Combination of molecular alterations and smoking intensity predicts bladder cancer outcome†
A report from the Los Angeles Cancer Surveillance Program
Version of Record online: 14 JAN 2013
Copyright © 2013 American Cancer Society
Volume 119, Issue 4, pages 756–765, 15 February 2013
How to Cite
Mitra, A. P., Castelao, J. E., Hawes, D., Tsao-Wei, D. D., Jiang, X., Shi, S.-R., Datar, R. H., Skinner, E. C., Stein, J. P., Groshen, S., Yu, M. C., Ross, R. K., Skinner, D. G., Cortessis, V. K. and Cote, R. J. (2013), Combination of molecular alterations and smoking intensity predicts bladder cancer outcome. Cancer, 119: 756–765. doi: 10.1002/cncr.27763
The authors are grateful to Peter A. Jones, PhD, DSc, for his guidance and leadership that, in large part, enabled this collaborative effort. This paper is dedicated to the memory of John P. Stein, MD, and Ronald K. Ross, MD.
- Issue online: 4 FEB 2013
- Version of Record online: 14 JAN 2013
- Manuscript Accepted: 10 JUL 2012
- Manuscript Revised: 30 MAY 2012
- Manuscript Received: 21 MAR 2012
- urinary bladder neoplasms;
- tumor suppressor protein p53;
- apoptotic protease-activating factor 1;
Traditional single-marker and multimarker molecular profiling approaches in bladder cancer do not account for major risk factors and their influence on clinical outcome. This study examined the prognostic value of molecular alterations across all disease stages after accounting for clinicopathological factors and smoking, the most common risk factor for bladder cancer in the developed world, in a population-based cohort.
Primary bladder tumors from 212 cancer registry patients (median follow-up, 13.2 years) were immunohistochemically profiled for Bax, caspase-3, apoptotic protease-activating factor 1 (Apaf-1), Bcl-2, p53, p21, cyclooxygenase-2, vascular endothelial growth factor, and E-cadherin alterations. “Smoking intensity” quantified the impact of duration and daily frequency of smoking.
Age, pathological stage, surgical modality, and adjuvant therapy administration were significantly associated with survival. Increasing smoking intensity was independently associated with worse outcome (P < .001). Apaf-1, E-cadherin, and p53 were prognostic for outcome (P = .005, .014, and .032, respectively); E-cadherin remained prognostic following multivariable analysis (P = .040). Combined alterations in all 9 biomarkers were prognostic by univariable (P < .001) and multivariable (P = .006) analysis. A multivariable model that included all 9 biomarkers and smoking intensity had greater accuracy in predicting prognosis than models composed of standard clinicopathological covariates without or with smoking intensity (P < .001 and P = .018, respectively).
Apaf-1, E-cadherin, and p53 alterations individually predicted survival in bladder cancer patients. Increasing number of biomarker alterations was significantly associated with worsening survival, although markers comprising the panel were not necessarily prognostic individually. Predictive value of the 9-biomarker panel with smoking intensity was significantly higher than that of routine clinicopathological parameters alone. Cancer 2013. © 2013 American Cancer Society.
Urothelial carcinoma of the bladder (UCB), the sixth most common cancer in the United States,1 develops through alterations in several cellular processes.2 Profiling such alterations can identify biomarkers that can improve risk assessment over traditional clinical and histopathological parameters. Early studies in UCB profiled single markers, or multiple markers within a single cellular process. More recently, we and others have used multimarker approaches to profile alterations across several pathways.3-6 However, most of these studies do not account for common UCB risk factors and their influence on outcome. Cigarette smoking is an established risk factor for UCB in the United States due to carcinogenic effects of aromatic amines in tobacco smoke.7
To our knowledge, this is one of the initial UCB studies to examine the prognostic value of molecular alterations and cigarette smoke exposure in a multicenter population-based cohort after accounting for routine clinicopathological criteria. Proteins central to key cellular processes commonly implicated in urothelial tumorigenesis such as apoptosis, cell cycle regulation, inflammation, angiogenesis, and invasion were immunohistochemically profiled7; several of these biomarkers have been previously identified as being individually prognostic in UCB, although none were combined in a broad multipathway panel.8-18 This was an effort to identify a marker panel that could predict outcome independent of standard criteria and smoking history.
MATERIALS AND METHODS
This study, approved by the University of Southern California Institutional Review Board, reports on 212 patients with primary UCB with available archived tissues recruited through the multi-institutional Los Angeles County Cancer Surveillance Program between 1987 and 1996.19 This program is part of the California Cancer Registry and the National Cancer Institute-sponsored Surveillance, Epidemiology and End Results program that prospectively collects population-based cancer patient data. Because Los Angeles, California, represents a microcosm of the United States's shifting demographics, this population may allow better understanding of UCB nationwide. Patients with distant metastasis at diagnosis were excluded. All patients provided informed consent.
Tumor staging and grading were standardized to the American Joint Committee on Cancer and World Health Organization systems, respectively.20, 21 A total of 166 (78%) patients had high-grade UCB. Most non-muscle invasive (Ta/CIS/T1,N−) tumor patients were treated by transurethral resection alone (n = 99, 71%) or in combination with adjuvant therapy (n = 30, 22%). In contrast, most patients with muscle-invasive (T2-4,N−) or node-positive (any T,N+) tumors were treated with radical cystectomy alone (n = 35, 48%) or in combination with adjuvant therapy (n = 19, 26%).
Defining Cigarette Smoke Exposure
Patients were interviewed in person regarding smoking history up to 2 years before UCB diagnosis using a structured questionnaire.22 A patient was classified as a smoker if he/she smoked at least 1 cigarette daily for 6 months or longer. For smokers (n = 184), duration of regular smoking until diagnosis and number of cigarettes smoked daily were also queried. To analyze impact of smoking at the pathological and molecular levels, a “smoking intensity” variable that combined raw smoking metrics was introduced, and the cohort was divided into 3 groups (Fig. 1A): group 1 = nonsmokers or patients who smoked ≤ 20 cigarettes/day for ≤ 30 years (n = 67, 32%); group 2 = patients who smoked for 31 to 40 years or > 20 cigarettes/day for ≤ 30 years (n = 93, 44%); group 3 = patients who smoked for > 40 years (n = 52, 24%). Nonsmokers were combined with light smokers in group 1, because a full sensitivity analysis revealed no substantive outcome differences between the 2 subgroups (log-rank P = .65 between subgroups).
Serial 5-μm sections of formalin-fixed, paraffin-embedded UCB tissues were obtained for each patient. Expressions of Bax, caspase-3, apoptotic protease-activating factor 1 (Apaf-1), Bcl-2, p53, p21, cyclooxygenase-2 (COX-2), vascular endothelial growth factor (VEGF), and E-cadherin were analyzed by immunohistochemistry following standardized procedures (Fig. 2).12 Table 1 lists procedural specifics for each marker. Antigen retrieval was performed in all cases. Every run included established positive controls, and substitution of primary antibody with rabbit immunoglobulin fraction served as negative control. Immunoreactivity was detected by avidin-biotin complex immunoperoxidase system. Diaminobenzidine was used as chromogen, and hematoxylin was used to counterstain.
|Marker||Cellular Process||Clone||Antibody Dilution||Definition of Altered Status||Cases Evaluated (%)|
|Baxab||Apoptosis||2D2c||1:400||Absent/weak staining, or ≤30% cells stained moderately, or <10% cells stained strongly||212 (100)|
|Caspase-3ab||Apoptosis||3CSP03c||1:200||Absent/weak staining, or ≤30% cells stained moderately, or <10% cells stained strongly||212 (100)|
|Apaf-1ab||Apoptosis||–d||1:200||≤75% cells stained||212 (100)|
|Bcl-2be||Apoptosis||124c||1:40||>25% cells stained weakly, or ≥10% cells stained moderately, or any strong staining||211 (99)|
|p53fg||Cell cycle regulation||1801c||1:3000||>25% cells stained strongly||206 (97)|
|p21fg||Cell cycle regulation||EA10c||1:100||<10% cells stained||211 (99)|
|COX-2ab||Inflammation||SP21h||1:100||>75% cells stained weakly, or >50% cells stained moderately, or >25% cells stained strongly||211 (99)|
|VEGFab||Angiogenesis||JH121c||1:100||>75% cells stained||203 (96)|
|E-cadherinij||Invasion||4A2C7c||1:100||Absent staining, or ≤25% cells stained weakly, or <10% cells stained moderately||211 (99)|
All cases were independently interpreted under light microscope by at least 2 investigators (A.P.M./D.H./S.R.S./R.J.C.), blinded to outcome and coreaders' results. In case of discrepant results (< 5% cases for each marker), the case was coreviewed with another investigator to reach a consensus. Reasons that prevented case evaluation (< 5% cases for each marker) included tissue loss or introduction of tissue folding artifacts during the staining process. Entire sections were scored for average staining intensity (absent/weak/moderate/strong), and percentage of cells showing immunoreactivity at that intensity. Marker expressions were categorized as wild-type or altered; alteration thresholds were defined through joint analysis of staining characteristics of normal urothelium, previously defined cutoffs, and biological functions of the proteins. These cutoffs were based on percentage of immunoreactive cells with or without the added value of average staining intensity, thus making the parameters semiquantitative (Table 1).
Survival outcome was calculated from date of surgery to date of death; surviving patients were censored at last follow-up. Contingency tables and Pearson's chi-square test were used to determine associations between baseline characteristics, smoking intensity, and marker alterations. Log-rank test was used to examine univariable associations with outcome.23 Gehan-Breslow-Wilcoxon (GBW) test was also used to examine associations between categorical variables and survival, giving more weight to events at early time points.24 This was done to account for early deaths that were possibly due to disease, rather than later events that could be due to other causes. Relative risks and associated 95% confidence intervals for univariable analyses were calculated as described by Berry et al.25
Multivariable Cox proportional hazards models were constructed to assess the prognostic value of altered markers individually and in combination. Logistic regression models at the 5-year survival status were constructed to assess predictive accuracies as determined by area under the receiver operating characteristic (ROC) curve when smoking and individual marker alterations were successively added to a base model that comprised standard clinicopathological predictors. Differences in predictive accuracies between 2 models were estimated by a z-score test statistic with the null hypothesis being that the datasets arose from binormal ROC curves with equal areas beneath them.26
Reported P values are 2-sided; level of significance was set at P = .050, whereas .050 < P < .100 was considered as approaching statistical significance. Analyses were performed using SAS, version 9.2, and Epilog Plus, version 1.0.
Clinicopathological Criteria, and Smoking History, Individual Markers, and Outcome
Patients included 196 (92%) whites and 168 (79%) males. Median age at diagnosis was 58.9 years (range, 30.5-64.9 years). Median follow-up was 13.2 years (range, 0.5-20 years), during which 90 (42%) patients died. Associations of clinicopathological parameters with smoking intensity are shown in Table 2.
|Group 1 n (row %)||Group 2 n (row %)||Group 3 n (row %)||Pb|
|All patients||212||67 (32)||93 (44)||52 (24)|
|<60 y||121||48 (40)||62 (51)||11 (9)|
|≥60 y||91||19 (21)||31 (34)||41 (45)|
|Non–muscle invasive||139||47 (34)||64 (46)||28 (20)|
|Muscle-invasive||55||13 (24)||24 (43)||18 (33)|
|Nodal metastasis||18||7 (39)||5 (28)||6 (33)|
|Transurethral resection||148||51 (34)||63 (43)||34 (23)|
|Radical cystectomy||64||16 (25)||30 (47)||18 (28)|
|Surgery only||150||52 (35)||62 (41)||36 (24)|
|Surgery+adjuvant therapy||62||15 (24)||31 (50)||16 (26)|
Increasing pathological stage was associated with increasing frequency of p53 (P < .001), E-cadherin (P = .002), p21 (P = .022), and Apaf-1 (P = .047) alterations. Expectedly, p53 (P < .001), E-cadherin (P = .001), and p21 (P = .046) expressions were also associated with surgery; surgical modality was associated with pathological stage (P < .001), where 129 (93%) patients with non-muscle invasive tumors had transurethral resection, and 54 (74%) patients with muscle-invasive or nodal metastasized disease had radical cystectomy.
Age (P < .001), pathological stage (P < .001), surgical modality (P = .050), and adjuvant therapy administration (P < .001) were associated with survival by log-rank analysis. These significant associations were also confirmed at early time points by the GBW test (Table 3).
|Parameter||Clinical Outcome (Univariable)||Clinical Outcome (Multivariablea)||Clinical Outcome (Multivariableb)|
|n||Relative Risk of Dying (95% CI)||Pc||Pd||Relative Risk of Dying (95% CI)||Pe||Relative Risk of Dying (95% CI)||Pe|
|<60 y||121||1.00 (Reference)||1.00 (Reference)||1.00 (Reference)|
|≥60 y||91||2.11 (1.39, 3.20)||1.37 (0.84, 2.23)||1.29 (0.81, 2.05)|
|Non–muscle invasive||139||1.00 (Reference)||1.00 (Reference)||1.00 (Reference)|
|Muscle-invasive||55||1.56 (0.98, 2.49)||1.44 (0.71, 2.91)||1.49 (0.78, 2.85)|
|Nodal metastasis||18||4.36 (2.32, 8.18)||5.83 (2.06, 16.54)||4.53 (1.72, 11.94)|
|Transurethral resection||148||1.00 (Reference)||1.00 (Reference)||1.00 (Reference)|
|Radical cystectomy||64||1.52 (0.99, 2.34)||0.76 (0.38, 1.51)||0.73 (0.38, 1.40)|
|Surgery only||150||1.00 (Reference)||1.00 (Reference)||1.00 (Reference)|
|Surgery+adjuvant therapy||62||2.06 (1.35, 3.13)||1.53 (0.96, 2.46)||1.69 (1.08, 2.66)|
|Group 1||67||1.00 (Reference)||1.00 (Reference)||1.00 (Reference)|
|Group 2||93||2.15 (1.17, 3.93)||2.59 (1.29, 5.21)||2.18 (1.15, 4.13)|
|Group 3||52||5.76 (3.08, 10.76)||6.11 (3.02, 12.37)||5.80 (2.87, 11.71)|
|Wild-type||33||1.00 (Reference)||1.00 (Reference)||–|
|Altered||179||1.88 (0.94, 3.74)||1.92 (0.91, 4.06)||–|
|Wild-type||38||1.00 (Reference)||1.00 (Reference)||–|
|Altered||174||0.77 (0.46, 1.28)||0.56 (0.32, 1.00)||–|
|Wild-type||91||1.00 (Reference)||1.00 (Reference)||–|
|Altered||121||1.82 (1.18, 2.81)||1.57 (0.96, 2.57)||–|
|Wild-type||159||1.00 (Reference)||1.00 (Reference)||–|
|Altered||52||1.24 (0.77, 2.00)||1.41 (0.84, 2.36)||–|
|Wild-type||161||1.00 (Reference)||1.00 (Reference)||–|
|Altered||45||1.66 (1.04, 2.66)||1.12 (0.66, 1.90)||–|
|Wild-type||188||1.00 (Reference)||1.00 (Reference)||–|
|Altered||23||1.73 (0.95, 3.14)||1.12 (0.53, 2.34)||–|
|Wild-type||66||1.00 (Reference)||1.00 (Reference)||–|
|Altered||145||1.22 (0.77, 1.94)||1.38 (0.81, 2.37)||–|
|Wild-type||50||1.00 (Reference)||1.00 (Reference)||–|
|Altered||153||1.49 (0.88, 2.50)||1.05 (0.61, 1.81)||–|
|Wild-type||196||1.00 (Reference)||1.00 (Reference)||–|
|Altered||15||2.23 (1.15, 4.32)||2.34 (1.09, 5.02)||–|
|Number of altered markers||<.001||<.001||–||.006|
|≤3||52||1.00 (Reference)||–||1.00 (Reference)|
|4-5||129||1.82 (1.03, 3.21)||–||1.38 (0.77, 2.50)|
|6-9||31||4.97 (2.54, 9.72)||–||3.22 (1.52, 6.84)|
Associations of Smoking With Individual Markers and Outcome
Patients who smoked > 30 years had greater risk of COX-2 alterations (P = .020; data not shown). Number of cigarettes smoked daily was associated with Bax expression (P = .008; data not shown). Increased smoking intensity was associated with poor outcome (log-rank, GBW P < .001; Table 3, Fig. 1B). In addition, increase in smoking duration and number of cigarettes smoked daily were associated with worse prognosis (log-rank P < .001 and P = .003, respectively). Smoking intensity was not associated with individual marker alterations.
Individual Marker Alterations
To investigate whether individual protein alterations could predict downstream effects, associations of perturbations in various molecular pathways with each other were examined. Expression patterns of several markers were associated with each other within the same and across different cellular processes (data not shown). Bax, p21, and p53 alterations were associated with concomitant alterations in caspase-3 (P < .001), Apaf-1 (P = .028), and E-cadherin (P = .002), respectively. However, Bcl-2 alterations were associated with wild-type caspase-3 (P < .001) and Bax (P = .032) phenotypes, thereby suggesting possible exclusivity of antiapoptotic and proapoptotic processes in these tumors. These expected associations of alterations across various pathways indicated that our observations were biologically valid.
Individual Markers and Clinical Outcome
Apaf-1 (P = .005), E-cadherin (P = .014), and p53 (P = .032) alterations were associated with poor prognosis by log-rank analysis; these associations were also significant by the GBW test (Table 3). Expression of p21 (P = .063) and Bax (P = .067) also approached significance by log-rank analysis; the prognostic value of p21 was significant at early time points (GBW P = .044).
A multivariable Cox proportional hazards model was then created to include all baseline clinicopathological prognostic criteria, smoking intensity, and individual marker alterations (Table 3). Pathological stage (P = .003) and smoking intensity (P < .001) were the only baseline characteristics that still retained their associations with outcome. E-cadherin was the only marker that remained predictive for outcome following this analysis (P = .040). Caspase-3 (P = .058), Apaf-1 (P = .067), and Bax (P = .068) alterations also approached significance for outcome by multivariable analysis.
Prognostic Performance of Combined Molecular Alterations Relative to Baseline Parameters
We next proceeded to analyze the combined prognostic value of alterations in all 9 biomarkers, irrespective of their individual prognostic potential. Patients were assigned to 1 of 3 groups: ≤ 3 markers altered (n = 52), 4 or 5 markers altered (n = 129), and 6 to 9 markers altered (n = 31). When thus divided, an increasing number of marker alterations was associated with poor prognosis (log-rank, test for trend, GBW P < .001; Table 3, Fig. 3A). When subset analyses were performed using this categorization, increasing number of marker alterations were also associated with poor outcome in patients with non–muscle invasive (log-rank P = .022, GBW P = .012) and muscle-invasive (log-rank P = .001, GBW P = .002) UCB. The association between number of altered markers and outcome remained significant in a multivariable Cox proportional hazards model that included all baseline clinicopathological covariates and smoking intensity (P = .006; Table 3).
Finally, to determine if there was any substantially increased prognostic value to adding smoking intensity and alterations in the 9 biomarkers over standard clinicopathological covariates, a base logistic regression model was created using age, pathological stage, surgical modality, and adjuvant therapy administration (Fig. 3B). The accuracy of this model alone in predicting survival, as measured by area under the ROC curve, was 75.6%. When smoking intensity was added to this model, its predictive accuracy increased to 80.9%. This increase in predictive accuracy, based solely on the addition of smoking intensity, was statistically significant (P = .012, Fig. 3C). Individual marker alterations were then added successively to the model, beginning with E-cadherin and progressively adding 1 marker at a time based on their decreasing significance of association with outcome as determined by the multivariable Cox analysis that included individual markers (Table 3). With each new marker addition, the predictive accuracy of each successive model continued to increase significantly over the base model, irrespective of the marker's individual association with outcome. Predictive accuracy of the final model that included base model covariates, smoking intensity, and alterations in all 9 markers was 85.4%, which was significantly higher than the base model alone (P < .001; Fig. 3B,C). Moreover, the final model's predictive accuracy was significantly higher than the model consisting of baseline covariates and smoking intensity (P = .018).
This study employed a multipathway-based approach to profile alterations in 9 proteins involved in cell cycle regulation, apoptosis, inflammation, angiogenesis, and invasion, which are all crucial processes in UCB pathogenesis.7 Associations of these alterations with smoking history and outcome were examined across all disease stages in a population-based cohort. The study highlighted the prognostic importance of smoking intensity, a novel variable that accounted for incrementally harmful effects of smoking. It also identified Apaf-1, E-cadherin, and p53 as individual markers of UCB outcome, with E-cadherin remaining prognostic after multivariable modeling. Importantly, combination of alterations in all 9 markers was robustly prognostic even after multivariable analysis. As with our previous observations,5, 13 this study indicates that number of alterations is directly proportional to UCB outcome in a progressive fashion, confirming that accumulation of molecular aberrations is more important than the individually measured alterations themselves. Patients with alterations in 6 to 9 markers had a very poor outcome, and these individuals clearly need more aggressive management. However, unlike prior investigations,5, 13 this study also indicates that the 9-biomarker panel can be predictive of outcome even when markers in the panel may not be individually prognostic.
Tobacco smoking is the most important risk factor for UCB in the developed world.7 Although traditional parameters such as smoking status and duration, and number of cigarettes smoked daily are reasonable measures of smoking habit, they may not gauge the severity and chronicity of exposure in a fashion most relevant to bladder tumorigenesis. Measurement of “pack years,” which is based on number of years smoked and average packs smoked daily, does not distinguish between individuals who smoke fewer packs for a longer duration from those who smoke more packs for a shorter duration, thereby poorly assessing chronicity of exposure.27 To reconcile this issue, we defined “smoking intensity” risk strata based on traditional smoking history parameters; patients were categorized into groups organized in the order of increasing detrimental effects of tobacco smoke. Although the association of smoking and UCB incidence is established,22 our analysis also revealed that increasing smoking intensity influences the disease's biological behavior with adverse effects on prognosis.
Previous studies have examined proteins interrogated in this report, either singly or in combination, with respect to UCB prognosis.8-18 However, none of these studies accounted for the influence of smoking on protein alterations or outcomes in their cohorts. Furthermore, these studies did not profile such a comprehensive biomarker panel involving multiple cellular processes on a single cohort. Survival analysis after adjustment of clinicopathological and smoking parameters permits characterization of biomarker panels that predict outcome independent of these prognostic criteria, thereby allowing individualized patient management.
Associations between individual markers revealed interactions between apoptotic, cell cycle regulation, and invasion processes in urothelial tumorigenesis. Besides validating our multipathway-based approach toward candidate biomarker selection, this also served as a biological assessment of data quality in this study. p53 alterations were absent in low-grade Ta tumors, and increased to 41% of all muscle-invasive tumors, confirming the protein's role in more aggressive disease (data not shown).28, 29
E-cadherin, p53, and Apaf-1 were individual outcome predictors. Decreased E-cadherin expression has been associated with recurrence, progression, and poor survival in UCB.18 Previous studies have suggested a role for p53 in predicting prognosis and chemotherapeutic response, although this observation was not validated in a prospective trial.11, 30, 31 Although prior reports have not described a prognostic role of Apaf-1 in UCB, studies on this proapoptotic protein were limited in terms of cohort size and range of pathological stages examined.10
Multivariable analysis found E-cadherin alterations prognostic independent of baseline clinicopathological parameters and smoking. Sequential addition of individual marker alterations revealed that a composite model containing all 9 biomarkers could more significantly and accurately predict UCB outcome than baseline parameters without or with smoking intensity. This underscores the added and independent prognostic value of these biomarkers over routine clinicopathological parameters. The prognostic importance of this 9-biomarker panel was apparent as increasing number of alterations was associated with worsening survival, irrespective of which specific proteins harbored those alterations.
This study is unique in examining molecular alterations in primary UCB associated with smoking in a population-based cohort where long-term survival was assessed. Protein expression was examined on serial tumor sections rather than tissue microarrays, which allowed examination of areas of heterogeneity that could have been potentially missed on microarray cores. This allowed semiquantitative marker assessment using percentage of cells showing altered protein expression and relative preponderance of altered molecules measured as staining intensity in an average tumor cell.
Variables that could potentially influence survival, including age, pathological stage, smoking, and treatment type, were accounted for by multivariable analysis. Although we were limited by the availability of disease-specific mortality, the long follow-up ensured that all events due to disease were included in the death statistics. Furthermore, the GBW test was equally or more significant in most cases where log-rank analysis was significant for outcome, indicating that the biomarkers were truly and specifically associated with UCB prognosis rather than merely related to simple measures of longevity.
In conclusion, these findings can potentially affect UCB management across all stages. Documentation of tobacco-use history is important; increasing smoking intensity was identified as a unique independent variable that may be associated with poor prognosis. Furthermore, the 9-biomarker panel can be employed as a tool to identify patients in need of more aggressive treatment, independent of routine clinicopathological parameters or smoking history. Although validation of this panel is needed, this study clearly demonstrates that a multipathway-based approach to constructing rational marker panels holds potential for future UCB management.
This study is funded by National Institutes of Health, National Cancer Institute grants CA-71921, CA-65726, and CA-86871.
CONFLICT OF INTEREST DISCLOSURE
The authors made no disclosure.
- 19http://www.usc.edu/lacsp. Accessed May 28, 2012., , . Cancer in Los Angeles County: incidence and mortality by race/ethnicity, 1988-2009.
- 20Urinary bladder. In: Edge SB, Byrd DR, Compton CC, Fritz AG, Greene FL, Trotti A, editors. AJCC Cancer Staging Manual. 7th edition. New York, NY: Springer; 2010: 497-506.
- 21Histologic Typing of Urinary Bladder Tumours. Geneva, Switzerland: World Health Organization; 1973., , .
- 23Survival analysis. In: Applied Probability and Statistics. Wiley Series in Probability and Mathematical Statistics. New York, NY: Wiley; 1981: 44-102., , .
- 24Hypothesis testing. In: Survival Analysis: Techniques for Censored and Truncated Data. New York, NY: Springer; 2003: 201-242., .
- 26A new approach for testing the significance of differences between ROC curves measured from correlated data. In: Deconinck F, editor. Information Processing in Medical Imaging. The Hague, Netherlands: Martinus Nijhoff; 1984: 432-445., , .
- 27Cohort studies. In: Rothman KJ, Greenland S, Lash TL, editors. Modern Epidemiology. Philadelphia, PA: Lippincott Williams & Wilkins; 2008: 100-110., .