Skip to main content

Prognostic nomogram for bladder cancer with brain metastases: a National Cancer Database analysis



This study aimed to establish and validate a nomogram for predicting brain metastasis in patients with bladder cancer (BCa) and assess various treatment modalities using a primary cohort comprising 234 patients with clinicopathologically-confirmed BCa from 2004 to 2015 in the National Cancer Database.


Machine learning method and Cox model were used for nomogram construction. For BCa patients with brain metastasis, surgery of the primary site, chemotherapy, radiation therapy, palliative care, brain confinement of metastatic sites, and the Charlson/Deyo Score were predictive features identified for building the nomogram.


For the original 169 patients considered in the model, the areas under the receiver operating characteristic curve (AUC) were 0.823 (95% CI 0.758–0.889, P < 0.001) and 0.854 (95% CI 0.785–0.924, P < 0.001) for 0.5- and 1-year overall survival respectively. In the validation cohort, the nomogram displayed similar AUCs of 0.838 (95% CI 0.738–0.937, P < 0.001) and 0.809 (95% CI 0.680–0.939, P < 0.001), respectively. The high and low risk groups had median survivals of 1.91 and 5.09 months for the training cohort and 1.68 and 8.05 months for the validation set, respectively (both P < 0.0001).


Our prognostic nomogram provides a useful tool for overall survival prediction as well as assessing the risk and optimal treatment for BCa patients with brain metastasis.


As the top ranked malignancy of the urinary system, bladder cancer (BCa) incidence data in the US shows an estimated 79,030 (8th among all sites) new cases and 16,870 (8th among all sites) deaths in 2017 [1]. Unfortunately, 10–15% of BCa patients already have metastasis at initial diagnosis and 15–30% high-grade BCa will eventually progress to advanced disease and lead to poor prognosis [2].

Despite an initial response to chemotherapy, patients with non-organ-confined disease fail to attain satisfactory survival [3]. Since no optimally effective chemotherapeutic modality has been found, patients with NOC disease can barely survive for more than 3–6 months [4]. According to a previous population-based study of the SEER database, only 4.1% (76/1862) BCa patients had brain metastases in a cohort of 1862 patients with metastatic sites [5]. Given the rarity of brain metastases at presentation, currently, there is no randomized phase II or III clinical trials exploring outcomes of this group. The survival prognosis of this subgroup calls for significant melioration when compared to those with cerebral metastasis from other malignancies [6].

Some reports have claimed that stereotactic radiosurgery and whole-brain irradiation can be a useful alternative approach for patients with brain metastasis in certain malignancies [7, 8]; however, suitable treatment for BCa patients with brain metastases remains unclear. A study conducted in 2002 in Cleveland retrospectively analyzed 16 BCa patients with brain metastases and suggested more aggressive treatment rather than radiation therapy alone [9]. However, the cohort was too small to extract robust clinical traits. In 2010, Fokas et al. found no significant difference in survival after comparing radiotherapy alone with radiotherapy plus surgery in 62 patients with brain metastases from BCa [10]. Therefore, reconsideration of current medical strategies is indispensable, since the role of surgeries of the primary tumor or radiotherapy of brain lesions in the treatment of metastatic BCa is still obscure.

Although previous studies have identified several prognostic factors of poor outcome in advanced BCa, such as the presence of visceral metastasis, anemia, and C-reactive protein (CRP) [3, 11, 12], it remains unknown whether they could be applied to the clinical assessment. Currently, prognostic nomograms are widely applied as prognostic devices in oncologic medicine. With the ability to incorporate clinical characteristics to generate individual probabilities of clinical events, nomograms can aid clinical decisions and facilitate our drive towards personalized medicine [13]. The purpose of our study was to create a nomogram predicting overall survival (OS) of BCa patients with brain metastasis and evaluate suitable therapeutic modalities for this cohort.

Materials and methods

Study population

The National Cancer Database (NCDB) was queried for patients initially diagnosed with histological confirmed BCa (topographical code C67, International Classification of Diseases for Oncology, 3rd edition) between 2004 and 2015. Patients with brain metastatic disease at the time of presentation were selected for the analysis.

Baseline medical traits (including age; sex, race; pathological grade; tumor histology, lymph node vascular invasion, and clinical stage [TNM] of the American Joint Committee on Cancer; surgical statuses of the primary and metastatic sites; chemotherapy; radiation therapy; and palliative care) were derived from medical records (Table 1). Other inclusion criteria were as follows: age > 18 years; BCa as the primary cancer diagnosis; brain metastasis; other distant metastatic sites including bone, liver, lung, and distant lymph node involvement; active follow-up; and patients with > 30 days of survival. Patients without sufficient information about distant metastatic sites or survival data were excluded. No detailed data were available regarding the specific types of chemotherapy or hormonal therapy or palliative care agents. Finally, we included 234 patients with the above-mentioned criteria. We used a computer-generated random seed to assign 169 of these patients to the training set, and 65 patients to the internal testing set for subsequent analysis. Work of flow is displayed in Fig. 1.

Table 1 Description of clinical characteristics and their values
Fig. 1

Flowchart of the analysis

Compliance with ethical standards

The NCDB is a hospital-based registry of cases treated at American College of Surgeons Commission on Cancer accredited cancer programs. Extraction of data from the NCDB did not require extra informed consent. All the data were downloaded at the Sidney Kimmel Cancer Center of Thomas Jefferson University.

Statistical analysis

For feature selection, we used the least absolute shrinkage and selection operator (LASSO) method, which is a machine learning method suitable for the reduction of high-dimensional data [14]. Eighteen variables were used to select the most useful predictive features from the primary data set. The LASSO regression model analysis was performed using the glmnet package of R (R Foundation for Statistical Computing, Vienna, Austria).

Univariate and multivariate Cox regression analysis were performed to explore the independent prognostic factors via the survival package of R. The Schoenfeld residuals method was applied to test the proportional hazards assumption for the Cox regression model fit. Each regression coefficient of selected variables was converted to a 0 to 100 scale proportion according to its contribution. These points were added across enrolled variables to generate total points, which were then transformed to predicted probabilities. For clinical use, the predictive performance of the nomogram was measured via time-dependent receiver operating characteristic (ROC) analysis with area under the curve (AUC) values. Calibration was employed with bootstrapping to decrease the bias of over-fitting. The x-axis represented the prediction calculated using the nomogram, and the y-axis the actual risk odds for the individual. The 45-degree line represented an ideal performance of the nomogram, in which the predicted outcome perfectly corresponded with the actual outcome. The model that incorporated the above independent predictors was developed and presented as the nomogram. Nomogram and calibration plots were obtained using the rms and nomogramEx packages of R.

Survival analysis was performed using the Kaplan–Meier method to probe the correlation between variables and OS, and the log-rank test was performed to compare survival variance in different groups. Decision curve analysis were performed to compare with the current AJCC TNM staging system. All statistical tests and analyses were performed in R software version 3.5.1. Statistical significance was set at < 0.05.


Clinicopathologic characteristics

During the study procedure, 268 consecutive BCa patients with brain-involvement were identified from the NCDB. Of these, 234 patients with brain metastasis in accordance with the inclusion criteria were enrolled, and 169 and 65 patients were randomly divided into the training and internal validation cohorts, respectively. The clinicopathologic characteristics and baseline data in the primary and validation cohorts are provided in Table 2. The median follow-up time was 3.38 (range: 1.08–61.21) months.

Table 2 Baseline characteristics and distribution of risk stratification of patients in the training and validation cohorts

Feature selection via LASSO

LASSO with tenfold cross-validation generated 7 variables out of 18 features: Grade, Surgery_Primary, Chemotherapy, Radiation_Therapy, Paliative_Care, Brain_Confined_Met, and CDCC_Score (Fig. 2a, b). The results of the univariate and multivariate Cox regression of the primary cohort are recorded in Table 3. Surgery_Primary, Chemotherapy, Radiation_Therapy, Paliative_Care, Brain_Confined_Met, and CDCC_Score were chosen for further analysis (apart from Radiation_Therapy, all other variables were independent prognostic factors in the LASSO Cox model; the reason for including this variable will be explained in “Discussion”). P values for Schoenfeld residuals method were all > 0.05 which fitted the proportional hazards assumption for the Cox model (Additional file 1: Fig. S1).

Fig. 2

Clinical trait selection via the least absolute shrinkage and selection operator (LASSO) cox regression model. a Tenfold cross-validated error (first vertical line equals the minimum error (lambda = 0.066), whereas the second vertical line shows the cross-validated error within 1 standard error of the minimum). b The profile of coefficients in the model at varying levels of penalization plotted against the log (lambda) sequence

Table 3 Univariate and multivariate Cox regression analysis of BCa patients based on clinicopathological characteristics derived from NCDB data in the training cohort

Development and validation of the prognostic nomogram

The prognostic nomogram that integrated all selected factors for OS in the primary cohort is shown in Fig. 3. We then derived a formula to calculate the risk score for odds of death for every patient based on their individual status of the selected variables above. To take one patient for example (purple track in Fig. 3), basing on the selected features, the total points adds up to 323 and thus the corresponding 0.5- and 1-year death probabilities are 0.647 and 0.9 respectively. The equation of each variable and computational formula is presented in Table 4. We further stratified those patients with an average or higher-than-average risk score into the high risk group and those with lower-than-average risk score into the low risk group (Table 1). In terms of 0.5- and 1-year OS of the training set, our six-clinical variable-based classifier demonstrated favorable discrimination with AUC values of 0.823 (95% confidence interval [CI] 0.758–0.889, P < 0.001) and 0.854 (95% CI 0.785–0.924, P < 0.001), respectively (Fig. 4a). The internal-bootstrapped calibration plot for the probability of survival at 0.5 or 1 year after surgery showed an optimal agreement between prediction by nomogram and actual observation (Fig. 4b, c). In the validation cohort, the nomogram displayed similar AUC values of 0.838 (95% CI 0.738–0.937, P < 0.001) and 0.809 (95% CI 0.680–0.939, P < 0.001) for the estimation of survival (Fig. 4d). There was also a well-behaved calibration curve for the survival estimation (Fig. 4e, f).

Fig. 3

Nomogram to estimate the risk and predict the survival of BCa patients with brain metastasis. Bars in blue display the distribution of patients in the training cohort. To calculate the total points of a specific patient, locate the value of each variable on the top point axis, add the points from all of the variables, and draw a vertical line from the total point axis to determine the 0.5 and 1 year death probabilities at the lower line of the nomogram. Purple track provided an example for the calculation of total-points-to-outcome (*P < 0.05, **P < 0.01, ***P < 0.001)

Table 4 The risk point of each variable and computational formula of OS
Fig. 4

Time-dependent ROC curves comparing the prognostic accuracy of nomogram in BCa patients with metastatic brain lesions in the training cohort (a) and validation set (d). Validity of the predictive performance of the nomogram in the training cohort (b, c) and validation set (e, f). Nomogram-predicted probability of overall survival is plotted on the x-axis; actual overall survival is plotted on the y-axis. ROC receiver operator characteristic, AUC area under the curve

Kaplan–Meier survival analysis yielded a significant difference in survival between the training cohort and validation set. The median survival of the High and Low risk groups were 1.91 and 5.09 months in the training cohort (Fig. 5a) and 1.68 and 8.05 months in the validation set (Fig. 5b), respectively (both P < 0.0001).

Fig. 5

Kaplan–Meier survival analysis for all patients according to our classifier stratified by clinicopathological risk factors. Survival curves show the overall survival of high risk (blue) and low risk (green) groups between the training cohort (a) and validation set (b). Confidence interval band and risk table are also added

Moreover, the decision curve analysis demonstrated that when the threshold probability was greater than 0.4, the nomogram presented more net benefit than TNM system in terms of OS prediction (Fig. 6).

Fig. 6

Decision curve analysis for OS. Black line: all victims dead. Gray line: none victims dead. Black dashed line: model of the nomogram. Red dashed line: staging system of TNM


In this study, we developed and validated a novel prognostic tool based on six clinical variables to improve the prediction of OS for patients with confirmed BCa with metastatic brain lesions. Our results showed that this tool can well categorize patients into high-risk and low-risk groups with large differences in OS.

Generally, in our research, prognostic factors are closely related to the choices of treatment modalities, as well as the comorbidities and metastatic conditions of the patient. Known as the best method of determining comorbidity conditions, higher Charlson/Deyo Score (CDCC_Score) is reported as a poor prognostic factor for overall mortality and cancer-specific mortality in metastatic BCa [15], consistent with our findings. In a previous study, multisite metastasis was found to be able to independently predict worse OS compared with single metastatic sites in BCa patients [5]. Our results are in line with the study above since brain-confined metastatic disease was related with better survival.

Treatment for the metastatic group is not beyond dispute. Our study included 4 treatment variables: Surgery_Primary, Surgery_Met, Chemotherapy, Paliative_Care and Radiation_Therapy.

A previous study indicated that surgical management of the primary BCa might contribute to long-term disease-free survival in selected patients [16]. Chen et al. also suggested that surgical management of the primary BCa might improve OS outcomes among patients [17]. Our study uncovered that brain metastatic BCa patients can still benefit from surgical operation of the primary site via minimal invasive surgery or otherwise. As for surgeries of the metastatic site, limited conclusions could be drawn for the lack of unanimous reporting elements and resection of solitary lung metastasis may result in OS improvement when integrated with chemotherapy [18]. As shown in the nomogram, though resistance may easily show up, chemotherapy still exerts maximal survival benefit for brain-metastatic BCa patients, which correlates to the first-line treatment of the European Association of Urology guidelines [19]. Given that more than half of patients with metastatic urothelial cancer are unfit for cisplatin-based chemotherapy, the choice of chemotherapy combination will have to depend on the health condition of patients. Consensus from an international survey among urologic experts was reached to define patients unfit for cisplatin-based chemotherapy, which was as follows: performance score > 1, glomerular filtration rate ≤ 60 ml/min/1.73 m2, grade 2 audiometric loss and peripheral neuropathy, and New York Heart Association class III heart failure [20, 21]. Palliative care is defined as any procedures to alleviate symptoms distinguishable from the same modality used for curative intent, which may include surgery, radiation therapy, systemic therapy, and/or other pain management drugs. Advanced BCa can be associated with problems like ureteral obstruction, persistent bleeding, pain, and/or voiding complaints; palliative care may prolong life expectancy in these patients [22]. The variable Radiation_Therapy was fitted into the analysis because although the P value 0.07 slightly surpassed 0.05 in the multivariate Cox model, it was 0.04 in the univariate analysis. Moreover, for brain metastatic cancer, conventional fractionated whole brain radiotherapy is still frequently used as a standard therapy [23]; thus, we included it in the prognostic nomogram for clinical consideration.

To the best of our knowledge, this is the largest cohort study exploring the prognostic significance of BCa with brain metastasis and the effect of various treatments on patients’ prognoses; however, several limitations are still noteworthy. For example, information regarding metastasectomy for specific metastatic sites was incomplete. In addition, there was a lack of details and sequences concerning chemotherapy, endocrine therapy, immunological treatment, and radiation therapy. As a retrospective study population from different medical facilities, some baseline characteristics may be non-uniform and external validation cohorts are needed to confirm the predictive accuracy of the nomogram.


By combining six clinical factors of brain-metastatic BCa patients, we constructed a prognostic nomogram. The model provides an optimal estimation of OS and reference for suitable treatments in BCa patients with brain metastasis.

Availability of data and materials

Original data for BCa patients with brain metastasis can be found at



bladder cancer


area under the receiver operating characteristic curve


C-reactive protein


overall survival


National Cancer Database


least absolute shrinkage and selection operator


confidence interval


Charlson/Deyo Score


  1. 1.

    Siegel RL, Miller KD, Jemal A. Cancer statistics, 2017. CA Cancer J Clin. 2017;67:7–30.

    Article  Google Scholar 

  2. 2.

    Stein JP, et al. Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054 patients. J Clin Oncol Off J Am Soc Clin Oncol. 2001;19:666–75.

    CAS  Article  Google Scholar 

  3. 3.

    von der Maase H, et al. Long-term survival results of a randomized trial comparing gemcitabine plus cisplatin, with methotrexate, vinblastine, doxorubicin, plus cisplatin in patients with bladder cancer. J Clin Oncol Off J Am Soc Clin Oncol. 2005;23:4602–8.

    Article  Google Scholar 

  4. 4.

    Sternberg CN, Vogelzang NJ. Gemcitabine, paclitaxel, pemetrexed and other newer agents in urothelial and kidney cancers. Crit Rev Oncol Hematol. 2003;46(Suppl):S105–15.

    Article  Google Scholar 

  5. 5.

    Dong F, et al. Prognostic value of site-specific metastases and therapeutic roles of surgery for patients with metastatic bladder cancer: a population-based study. Cancer Manag Res. 2017;9:611–26.

    Article  Google Scholar 

  6. 6.

    Tsao MN, et al. Radiotherapeutic and surgical management for newly diagnosed brain metastasis(es): an American Society for Radiation Oncology evidence-based guideline. Pract Radiat Oncol. 2012;2:210–25.

    Article  Google Scholar 

  7. 7.

    Rades D, et al. Comparison of stereotactic radiosurgery (SRS) alone and whole brain radiotherapy (WBRT) plus a stereotactic boost (WBRT + SRS) for one to three brain metastases. Strahlenther Onkol Organ Dtsch Rontgengesellschaft Al. 2008;184:655–62.

    Article  Google Scholar 

  8. 8.

    Brown PD, et al. Effect of radiosurgery alone vs radiosurgery with whole brain radiation therapy on cognitive function in patients with 1 to 3 brain metastases: a randomized clinical trial. JAMA. 2016;316:401–9.

    Article  Google Scholar 

  9. 9.

    Mahmoud-Ahmed AS, et al. Brain metastases from bladder carcinoma: presentation, treatment and survival. J Urol. 2002;167:2419–22.

    Article  Google Scholar 

  10. 10.

    Fokas E, Henzel M, Engenhart-Cabillic R. A comparison of radiotherapy with radiotherapy plus surgery for brain metastases from urinary bladder cancer: analysis of 62 patients. Strahlenther Onkol Organ Dtsch Rontgengesellschaft Al. 2010;186:565–71.

    Article  Google Scholar 

  11. 11.

    Bellmunt J, et al. Prognostic factors in patients with advanced transitional cell carcinoma of the urothelial tract experiencing treatment failure with platinum-containing regimens. J Clin Oncol Off J Am Soc Clin Oncol. 2010;28:1850–5.

    Article  Google Scholar 

  12. 12.

    Ishioka J, et al. Development of a nomogram incorporating serum C-reactive protein level to predict overall survival of patients with advanced urothelial carcinoma and its evaluation by decision curve analysis. Br J Cancer. 2012;107:1031–6.

    CAS  Article  Google Scholar 

  13. 13.

    Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16:e173–80.

    Article  Google Scholar 

  14. 14.

    Tibshirani R. Regression shrinkage and selection via the Lasso. J R Stat Soc Ser B Methodol. 1996;58:267–88.

    Google Scholar 

  15. 15.

    Mayr R, et al. Predictive capacity of four comorbidity indices estimating perioperative mortality after radical cystectomy for urothelial carcinoma of the bladder. BJU Int. 2012;110:E222–7.

    Article  Google Scholar 

  16. 16.

    Herr HW, Donat SM, Bajorin DF. Post-chemotherapy surgery in patients with unresectable or regionally metastatic bladder cancer. J Urol. 2001;165:811–4.

    CAS  Article  Google Scholar 

  17. 17.

    Chen C, Hu L, Chen Y, Hou J. The prognostic value of histological subtype in patients with metastatic bladder cancer. Oncotarget. 2017;8:28408–17.

    PubMed  PubMed Central  Google Scholar 

  18. 18.

    Nakagawa T, et al. Oncologic outcome of metastasectomy for urothelial carcinoma: who is the best candidate? Ann Surg Oncol. 2017;24:2794–800.

    Article  Google Scholar 

  19. 19.

    Alfred Witjes J, et al. Updated 2016 EAU guidelines on muscle-invasive and metastatic bladder cancer. Eur Urol. 2017;71:462–75.

    CAS  Article  Google Scholar 

  20. 20.

    Hussain SA, et al. A study of split-dose cisplatin-based neo-adjuvant chemotherapy in muscle-invasive bladder cancer. Oncol Lett. 2012;3:855–9.

    CAS  PubMed  PubMed Central  Google Scholar 

  21. 21.

    Morales-Barrera R, et al. Cisplatin and gemcitabine administered every two weeks in patients with locally advanced or metastatic urothelial carcinoma and impaired renal function. Eur J Cancer Oxf Engl. 2012;1990(48):1816–21.

    Article  Google Scholar 

  22. 22.

    Ghahestani SM, Shakhssalim N. Palliative treatment of intractable hematuria in context of advanced bladder cancer: a systematic review. Urol J. 2009;6:149–56.

    PubMed  Google Scholar 

  23. 23.

    Tsao MN, et al. International practice survey on the management of brain metastases: third international consensus workshop on palliative radiotherapy and symptom control. Clin Oncol. 2012;24:e81–92.

    CAS  Article  Google Scholar 

Download references


Not applicable.


This work was supported by the National Natural Science Foundation of China (81672515), Shanghai Pujiang Program (16PJD039) and Shanghai Municipal Education Commission—Gaofeng Clinical Medicine Grant (20172019).

Author information




ZXY and ZZ contributed equally to the formation this manuscript. ZHL, YYH and WYS designed and supervised the procedure. SG, ZXY and ZZ collected and analyzed the data. KW and ZXY visualized the result. RJW and XYM provided the statistical guidance. FS and XW polished the manuscript. All authors read and approved the final manuscript.

Corresponding authors

Correspondence to Wenyin Shi or Yinyan He or Zhihong Liu.

Ethics declarations

Ethics approval and consent to participate

Extraction of data from the NCDB did not require extra informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (, which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Yao, Z., Zheng, Z., Ke, W. et al. Prognostic nomogram for bladder cancer with brain metastases: a National Cancer Database analysis. J Transl Med 17, 411 (2019).

Download citation


  • Bladder cancer
  • Brain metastasis
  • Machine learning
  • Nomogram
  • Overall survival