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

J Transl Med 2019 AI 7 Explanations View Original
Original Paper (PDF)

Unable to display PDF. Download it here or view on PMC.

Plain-English Explanations
Pages 1-2
Bladder Cancer Brain Metastases: A Rare and Deadly Scenario

The Severity of Metastatic Bladder Cancer. Bladder cancer is the eighth most common cancer in the US, with an estimated 79,030 new cases and 16,870 deaths in 2017. Between 10-15% of patients already have metastatic disease at diagnosis, and 15-30% of high-grade cases eventually progress to advanced disease. Patients with non-organ-confined disease rarely survive more than 3-6 months even with chemotherapy.

Brain Metastases Are Rare But Particularly Devastating. According to prior population-based studies, only about 4.1% of patients with metastatic bladder cancer develop brain metastases. This rarity means no randomized clinical trials have specifically studied treatment strategies for this subgroup, leaving clinicians without evidence-based guidance on optimal management. Prior small case series (16 and 62 patients) produced conflicting conclusions about whether radiation alone or radiation plus surgery provides better outcomes.

The Need for a Prognostic Tool. No prognostic model specifically designed for bladder cancer patients with brain metastases existed before this study. General prognostic models for metastatic bladder cancer do not account for the unique biology and treatment challenges of brain involvement. A validated tool to predict survival and guide treatment selection for this rare group was urgently needed.

Nomograms as Clinical Decision Tools. Nomograms are graphical computational tools that combine multiple clinical variables into a personalized probability estimate. Unlike staging systems that group patients into broad categories, nomograms generate individualized survival probabilities that reflect each patient's specific constellation of disease characteristics and treatments, supporting personalized clinical decisions.

TL;DR: Brain metastases occur in approximately 4% of metastatic bladder cancer patients, with extremely poor prognosis and no dedicated evidence base to guide treatment, motivating development of the first prognostic nomogram for this population.
Pages 2-3
National Cancer Database Study Design

Data Source. The National Cancer Database (NCDB) was queried for patients diagnosed with histologically confirmed bladder cancer between 2004 and 2015 who had brain metastatic disease at the time of presentation. The NCDB is a hospital-based registry of cases treated at American College of Surgeons Commission on Cancer-accredited cancer programs, representing approximately 70% of all newly diagnosed cancers in the United States.

Patient Selection. Of 268 bladder cancer patients with brain involvement identified, 234 met the inclusion criteria: age above 18, bladder cancer as the primary diagnosis, confirmed brain metastasis, information on other distant metastatic sites (bone, liver, lung, distant lymph nodes), active follow-up, and survival exceeding 30 days. This is the largest published cohort of bladder cancer patients with brain metastases analyzed for prognostic purposes.

Training and Validation Split. Using a computer-generated random seed, 169 patients were assigned to the training cohort and 65 to the internal validation cohort. Eighteen clinical variables were collected, including demographic characteristics, tumor pathology, TNM stage, metastatic sites, and four treatment modalities (surgery of primary site, chemotherapy, radiation therapy, and palliative care).

Machine Learning Feature Selection. The LASSO (Least Absolute Shrinkage and Selection Operator) method, a machine learning approach for high-dimensional variable reduction, was applied to identify the most predictive variables from the 18 candidates. LASSO applies a penalty that shrinks less important variable coefficients to zero, automatically selecting the minimal set of features with the greatest predictive value while preventing overfitting.

TL;DR: 234 bladder cancer patients with brain metastases from the National Cancer Database were analyzed, with LASSO machine learning used to select the most predictive variables from 18 candidates.
Pages 3-4
Key Prognostic Variables Identified

Six Variables Selected by LASSO. From 18 candidate variables, LASSO with tenfold cross-validation identified seven: tumor grade, surgery of primary site, chemotherapy, radiation therapy, palliative care, brain-confined metastasis status, and Charlson/Deyo Comorbidity Score (CDCC_Score). After Cox regression confirmation, six were incorporated into the final nomogram (grade was dropped as it did not independently predict survival on multivariate analysis).

Chemotherapy Was the Strongest Protective Factor. Chemotherapy had the strongest independent association with improved survival, with a hazard ratio of 0.213 (p less than 0.001) in multivariate analysis, meaning patients receiving chemotherapy had an approximately 79% reduction in the hazard of death. This aligns with European Association of Urology guidelines making chemotherapy the first-line treatment for metastatic urothelial carcinoma, though selection of eligible patients is complex.

Surgery of the Primary Site Paradoxically Associated with Worse Survival. Patients who received no surgery of the primary site had a hazard ratio of 2.529 (p less than 0.001) compared to minimal invasive surgery. This likely reflects selection bias: patients well enough to receive surgery had better underlying fitness, while those managed without surgery were already too ill, making surgery appear protective as a surrogate for overall health status.

Brain-Confined Metastasis and Comorbidity Score. Patients whose metastatic disease was confined to the brain (without simultaneous bone, liver, or lung involvement) had better survival than those with multi-site metastases, with a hazard ratio of 2.229 for non-brain-confined disease (p = 0.02). Higher Charlson/Deyo Comorbidity Score (indicating more severe baseline comorbidities) was associated with worse survival, particularly at the highest score level (p = 0.04).

TL;DR: Chemotherapy, surgery of the primary site, comorbidity score, and brain confinement of metastatic disease were the strongest independent predictors of survival in bladder cancer patients with brain metastases.
Pages 4-6
Nomogram Performance and Validation

Strong Discriminatory Performance. The six-variable nomogram demonstrated strong predictive discrimination in the training cohort, with area under the ROC curve (AUC) values of 0.823 (95% CI 0.758-0.889) for 6-month survival and 0.854 (95% CI 0.785-0.924) for 1-year survival prediction. AUC values above 0.8 indicate good-to-excellent discriminatory ability in clinical prognostication.

Validated in Independent Cohort. In the internal validation cohort of 65 patients, the nomogram maintained similar performance with AUC values of 0.838 (95% CI 0.738-0.937) for 6-month survival and 0.809 (95% CI 0.680-0.939) for 1-year survival. The consistency between training and validation performance indicates the model is generalizable and not simply overfitted to training data characteristics.

Survival Stratification. The nomogram successfully stratified patients into clinically meaningful risk groups. In the training cohort, high-risk patients had a median survival of 1.91 months compared to 5.09 months for low-risk patients (p less than 0.0001). In the validation cohort, the separation was even greater: 1.68 months versus 8.05 months (p less than 0.0001), a fourfold difference in median survival between risk groups.

Superiority Over TNM Staging. Decision curve analysis demonstrated that when the threshold probability for clinical decision-making exceeded 0.4, the nomogram provided greater net benefit than the standard AJCC TNM staging system. This indicates the nomogram provides clinically actionable survival probability estimates beyond what staging alone can offer, particularly for patients with intermediate to high risk of early death.

TL;DR: The nomogram achieved AUC values of 0.82-0.85 in training and validation, separated median survival by 3-5 months between risk groups, and outperformed TNM staging for clinical decision support.
Pages 8-9
Treatment Implications for Brain-Metastatic Bladder Cancer

Chemotherapy Remains Central. Despite the poor prognosis of brain-metastatic bladder cancer, the data confirm that systemic chemotherapy provides the greatest survival benefit. The choice of regimen must account for patient fitness, kidney function, and performance status, as more than half of patients with metastatic urothelial cancer are ineligible for cisplatin-based therapy due to impaired renal function, poor performance, neuropathy, or heart failure.

Role of Brain Radiation. Radiation therapy to the brain was included in the nomogram despite borderline multivariate significance (p = 0.07) because conventional whole brain radiotherapy remains standard practice for brain metastases from many malignancies and carries clinical relevance for this patient group. The univariate association with improved survival (p = 0.04) justifies its inclusion as a consideration in the prognostic calculation.

Palliative Care Contributes to Survival. Palliative care interventions were significantly associated with improved survival (HR 0.631, p = 0.03), reflecting that symptom management -- including relief of ureteral obstruction, control of bleeding, and pain management -- may directly prolong survival in this debilitated population by preventing complications that accelerate death.

Brain-Confined Disease Warrants More Aggressive Treatment. Patients with metastases confined exclusively to the brain had better survival than those with simultaneous involvement of other organs. This subgroup may represent candidates for more aggressive local treatment of brain lesions (stereotactic radiosurgery, whole brain radiation, or even resection) in combination with systemic therapy, analogous to oligometastatic disease management in other cancers.

TL;DR: Chemotherapy provides the greatest survival benefit in brain-metastatic bladder cancer; patients with brain-only metastases and good performance status may be candidates for more aggressive treatment combinations.
Pages 8-9
Limitations and Future Directions

Largest Cohort to Date, But Still Limited. With 234 patients, this represents the largest study of bladder cancer patients with brain metastases, but the absolute number remains small, reflecting the rarity of this condition. Small sample sizes limit the statistical power to detect effects of less common treatment modalities and limit the precision of survival estimates.

Retrospective Database Limitations. The NCDB does not capture detailed information about specific chemotherapy regimens, sequences of treatment, immunotherapy, endocrine therapy, or specific radiation dosing. Treatment decisions in this population were not randomized, creating significant confounding by indication. Sicker patients were less likely to receive aggressive treatments, and their worse outcomes reflect their underlying condition rather than treatment inefficacy.

External Validation Needed. The nomogram was internally validated using a random split of the NCDB cohort, but external validation using patients from a different data source or institution is needed to confirm generalizability. Patient populations, treatment practices, and imaging capabilities vary across institutions and countries, and models can lose performance when applied outside their development environment.

Emerging Therapies Not Captured. The study period (2004-2015) predates the widespread adoption of immune checkpoint inhibitors in bladder cancer treatment. Pembrolizumab, atezolizumab, and other checkpoint inhibitors approved from 2016 onward have changed the treatment landscape for metastatic urothelial carcinoma, and their impact on outcomes for brain-metastatic patients is not reflected in this nomogram's underlying data.

TL;DR: Despite being the largest study in this rare population, the retrospective database design limits causal conclusions, external validation is needed, and emerging immunotherapy agents are not captured in the model's underlying data.
Page 11
A Tool for a Neglected Patient Population

First Validated Prognostic Tool. This study provides the first validated prognostic nomogram specifically designed for bladder cancer patients with brain metastases, filling an important gap in a population previously managed without any evidence-based survival estimation tools. The tool can now support individualized prognostic counseling and treatment planning discussions.

Clinical Utility of the Six-Variable Model. The nomogram's six variables -- surgery of primary site, chemotherapy, radiation therapy, palliative care, brain confinement of metastasis, and Charlson/Deyo comorbidity score -- are all routinely available in clinical practice without requiring additional testing. This simplicity makes the tool immediately implementable in clinical settings where treatment decisions for this difficult population must be made quickly.

Guiding Personalized Treatment Decisions. By generating individualized survival probability estimates, the nomogram can help clinicians identify patients most likely to benefit from aggressive treatment (chemotherapy, radiation) versus those for whom purely supportive care is most appropriate. This supports shared decision-making conversations that align treatment intensity with patient goals and prognosis.

TL;DR: The first validated nomogram for bladder cancer with brain metastases uses six readily available clinical variables to generate individualized survival probabilities, directly supporting treatment decisions for this neglected patient population.
Citation: Open Access, 2019. Available at: PMC6902467.