number of tumors, tumor size, shape of tumor(s), tumor location, clinical T stage, concomitant presence of CIS, pretreatment intravesical therapy, number of TURBT procedures and the presence of hydronephrosis. The clinical T stage of bladder tumors was determined according to the 2002 Union for International Cancer Control tumor–nodes–metastasis classification of bladder tumors. Hydronephrosis, defined as any degree of ureteral and/or renal collecting system dilatation, was diagnosed by ultrasonography, intravenous urography, CT and/or MRI before RC, as reported previously.[8] Tumor size was defined as the maximum tumor dimension and was estimated by preoperative clinical imaging studies, such as CT, MRI and ultrasonography, and/or at the time of TURBT. The number and shape of tumors were determined in the same way. The concomitant presence of CIS was shown by the pathological reports on the TURBT specimens.

FSA was carried out to evaluate tumor involvement of the distal ureteral margin in each patient. When the margin was positive, additional resection of the ureter was carried out until the ureteral margin became tumor-free. The ureteral involvement was deemed positive if at least one FSA of the distal ureteral margin was positive and/or tumors were present in the resected distal ureter on the post-cystectomy pathological reports.

The correlation of the aforementioned clinical variables with ureteral involvement was analyzed by univariate analysis with Pearson's χ2-test. As reported previously,[9] continuous pretreatment measurements (e.g. number of tumors, tumor size, etc.) were analyzed as dichotomous variables according to approximately “optimal” cut-points, determined as follows. The value best discriminating between the presence and absence of ureteral involvement (i.e. which recorded the most significant P-value on a Pearson's χ2-test) was determined by testing all possible cut-points. All such cut-points were then rounded to clinically relevant (i.e. convenient) values. Multivariate logistic regression analysis was then carried out to identify a significant set of independent predictors of the presence of ureteral involvement. The multivariate analysis included all of the possible predictive factors, and a stepwise selection procedure was used to obtain a multivariate model with maximum precision for the important variables. The predicted probability of ureteral involvement (P) was estimated with the formula P = 1 / (1 + exp [–k]). Logistic regression produces a score (k), where k = a + b1* × 1 + b2* × 2+ … is a linear combination of the predictors (×1, ×2 …) in the model. The model coefficients (a, b1, b2 …) were chosen to optimize the model's ability to predict the probability of ureteral involvement. Odds ratios were also calculated. The variables of the different groups were compared using the Mann–Whitney U-test or anova. The correlation of the possible predictive factors with ureteral involvement of malignancy was assessed using receiver operating characteristics analysis, as previously described;[10] this involves plotting the true positive rate (sensitivity) against the false-positive rate (1-specificity) using all different possible cut-off values to define positive and negative test results. The test's performance was measured by the AUC, with a perfect test having an area of 1.0. Results were presented as the mean value plus or minus the standard error. A P-value of less than 0.05 was considered statistically significant. All statistical analyses were carried out using StatView version 5.0 (sas, Cary, NC, USA) and stata version 11.0 (StataCorp, College Station, TX, USA).

Results

Patient characteristics are shown in Table 1. The clinical T stage was ≤T1, T2, T3 and T4 in 62, 80, 41 and 14 patients, respectively. A total of 38 patients (19.3%) had a history of intravesical therapy with BCG before RC. Preoperative imaging showed hydronephrosis in 38 patients (19.3%). The number of tumors, tumor size and shape of tumor(s) were ≤2 or ≥3, <3 cm or ≥3 cm, and papillary or non-papillary in 148 (75.1%) or 49 (24.9%), 63 (32.0%) or 134 (68.0%), and 56 (28.4%) or 141 (71.6%) patients, respectively. The total number of TURBT procedures before RC was 1 or ≥2 in 140 (71.1%) or 57 (28.9%) patients, respectively. A total of 68 (34.5%) of 197 patients had bladder cancer involving the vesical trigone. The concomitant presence of CIS was observed in 40 (20.3%) patients. Positive ureteral involvement was observed in 38 (19.3%) patients.

Table 1. Patient characteristics of 197 consecutive patients with bladder cancer who underwent radical cystectomy at Tokyo Medical University Hospital, Tokyo, Japan
VariableNo. (%) patients
Age (years) 
Mean (Median)68.5 (70)
Range38–89
Sex 
Male165 (83.6)
Female32 (16.4)
Clinical T stage 
≤T162 (32.8)
T280 (40.2)
T341 (20.1)
T414 (6.9)
Tumor grade on TURBT 
≤221 (11.1)
3176 (88.9)
Histological features upon TURBT 
Pure UC166 (84.3)
With non-UC component31 (15.7)
No. tumors 
≤2148 (75.1)
≥349 (24.9)
Tumor size 
<3 cm63 (32.0)
≥3 cm134 (68.0)
Shape of tumor(s) 
Papillary56 (28.4)
Non-papillary141 (71.6)
Involvement of vesical trigone 
Yes68 (34.5)
No129 (65.5)
Presence of CIS 
Yes40 (20.3)
No157 (79.7)
History of intravesical therapy 
Yes38 (19.3)
No159 (80.7)
Number of TURBT procedures 
1140 (71.1)
≥257 (28.9)
Hydronephrosis 
With38 (19.3)
Without159 (80.7)

In the univariate analysis, tumor location (involving vesical trigone), clinical T stage (≥cT3) and number of tumors (≥3) were significantly associated with ureteral involvement of malignancy (P = 0.0030, 0.0151 and 0.004, respectively; Table 2). In contrast, sex, tumor grade and histological features on TURBT, shape of tumors, concomitant presence of CIS, tumor size, pretreatment intravesical therapy, number of TURBT procedures and presence of hydronephrosis were not significantly correlated with ureteral involvement. Multivariate logistic regression analysis with a stepwise selection procedure showed that tumor location (involving the vesical trigone), clinical T stage (≥cT3) and number of tumor(s) (≥3) were significant independent predictors of positive ureteral involvement (Table 2). The probability of ureteral involvement was calculated with the logistic regression formula for tumor location, clinical T stage and number of tumors as parameters. The logistic regression equation is: P = 1 / (1 + exp [2.567 – 1.177 × number of tumors − 0.879 × tumor location − 1.160 × clinical T stage]). In this equation, the absence and presence of vesical trigone involvement equalled 0 and 1, clinical T stage <T3 and ≥cT3 equalled 0 and 1, and the number of tumors ≤2 and ≥3 equalled 0 and 1, respectively. These three parameters were used to construct a nomogram predicting the ureteral involvement of malignancy (Table 3). When these three precystectomy parameters were combined to predict ureteral tumor involvement, the predicted and observed risks of ureteral involvement were well correlated (P = 0.0021 and regression coefficient = 0.904). The AUC calculated for our predictive model shown in Table 3 was 0.7356, which was significantly better than those for the respective significant variables individually (AUC values for tumor location, clinical T stage, and number of tumors were 0.6285, 0.6042 and 0.6231 with P = 0.02, 0.0107 and 0.0005, respectively).

Table 2. Univariate and multivariate stepwise logistic regression analyses of precystectomy parameters for the detection of ureteral involvement
ParameterUnivariate P-valueMultivariate
CoefficientOdds ratio95% CIP-value
Sex (male vs female)>0.999    
Tumor grade on TURBT (≤2 vs 3)0.3760    
Histological features on TURBT (pure UC vs with non-UC component)0.4581    
No. tumors (≤2 vs ≥3)0.00301.1773.2451.431–7.3630.0048
Tumor size (<3 cm vs ≥3 cm)0.4453    
Shape of tumor(s) (papillary vs non-papillary)0.8428    
Tumor location: involvement of vesical trigone (yes vs no)0.00410.8792.4081.116–5.2000.0252
Clinical T stage (≤cT2 vs ≥cT3)0.01511.1603.1901.432–7.1050.0045
Concomitant presence of CIS (yes vs no)0.6535    
Pretreatment with intravesical therapy (yes vs no)0.6661    
No. TURBT procedures (1 vs ≥2)0.1623    
Hydronephrosis (yes vs no)0.1099    
Table 3. Predicted probabilities and observed rates of ureteral involvement of malignancy
LocationClinical TNumberPredicted probability (%)Observed rate (%)
  1. Clinical T, clinical T stage ≥cT3; Location, tumor location (involvement of vesical trigone); Number, ≥3 tumors present.
Neg.Neg.Neg.7.15.3 (4/75)
Pos.Neg.Neg.15.617.9 (5/28)
Neg.Pos.Neg.19.725.8 (8/31)
Neg.Neg.Pos.19.915.8 (3/19)
Pos.Pos.Neg.37.128.6 (4/14)
Pos.Neg.Pos.37.545.0 (9/20)
Neg.Pos.Pos.44.350.0 (2/4)
Pos.Pos.Pos.65.750.0 (3/6)

Discussion

Some forms of urinary diversion, such as the creation of some types of neobladder with tensionless ureterointestinal anastomosis, require longer ureters than do others. However, preserving longer ureters might increase the risk of residual cancer on the distal ureter in some patients. Therefore, it is important to predict the risk of ureteral involvement before RC and to know intraoperatively the length of ureter that can be preserved for appropriate urinary diversion. FSA has been considered routine practice and has been reported to accurately detect the malignant involvement of the distal ureter intraoperatively.[7] However, FSA cannot routinely be carried out in hospitals where no full-time pathologist is available. If we can predict the risk of ureteral involvement preoperatively, we can omit FSA for patients with lower risk of ureteral involvement at RC, thereby eliminating unnecessary medical expenses. Furthermore, the prediction of the risk of ureteral involvement might help us to identify patients who should have the maximum possible amount of ureter resected in order to prevent residual malignancy in the distal ureter even in hospitals with no full-time pathologists. To our knowledge, however, no predictive tool for ureteral tumor involvement at RC has been reported to date.

Johnson et al.[11] reported that FSA should be reserved only for patients with multifocal CIS or prostatic duct urothelial carcinoma,[11] and Schumacher et al.[12] also recommended FSA for patients with CIS of the bladder. Osman et al. carried out multivariate analysis to find risk factors for distal ureteral malignancy and concluded that male sex was a significantly associated predictor.[7] Therefore, the risk factors for ureteral involvement of malignancy remain controversial. CIS was not associated with ureteral involvement in the present study (P = 0.6535). In contrast, multifocality of the tumor (number of tumors, ≥3) was an independent predictor of ureteral involvement of malignancy in the multivariate analysis (P = 0.003) and was significantly correlated with the concomitant presence of CIS (P = 0.0228). The present results showing that CIS was not a significant predictive factor seem inconsistent compared with those of previous studies.[11, 12] Indeed, of the 38 cases of positive ureteral involvement, just nine (23.7%) had concomitant CIS that were preoperatively detected. In contrast, the post-cystectomy final pathological reports showed that concomitant CIS lesions were found in 28 of 38 (73.7%) patients with positive ureteral involvement (data not shown). Discrepancies between the accuracy of pre- and post-cystectomy detection of concomitant CIS might represent real-world clinical practices. As indicated by Nuhn et al., the concomitant CIS in cystectomy specimens does not add additional information for the management of bladder cancer.[13] Therefore, the patients who would be candidates for RC might not benefit from the preoperative detection of concomitant CIS. In the present study, the incidence of CIS was significantly higher in ≤cT2 patients than in ≥cT3 patients (23.9% in ≤cT2 patients vs 10.9% in ≥cT3 patients, respectively, P = 0.0484). This might partly be because we did not routinely carry out random biopsies of the surrounding area of main tumors and/or other abnormal mucosa, especially for clinically extravesical (≥cT3) disease, as detecting concomitant CIS in such patients with advanced tumor might not improve their clinical outcomes. Thus, the present study, which aimed at preoperative prediction of ureteral involvement of malignancy on RC, showed that multifocality rather than concomitant presence of CIS was a significant indicator of ureteral involvement of malignancy.

In the present study, the multivariate analysis showed that clinical T stage, tumor location and tumor number were significant independent predictors of ureteral involvement. We used these three risk factors to generate a preoperative model for predicting the ureteral involvement of malignancy. Patients with these risk factors might be candidates for intraoperative FSA, and as much length of their ureters as possible should be resected. In contrast, FSA could possibly be omitted in patients with none of the significant predictive factors, which would presumably eliminate some unnecessary medical expenses, and RC without examination of the distal ureter by FSA could be carried out in hospitals without full-time pathologists.

The present study might have had limitations because of its retrospective nature and the small sample size. Although the present predictive model might not be perfect, all of the patients in the present study were preoperatively investigated and postoperatively observed at a single institution. Therefore, all data were universally available in our databases, and the data quality might be better. As reported previously,[14-19] suboptimal predictive and prognostic models are reported and used based on the contention that in the absence of perfectly accurate models it is better to use a systematic rule with known accuracy and performance characteristics than to predict outcomes in a less accurate and possibly haphazard fashion. Therefore, our predictive model might qualify as a stepping stone towards a structured, standardized, methodological and objective risk stratification of ureteral involvement before RC, and could aid development of more accurate tools in the future. Despite some limitations, this model might help clinicians in clinical practice and seems preferable to making similar predictions with no tools at all. Further prospective studies will be expected to confirm our findings.

The present study showed that the tumor location, clinical T stage and number of tumors were significant independent predictors of ureteral involvement of malignancy at the time of RC in patients with bladder cancer, and that the combination of these three parameters might be useful for preoperative prediction of ureteral tumor involvement.

Conflict of interest

None declared.

Ancillary