Association of Health-Care System and Survival in African American and Non-Hispanic White Patients With Bladder Cancer

Abstract

Background

African American patients with bladder cancer have inferior outcomes compared with non-Hispanic White (White) patients. We hypothesize that access to health care is a primary determinant of this disparity. We compared outcomes by race for patients with bladder cancer receiving care within the predominant hybrid-payer health-care model of the United States captured in the Surveillance, Epidemiology, and End Results (SEER) database with those receiving care within the equal-access model of the Veterans’ Health Administration (VHA).

Methods

African American and White patients diagnosed with bladder cancer were identified in SEER and VHA. Stage at presentation, bladder cancer–specific mortality (BCM), and overall survival (OS) were compared by race within each health-care system.

Results

The SEER cohort included 122 449 patients (93.7% White, 6.3% African American). The VHA cohort included 36 322 patients (91.0% White, 9.0% African American). In both cohorts, African American patients were more likely to present with muscle-invasive disease and metastases, but the differences between races were statistically significantly smaller in VHA. In SEER multivariable models, African American patients had worse BCM (hazard ratio [HR] = 1.22, 95% confidence interval [CI] = 1.15 to 1.29) and OS (HR = 1.26, 95% CI = 1.20 to 1.31). In contrast within the VHA, African American patients had similar BCM (HR = 0.97, 95% CI = 0.88 to 1.07) and OS (HR = 0.99, 95% CI = 0.93 to 1.05).

Conclusions

In this study of contrasting health-care models, receiving medical care in an equal-access system was associated with reduced differences in stage at presentation and eliminated disparities in survival outcomes for African American patients with bladder cancer. Our findings highlight the importance of reducing financial barriers to care to notably improve health equity and oncologic outcomes for African American patients.

African American patients with bladder cancer present with more advanced disease (1-3) and have higher mortality than non-Hispanic White (White) patients (2-6). Common theories for this disparity comprise multifactorial explanations, including attribution to differences in risk factors or predisposing genetic factors in African American patients alongside individual- and system-originating, covert and overt racism (2,3,5-8). However, African American patients tend to experience increased barriers at several different levels to receiving equitable medical care (9-11), which could lead to delayed diagnosis and poorer outcomes for patients with bladder cancer. Although the determinants are yet to be well understood, this knowledge is critical to developing strategies to eliminate the disparity in bladder cancer outcomes.

We hypothesize that access to health care is a primary determinant of inferior outcomes for African American patients with bladder cancer. To test this hypothesis, we compared outcomes by race for patients receiving care within the predominant hybrid-payer model of the United States in the Surveillance, Epidemiology, and End Results (SEER) database with those of patients receiving care within the Veterans’ Health Administration (VHA), which greatly reduces financial barriers to care.

Methods

Data Source

The SEER cohort included data from the SEER 18 registry, which covers approximately 35% of the US population (12). SEER is the largest comprehensive source of population-based cancer data within the United States. As such, SEER is broadly representative of the US health-care system. The VHA cohort was gathered from VA Informatics and Computing Infrastructure (VINCI), an informatics platform that enables access to patient-level electronic health records and administrative data throughout the national VHA. VINCI includes more than 21 million veterans’ data, thereby capturing an estimated 90% of cancers within the VHA (13–15). The protocol was approved by the San Diego VA Institutional Review Board (#150169). Informed signed consent was waived by the institutional review board given that this is a retrospective analysis with minimal risk to the rights and welfare of participants.

Patient Selection and Covariables

For the SEER and VHA cohorts, White and African American patients with histologically confirmed bladder cancer of any stage or histology diagnosed between 2004 and 2015 (SEER) or 2000 and 2017 (VHA) were included. Patients without complete tumor or survival information were excluded. The final SEER and VHA cohorts included 122 449 and 36 322 patients, respectively (Supplementary Figure 1, available online).

Covariables of interest included age, body mass index, sex, Charlson comorbidity index score, smoking status, marital status, zip code–level median income and education (bachelor’s degree) level based on American Community Survey census data (16), Yost index (a census tract–level socioeconomic status index) (17), insurance status, estimated glomerular filtration rate, histology, TNM (tumor, nodes, metastases) staging, and American Joint Commission on Cancer 8th edition stages when available. In VINCI, creatinine laboratory values within 1 year before diagnosis were obtained for each patient and used to calculate glomerular filtration rate (18). In VINCI, linked administrative data provided International Classification of Diseases-9 and 10 codes for calculation of Charlson scores (19,20). Age at death and cause-specific mortality information were available in both SEER and VHA. All patients were followed up until death or last follow up, with latest possible date in either database of December 31, 2017.

Outcomes and Statistical Analysis

The primary outcome was bladder cancer–specific mortality (BCM). Secondary outcomes were overall survival (OS) and disease stage at diagnosis. The start date for all time-dependent endpoints was the date of diagnosis. Baseline characteristics were compared between racial cohorts within each database using χ² test and Wilcoxon’s rank sum test for categorical and continuous variables, respectively. Comparisons of BCM between racial cohorts were evaluated using a competing risk analysis framework to account for nonbladder cancer mortality. BCM was assessed with cumulative incidence analysis for unadjusted models and with Fine-Gray regression analysis for multivariable models. OS was assessed with Kaplan-Meier analysis for unadjusted models and with Cox proportional hazards analysis for multivariable models. All multivariable models were chosen a priori based on potential mechanism for confounding. Due to differences in availability of aforementioned covariables within each database, primary multivariable analysis involved slightly different models within SEER and VHA that included as many relevant covariables as available. To overcome inherent limitations of comparing different multivariable models in the primary analysis, secondary multivariable analysis was conducted on the pooled cohort of SEER and VHA patients with models that included only common covariables between the databases. In multivariable BCM models, to assess the possibility of unmeasured confounding associated with retrospective cohort studies, we calculated the E-value (evidence for causality) (21). This metric assesses the degree of unmeasured confounding needed to nullify the findings and represents this as a hazard ratio directly comparable with that of other variables in the model (22,23). Finally, direct comparisons between health-care systems were assessed using interaction terms in models including only race, health-care system, and the interaction term. For all survival analysis, hazard ratios (HRs) and 95% confidence intervals (CIs) were reported.

Mediation analysis was conducted to assess the potential mediation effect of disease stage between the variable of question (race) and the primary outcome (BCM) (Supplementary Figure 2, available online). Common in epidemiology for mediation analysis, the proportional hazards model with a counterfactual approach and the difference method, which examines an outcome model with and without the mediator, were selected (24,25). The multiple assumptions to establish mediation (24) and confirm no confounding (25) were satisfied. Relative attenuation of the subdistribution hazard ratios (SHR) for race in each sequential model, within SEER and VHA cohorts separately, was calculated as RA = [SHR_Race − SHR_{Race + X}] ÷ [SHR_Race − 1], where X is any number of covariables added to the univariate model (25).

All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC), with 2-sided P values less than .05 considered statistically significant.

Results

Baseline Characteristics

The SEER cohort included 122 449 patients, of which 7662 (6.3%) were African American and 114 787 (93.7%) were White. The median follow up time was 53 months. African American patients were more likely to be female (34.0% vs 23.7%, P < .001), be unmarried (59.6% vs 39.8%, P < .001), belong to the lowest Yost/socioeconomic status tertile (41.4% vs 29.0%, P < .001), and have nonurothelial histology (8.0% vs 4.0%, P < .001) than White patients (Table 1). The VHA cohort included 36 322 patients, of which 3250 (9.0%) were African American and 33 072 (91.0%) were White. The median follow up time was 48 months. African American patients were more likely than White patients to have any comorbidities (73.9% vs 70.1%, P < .001), be unmarried (60.2% vs 48.0%, P < .001), live in less affluent (69.5% vs 55.1%, P < .001) and less educated (60.8% vs 54.3%, P < .001) zip codes, and have nonurothelial histology (5.0% vs 3.7%, P < .001) (Table 1).

Table 1.

Baseline patient and tumor characteristics of each racial cohort in VHA and SEER datasets

Variable	SEER			VHA
	African American race (n = 7662)	White race (n = 114 787)	Pa	African American race (n = 3250)	White race (n = 33 072)	Pa
Median age (range), y	67 (19-105)	71 (18-114)	<.001	68 (27-99)	70 (23-102)	<.001
Median BMI (range), kg/m2	—	—	—	25.9 (14.1-54.3)	26.4 (14.0-69.6)	<.001
Sex, No. (%)			<.001			.14
Male	5054 (66.0)	87 563 (76.3)		3214 (98.9)	32 790 (99.2)
Female	2608 (34.0)	27 224 (23.7)		36 (1.1)	282 (0.8)
Charlson score, No. (%)			—			<.001
0	—	—		849 (26.1)	9884 (29.9)
1+	—	—		2401 (73.9)	23 188 (70.1)
Smoker at diagnosis, No. (%)			—			<.001
Yes	—	—		1413 (34.5)	13 200 (39.9)
No	—	—		1837 (56.5)	19 872 (60.1)
Married, No. (%)			<.001			<.001
Yes	3095 (40.4)	69 136 (60.2)		1294 (39.8)	17 186 (52.0)
No	4567 (59.6)	45 651 (39.8)		1956 (60.2)	15 886 (48.0)
Median income in zip code, No. (%), US $			—			<.001
<50K	—	—		2260 (69.5)	18 228 (55.1)
≥50K	—	—		990 (30.5)	14 844 (44.9)
Population with bachelor’s in zip code, No. (%)			—			<.001
>15%	—	—		1274 (39.2)	15 105 (45.7)
≤15%	—	—		1976 (60.8)	17 967 (54.3)
Yost Index tertile, No. (%)			<.001			—
High	1825 (23.8)	41 472 (36.1)		—	—
Medium	2664 (34.8)	40 053 (34.9)		—	—
Low	3173 (41.4)	33 262 (29.0)		—	—
Insurance, No. (%)			<.001			—
Yes	5512 (71.9)	83 125 (72.4)		—	—
No	247 (3.2)	1477 (1.3)		—	—
Unknown	1903 (24.9)	30 185 (26.3)		—	—
GFR, No. (%)			—			.19
>50	—	—		2552 (78.5)	25 639 (77.5)
≤50	—	—		698 (21.5)	7433 (22.5)
Histology, No. (%)			<.001			<.001
Urothelial	7046 (92.0)	110 225 (96.0)		3087 (95.0)	31 843 (96.3)
Nonurothelial	616 (8.0)	4562 (4.0)		163 (5.0)	1229 (3.7)
T stage, No. (%)			<.001			<.001
A/IS	3403 (44.3)	64 230 (56.0)		1496 (46.0)	16 906 (51.2)
1	1890 (24.7)	26 308 (22.9)		1021 (31.5)	10 002 (30.2)
2	1377 (18.0)	15 679 (13.7)		544 (16.7)	4673 (14.1)
3	451 (5.9)	4742 (4.1)		68 (2.1)	652 (2.0)
4	541 (7.1)	3828 (3.3)		121 (3.7)	839 (2.5)
Muscle-invasive (T stages 2-4), No. (%)			<.001			<.001
No	5293 (69.0)	90 538 (78.9)		2517 (77.5)	26 908 (81.4)
Yes	2369 (31.0)	24 249 (21.1)		733 (22.5)	6164 (18.6)
N stage, No. (%)			<.001			<.001
0	7082 (92.4)	110 034 (95.8)		3124 (96.1)	32 281 (97.6)
1	261 (3.4)	2278 (2.0)		48 (1.5)	293 (0.9)
2	272 (3.6)	2062 (1.8)		61 (1.9)	404 (1.2)
3	47 (0.6)	4134 (0.4)		17 (0.5)	94 (0.3)
M stage, No. (%)			<.001			<.001
0	7110 (92.8)	110 659 (96.4)		3121 (96.0)	32 244 (97.5)
1	552 (7.2)	4128 (3.6)		129 (4.0)	828 (2.5)
AJCC stage, No. (%)			<.001			<.001
0	3316 (43.3)	63 541 (55.4)		1496 (46.0)	16 906 (51.1)
1	1778 (23.2)	25 599 (22.3)		999 (30.7)	9871 (29.9)
2	1105 (14.4)	13203 (11.5)		457 (14.1)	4047 (12.2)
3	718 (9.4)	6994 (6.1)		169 (5.2)	1420 (4.3)
4	745 (9.7)	5450 (4.7)		129 (4.0)	828 (2.5)

a

χ² test and Wilcoxon’s rank sum test were used to calculate 2-sided P values for categorical and continuous variables, respectively. AJCC = American Joint Commission on Cancer; BMI = body mass index; GFR = glomerular filtration rate; M stage = categorization of metastases; N stage = categorization of involved lymph nodes; SEER = Surveillance, Epidemiology, and End Results; T stage = categorization of primary tumor; VHA = Veterans’ Health Administration.

Stage at Presentation

In both the SEER and VHA cohorts, African American patients were more likely to present with muscle-invasive disease, nodal involvement, and metastases at diagnosis (Table 1). However, the disparity between races was statistically significantly smaller in the VHA for muscle-invasive disease (absolute difference: SEER 9.9% vs VHA 3.9%) and metastases at diagnosis (absolute difference: SEER 3.6% vs VHA 1.5%). The odds ratios of African American vs White race were statistically significantly higher in SEER than in VHA for muscle-invasive disease (OR = 1.67 [95% CI = 1.59 to 1.76] vs OR = 1.27 [95% CI = 1.17 to 1.39], P_interaction < .001) and metastatic disease at diagnosis (OR = 2.08 [95% CI = 1.90 to 2.28] vs OR = 1.61 [95% CI = 1.33 to 1.95], P_interaction = .02).

Bladder Cancer–Specific Mortality

In the SEER cohort, there were 45 387 deaths, of which 22 225 were attributed to bladder cancer. In the VHA cohort, there were 16 237 deaths, of which 6160 were attributed to bladder cancer. The 4-year cumulative incidence of BCM was higher for African American patients in both the SEER cohort (30.0% vs 18.4% for African Americans and Whites, respectively, P < .001; Figure 1, A) and the VHA cohort (20.0% vs 17.0% for African Americans and Whites respectively, P < .001; Figure 1, B), though the absolute differences were statistically significantly lower in the VHA (11.6% vs 3.0%, for SEER and VHA, respectively, P_interaction < . 001).

Figure 1.

Mortality and survival curves in Surveillance, Epidemiology, and End Results (SEER) and Veterans’ Health Administration (VHA) cohorts. A) Cumulative incidence curves for bladder cancer–specific mortality (BCM) in the SEER cohort, stratified by race. B) Cumulative incidence curves for BCM in the VHA cohort, stratified by race. C) Kaplan-Meier curves for overall survival (OS) in the SEER cohort, stratified by race. D) Kaplan-Meier curves for OS in the VHA cohort, stratified by race. Cumulative incidence analysis was used to calculate 2-sided P values for cumulative incidence models. Kaplan-Meier analysis was used to calculate 2-sided P values for OS models.

In multivariable analysis in the SEER cohort, African American race was statistically significantly associated with worse BCM (HR = 1.22, 95% CI = 1.15 to 1.29, P < .001) (Table 2). In multivariable analysis in the VHA cohort, African American race was not associated with a difference in BCM (HR = 0.97, 95% CI = 0.88 to 1.07, P = .54) (Table 2). To assess the robustness of the calculated hazard ratios for race on BCM, we calculated E-values. In SEER, the E-value was 1.56. In VHA, the E-value was 1.17. Compared with the hazard ratios of known important covariables in the respective models (eg, N1 disease in SEER with HR = 1.52), a moderate amount of unmeasured confounding beyond the extensive covariables already corrected for would be necessary to nullify these notable associations between race and BCM. In a sensitivity analysis of only male patients in both datasets, multivariable analyses revealed similar associations between race and BCM (Supplementary Table 1, available online).

Table 2.

Multivariable a priori regressions on BCM in SEER and VHA cohorts

Variable	SEER BCM		VHA BCM
	HR (95% CI)	Pa	HR (95% CI)	Pa
Race
African American	1.22 (1.15 to 1.29)	<.001	0.97 (0.88 to 1.07)	.54
White	1.00 (Referent)		1.00 (Referent)
Age	1.03 (1.03 to 1.03)	<.001	1.01 (1.01 to 1.02)	<.001
BMI	—	—	0.95 (0.94 to 0.95)	<.001
Sex
Male	1.00 (Referent)		1.00 (Referent)
Female	1.08 (1.04 to 1.11)	<.001	1.19 (0.87 to 1.61)	.28
Charlson score
0	—	—	1.00 (Referent)
1+	—	—	0.95 (0.85 to 1.05)	.30
Smoker at diagnosis
No	—	—	1.00 (Referent)
Yes	—	—	0.98 (0.93 to 1.04)	.56
Married
No	1.00 (Referent)		1.00 (Referent)
Yes	0.79 (0.77 to 0.82)	<.001	0.97 (0.91 to 1.02)	.21
Yost Index tertile
High	1.00 (Referent)		—	—
Medium	1.04 (1.01 to 1.08)	.03	—	—
Low	1.11 (1.07 to 1.15)	<.001	—	—
Insurance
Yes	1.00 (Referent)		—	—
No	1.58 (1.41 to 1.77)	<.001	—	—
Unknown	1.06 (1.03 to 1.10)	<.001	—	—
Median income in zip code, US $
<50K	—	—	1.00 (Referent)
≥50K	—	—	1.03 (0.96 to 1.10)	.41
Population with bachelor’s in zip code
≤15%	—	—	1.00 (Referent)
>15%	—	—	0.99 (0.93 to 1.05)	.74
GFR
>50	—	—	1.00 (Referent)
≤50	—	—	1.38 (1.29 to 1.47)	<.001
T stage
A/IS	1.00 (Referent)		1.00 (Referent)
1	2.88 (2.76 to 2.99)	<.001	2.47 (2.30 to 2.65)	<.001
2	7.68 (7.38 to 8.01)	<.001	6.81 (6.32 to 7.33)	<.001
3	8.33 (7.86 to 8.82)	<.001	8.69 (7.58 to 9.95)	<.001
4	10.86 (10.20 to 11.56)	<.001	8.95 (7.83 to 10.23)	<.001
N stage
0	1.00 (Referent)		1.00 (Referent)
1	1.52 (1.42 to 1.63)	<.001	1.24 (1.02 to 1.51)	.04
2	1.78 (1.66 to 1.92)	<.001	1.47 (1.23 to 1.75)	<.001
3	0.87 (0.75 to 1.00)	.06	1.79 (1.27 to 2.53)	.01
M stage
0	1.00 (Referent)		1.00 (Referent)
1	4.35 (4.12 to 4.60)	<.001	3.14 (2.76 to 3.57)	<.001
Histology
Urothelial	1.00 (Referent)		1.00 (Referent)
Nonurothelial	1.68 (1.59 to 1.77)	<.001	1.24 (1.09 to 1.41)	.01

a

Fine-Gray regression analysis was used to calculate 2-sided P values. BCM = bladder cancer–specific mortality; BMI = body mass index; GFR = glomerular filtration rate; M stage = categorization of metastases; N stage = categorization of involved lymph nodes; SEER = Surveillance, Epidemiology, and End Results; T stage = categorization of primary tumor; VHA = Veterans’ Health Administration.

In a secondary analysis of the pooled cohort of SEER and VHA patients, the interaction term between race and database was statistically significant (P < .001) because African American race was associated with worse BCM in SEER (HR = 1.25, 95% CI = 1.18 to 1.32) but was associated with similar BCM in VHA (HR = 0.98, 95% CI = 0.88 to 1.08) (Supplementary Table 2, available online).

Overall Survival

The 4-year OS was lower for African American patients within the SEER cohort (57.5% vs 68.3% for African Americans and Whites, respectively, P < .001; Figure 1, C) but not within the VHA cohort (62.4% vs 65.2% for African Americans and Whites, respectively, P = .10; Figure 1, D). In multivariable analysis in the SEER cohort, African American race was statistically significantly associated with worse OS (HR = 1.26, 95% CI = 1.20 to 1.31, P < .001) (Table 3). In multivariable analysis in the VHA cohort, African American race was not associated with a difference in OS (HR = 0.99, 95% CI = 0.93 to 1.05, P = .76) (Table 3).

Table 3.

Multivariable a priori regressions on OS in SEER and VHA cohorts

Variable	SEER OS		VHA OS
	HR (95% CI)	Pa	HR (95% CI)	Pa
Race
African American	1.26 (1.20 to 1.31)	<.001	0.99 (0.93 to 1.05)	.76
White	1.00 (Referent)		1.00 (Referent)
Age	1.06 (1.06 to 1.06)	<.001	1.05 (1.04 to 1.05)	<.001
BMI	—	—	0.97 (0.96 to 0.97)	<.001
Sex
Male	1.00 (Referent)		1.00 (Referent)
Female	0.85 (0.83 to 0.87)	<.001	0.92 (0.75 to 1.12)	.38
Charlson score
0	—		1.00 (Referent)
1+	—	—	1.19 (1.14 to 1.24)	<.001
Smoker at diagnosis
No	—	—	1.00 (Referent)
Yes	—	—	1.22 (1.17 to 1.26)	<.001
Married
No	1.00 (Referent)		1.00 (Referent)
Yes	0.76 (0.74 to 0.77)	<.001	0.96 (0.93 to 1.00)	.03
Yost Index tertile
High	1.00 (Referent)		—	—
Medium	1.08 (1.05 to 1.10)	<.001	—	—
Low	1.19 (1.16 to 1.22)	<.001	—	—
Insurance
Yes	1.00 (Referent)		—	—
No	1.61 (1.46 to 1.78)	<.001	—	—
Unknown	1.02 (1.00 to 1.04)	.09	—	—
Median income in zip code, US $
<50K	—	—	1.00 (Referent)
≥50K	—	—	0.95 (0.92 to 0.99)	.01
Population with bachelor’s in zip code
≤15%	—	—	1.00 (Referent)
>15%	—	—	0.95 (0.91 to 0.98)	.01
GFR
>50	—	—	1.00 (Referent)
≤50	—	—	1.54 (1.48 to 1.61)	<.001
T stage
A/IS	1.00 (Referent)		1.00 (Referent)
1	1.57 (1.53 to 1.60)	<.001	1.41 (1.36 to 1.46)	<.001
2	3.15 (3.06 to 3.24)	<.001	2.88 (2.74 to 3.03)	<.001
3	3.29 (3.12 to 3.47)	<.001	3.71 (3.27 to 4.22)	<.001
4	4.75 (4.49 to 5.02)	<.001	4.18 (3.70 to 4.73)	<.001
N stage
0	1.00 (Referent)		1.00 (Referent)
1	1.54 (1.42 to 1.66)	<.001	1.26 (1.02 to 1.55)	.03
2	1.79 (1.66 to 1.94)	<.001	1.40 (1.14 to 1.72)	.01
3	0.90 (0.78 to 1.05)	.17	1.56 (1.09 to 2.24)	.02
M stage
0	1.00 (Referent)		1.00 (Referent)
1	4.45 (4.18 to 4.74)	<.001	3.93 (3.43 to 4.50)	<.001
Histology
Urothelial	1.00 (Referent)		1.00 (Referent)
Nonurothelial	1.63 (1.56 to 1.72)	<.001	1.18 (1.06 to 1.31)	.01

a

Cox proportional hazards analysis was used to calculate 2-sided P values. BMI = body mass index; GFR = glomerular filtration rate; M stage = categorization of metastases; N stage = categorization of involved lymph nodes; OS = overall survival; SEER = Surveillance, Epidemiology, and End Results; T stage = categorization of primary tumor; VHA = Veterans’ Health Administration.

In a secondary analysis of the pooled cohort, the interaction term between race and database was statistically significant (P < .001) because African American race was associated with worse OS in SEER (HR = 1.32, 95% CI = 1.26 to 1.37) but was associated with similar OS in VHA (HR = 1.01, 95% CI = 0.95 to 1.08) (Supplementary Table 2, available online).

Mediation Analysis

In SEER, univariate analysis demonstrated African American race to be associated with worse BCM (SHR_Race = 1.71, 95% CI = 1.63 to 1.79, P < .001) (Figure 2). When TNM staging was added to the model, the association of African American race with worse BCM was attenuated by 68% (SHR_Race = 1.23, 95% CI = 1.16 to 1.29, P < .001). The addition of the rest of the patient and tumor variables did not substantially change the attenuation (SHR_Race = 1.22, 95% CI = 1.15 to 1.29, P < .001). This indicates that although disease stage is a major driver of mortality, the BCM disparity by race in SEER is not completely accounted for by this factor or the other available variables. Therefore, disease stage is a partial mediator between race and BCM in SEER.

Figure 2.

Forest plot of subdistribution hazard ratios (SHR) of race in sequential bladder cancer–specific mortality (BCM) models within Surveillance, Epidemiology, and End Results (SEER) and Veterans’ Health Administration (VHA) cohorts. Relative attenuation (RA) calculated per formula [SHR_Race − SHR_{Race + X}] ÷ [SHR_Race − 1], X is any number of covariables added to the model. Error bars depict the 95% confidence interval of SHR. M (stage) = categorization of metastases; N (stage) = categorization of involved lymph nodes; T (stage) = categorization of primary tumor.

In VHA, univariate analysis demonstrated African American race to be associated with worse BCM (SHR_Race = 1.17, 95% CI = 1.07 to 1.27, P < .001) (Figure 2). When TNM staging was added, the association of African American race with BCM was attenuated by 100% and was no longer statistically significant (SHR_Race = 1.00, 95% CI = 0.90 to 1.08, P = .80). The addition of the rest of the patient and tumor variables did not statistically significantly change the attenuation (SHR_Race = 0.97, 95% CI = 0.88 to 1.07, P = .54). This indicates that, as within SEER, disease stage within VHA is a major driver for the disparity in BCM by race observed in unadjusted analysis. However, unlike within SEER, this disparity is completely accounted for and disappears in the final multivariable model. Therefore, disease stage is a complete mediator between race and BCM in VHA.

Discussion

In this study of 2 contrasting health-care models, receipt of care within the VHA was associated with reduced differences in bladder cancer stage at presentation and the absence of survival disparities in multivariable analyses. These results argue against the hypothesis that poorer outcomes for African American patients are primarily driven by biologically more aggressive bladder cancer, although this possibility cannot be excluded based on these data. Instead, our findings highlight the importance of reducing financial and other barriers to health care to notably improve health equity and oncologic outcomes for African American patients.

Painless hematuria is a common presentation for patients with bladder cancer. The ability to see a primary care physician and/or specialists likely influences whether the cancer is diagnosed at an earlier, less advanced stage or at a later, more advanced stage. This diagnostic process is influenced by a multitude of clinical and nonclinical factors, namely financial and social barriers to health care. Within SEER, African American patients were substantially more likely to present with muscle-invasive disease and metastatic disease than their White counterparts. In contrast, the respective differences in stage at presentation within the VHA were less than one-half of that seen in SEER for muscle-invasive disease (9.9% vs 3.9%) and metastatic disease (3.6% vs 1.5%). Although improvements in the equal-access system are important, the remaining difference in stage at presentation merits further consideration with a close look as to what extent potentially biologically more aggressive disease in African Americans influences this outcome. Further research is needed to better understand these residual barriers to early diagnosis in the VHA and implement appropriate measures to achieve health equity. Ultimately, in all health-care systems, among additional reasons for the highlighted disparities, there is a need to raise awareness of bladder cancer and early diagnosis and to foster patient-physician relationships among the African American community (26–28).

The importance of disease stage at presentation is clearly demonstrated by the mediation analysis. Within the VHA, disease stage was a complete mediator because there was no difference in BCM by race after controlling for this variable. This indicates that stage-for-stage bladder cancer is not more aggressive in African American patients. Rather, stage at presentation is the driving feature of poorer outcomes for African American patients. Within SEER, stage at presentation was also the most important factor in explaining poorer outcomes for African American patients, but it was only a partial mediator in this model. Given that BCM was worse for African American patients even after accounting for stage at diagnosis and other sociodemographic and clinical variables, additional sources of poorer outcomes are present and should be investigated within the broader US health-care system. Specifically, we hypothesize that effective and timely treatments for bladder cancer are not used equally for African American and White patients. However, treatment variables were not assessed in this analysis due to the broad spectrum of disease stages and the difficulty in accurately ascertaining treatments in SEER (29). Further research examining disparities in bladder cancer treatment by race are warranted in light of our findings and previous literature (6,8,30,31).

To our knowledge, this is the first study to directly compare racial equity for patients with bladder cancer within contrasting health-care models. Whether it is the most comprehensive multivariable models within each database (primary analysis) or an identical multivariable model for both databases (secondary analysis), the results of our analyses clearly demonstrate inferior outcomes for African American patients in SEER but not in VHA. Our results are consistent with prior work from the SEER database and on bladder cancer in the United States (1–3,5–8). In 2016, a small study including 154 Black patients from the Department of Defense’s equal-access health-care system demonstrated similar patterns, with diminished disparity in stage at presentation and no differences in survival on multivariable analysis (32). In 2019, Cole et al. (31) suggested similar concepts with their findings that access related variables explained more of the race-based disparities in bladder cancer outcomes than tumor-related variables within a nationally representative cancer database. Of note, epidemiologic analyses of SEER have shown a lower incidence of bladder cancer in African American patients despite the higher disease stage for those who are diagnosed (1,3). This reinforces our hypothesis of substantial underdiagnosis of early-stage bladder cancer in African American patients and reemphasizes the need to raise awareness of this disease, especially within the African American community. Our study builds on previous literature to stand as the first, to our knowledge, comprehensive analysis of bladder cancer outcomes by race within 2 of the largest oncologic databases from contrasting health-care models.

Several limitations of the study are worth noting. First, the retrospective nature of the study may lead to biases that affect these results. We attempted to limit the impact of these biases through comprehensive modeling, including as many potentially confounding covariables as available in the primary analysis, though it is impossible to eliminate all biases. Second, although common in epidemiology for mediation analysis, the theoretical basis of the employed difference method was established primarily for linear models with continuous outcomes. In light of our results, additional rigorous mediation analyses are warranted to further elucidate the pathways of interest between race and bladder cancer outcomes. Third, although there are increased risk factors within the veteran population (33), we do not believe this limits the study’s external validity because it is unlikely that equal-access health care would lead to a differential impact on race-based disparities in bladder cancer within the veteran population compared with the general population. Finally, we recognize that our framework of race stands as a surrogate for the underlying structural sociodemographic features, which have been directly shown to be predictors of mortality in this study and in prior studies. Although we focused only on White and African American patients in order to have well-defined cohorts to conduct a large retrospective analysis, we hope our work will be expanded to other races and ethnicities to investigate disparities in those populations as well.

It is important to note that although the VHA substantially reduces financial barriers to medical care, there are additional barriers to achieving health equity. Attention should be drawn to improve facets including cultural competence or unconscious bias among providers, social stigma and health literacy and advocacy, and patient- and community-centered lack of trust in the health-care system (34). The prevalence of these factors and others has led to the appropriate questioning of the term “equal access” to describe the VHA. Consequently, although we use the term “equal access” for ease of understanding, the VHA likely does not lead to completely equitable health use and outcomes but rather represents an appreciable step towards that possibility.

Within the United States’ predominant hybrid-payer health-care model, African American patients with bladder cancer suffer inferior outcomes compared with White patients. In contrast, receiving medical care in an equal-access health-care system was associated with reduced disparities in outcomes for African American patients with bladder cancer. These findings highlight the importance of reducing financial and other barriers to care to notably improve health equity and oncologic outcomes for African American patients.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Notes

Role of the funder: Not applicable.

Disclosures: Dr Abhishek Kumar reports ownership stake in Sympto Health. Dr A. Karim Kader reports personal fees from Stratify Genomics, personal fees from Pellficure, intellectual property interests in prostate cancer genetics, outside the submitted work. Dr Rana R. McKay reports grants from Bayer, Pfizer, Tempus, personal fees from AstraZeneca, Bayer, Bristol Myers Squibb, Calithera, Exelixis, Janssen, Merck, Novartis, Pfizer, Sanofi, Tempus, personal fees from Dendreon, Vividion, personal fees from Caris, outside the submitted work. Dr Tyler F. Stewart has served as a paid consultant for Seattle Genetics, outside the submitted work. All remaining authors have declared no conflicts of interest.

Author contributions: All authors meet the ICMJE’s 4 criteria for authorship credit. Specific contributions are as follows. Study conception and design: NVK, AK, AKK, RRM, TFS, BSR. Data gathering, statistical analysis, interpretation of results: NVK, AK, EQ, ASQ, RSV, VN, TFS, BSR. Manuscript writing and preparation: NVK, AK, EQ, ASQ, RSV, AKK, RRM, TFS, BSR. All authors approved the final version of the manuscript.

Prior presentations: A version of this study was accepted and delivered as an oral presentation at American Society of Clinical Oncology (ASCO) 2021 Genitourinary Cancers Symposium on February 12, 2021.

Data Availability

SEER data used in this study are publicly available. VINCI data used in this study are available from the US Veterans Affairs Administration, but restrictions apply to the availability of these data, which were used under appropriate permissions for the current study and thus are not publicly available. Data are available, however, from the corresponding author upon reasonable request and with permission of these authorities.