Abstract
Rationale & Objective:
Community racial composition has been shown to be associated with mortality in patients receiving maintenenance dialysis. It is unclear whether living in communities with predominantly Black residents is also associated with risk for hospitalization among patients receiving hemodialysis.
Study Design:
Retrospective analysis of prospectively collected data from a cohort of patients receiving hemodialysis.
Setting & Participants:
4,567 patients treated in 154 dialysis facilities located in 127 unique zip codes and enrolled in US Dialysis Outcomes and Practice Patterns Study (DOPPS) phases 4 to 5 (2010–2015).
Exposure:
Tertile of percentage of Black residents within zip code of patients’ dialysis facility, defined through a link to the American Community Survey.
Outcome:
Rate of hospitalizations during the study period.
Analytic Approach:
Associations of patient-, facility-, and community-level variables with community’s percentage of Black residents were assessed using analysis of variance, Kruskal-Wallis, or χ2/Fisher exact tests. Negative binomial regression was used to estimate the incidence rate ratio for hospitalizations between these communities, with and without adjustment for potential confounding variables.
Results:
Mean age of study patients was 62.7 years. 53% were White, 27% were Black, and 45% were women. Median and threshold percentages of Black residents in zip codes in which dialysis facilities were located were 34.2% and ≥14.4% for tertile 3 and 1.0% and ≤1.8% for tertile 1, respectively. Compared with those in tertile 1 facilities, patients in tertile 3 facilities were more likely to be younger, be Black, live in urban communities with lower socioeconomic status, have a catheter as vascular access, and have fewer comorbid conditions. Patients dialyzing in communities with the highest tertile of Black residents experienced a higher adjusted rate of hospitalization (adjusted incidence rate ratio, 1.32; 95% CI, 1.12–1.56) compared with those treated in communities within the lowest tertile.
Limitations:
Potential residual confounding.
Conclusions:
The risk for hospitalization for patients receiving maintenance dialysis is higher among those treated in communities with a higher percentage of Black residents after adjustment for dialysis care, patient demographics, and comorbid conditions. Understanding the cause of this association should be a priority of future investigation.
Hospitalizations in patients receiving kidney replacement therapy account for up to 33% of Medicare expenditures for this group of patients and contribute to high morbidity and mortality.1 According to the US Renal Data System (USRDS), risk factors for hospitalization include female sex, Black race, and young age.1–4 However, hospitalization rates vary dramatically by health service area, and observed risks for hospitalizations are not consistent across different types of communities, as defined by racial composition and rural/urban continuum, in the United States.1,5–7
Observational studies have revealed disparities in predialysis care, quality of dialysis care, and access to transplantation in patients living in Black communities.6 These studies include national databases without patient-level clinical parameters, limiting inferences across communities.8 Our group previously showed that risk factors for hospitalization defined by the USRDS did not apply to an inner-city minority population.9 Hospitalization risk has been shown to be higher in Black patients receiving hemodialysis, but it is unclear whether this higher risk for hospitalization relates to the risks inherent in living in communities with a high percentage of Black residents.6
Dialysis quality disparities shown in previous studies were before the implementation in 2010 of the Centers for Medicare & Medicaid Services (CMS) Quality Incentive Program (QIP), and less is known about the impact of dialysis facility-level outcomes since the standardization of dialysis dose and bundling of dialysis medications by CMS as part of the QIP. We used the Dialysis Outcomes and Practice Patterns Study (DOPPS), with its collection of granular data for comorbid conditions, adherence, clinical and laboratory results, and medications, during a period after adoption of the QIP to better evaluate the association of community racial composition (as defined by community’s percentage of Black residents) and hospitalization. We hypothesized that patients receiving hemodialysis being treated at facilities in communities with more Black residents have a higher risk for hospitalization and that this association partially accounts for the higher risk for hospitalization attributed to Black race by the USRDS.
Methods
We obtained Institutional Review Board approval for this study from the Albert Einstein College of Medicine. Individual-level informed consent was not deemed necessary because this study is a secondary analysis of a deidentified DOPPS data set.
Design, Setting, & Participants
In this retrospective cohort analysis, we used patient-level data from the prospectively collected US sample of the DOPPS database during phases 4 and 5 (2010–2015).10,11 The DOPPS randomly selected 20 to 40 incident and prevalent patients per facility within a nationally representative random sample of dialysis facilities in the United States.11–13 After excluding 6 facilities that did not follow up patients for hospitalization, the cohort included 4,567 patients from 154 DOPPS dialysis facilities (Fig 1). Race and ethnicity were self-reported.
Figure 1.
Study cohort.
Outcome Variable
We examined the hospitalization incidence rate, defined as the number of hospitalizations per participant per year during enrollment in DOPPS. Follow-up data for death or disenrollment were available through June 2015. Hospitalizations were recorded prospectively by the DOPPS coordinator during follow-up visits.
Exposure Variable
The exposure variable was Census Bureau-derived percentage of residents who identified as Black (presented as tertiles) in communities defined by dialysis facility zip codes. For secondary analyses, the percentage of White and Hispanic residents was used as a continuous variable to test the community composition of each group with outcome. Individual self-reported race and ethnicity data were gathered by DOPPS and used to define 4 groups: (1) non-Hispanic Black, (2) non-Hispanic White, (3) Hispanic, and (4) other. The dialysis facility data (N = 127) were linked to the American Community Survey data gathered during 2011 to 2015 (urban/rural data gathered in 2010). The investigators were blinded to zip codes. The exposure was characterized into tertiles, with cut-points at 1.8% Black residents for tertile 1 (median, 1.0%), 14.4% for tertile 2 (median, 5.5%), and 92.6% (median, 34.2%) Black residents for tertile 3 communities. These tertiles are referred to as “community racial composition” in this article.
Other Measures
Information for patient demographics, comorbid condition history, laboratory values, dialysis treatment parameters and adherence, and medication prescriptions was abstracted from medical records at DOPPS enrollment. Clinical comorbid conditions, cause of kidney failure, and other case-mix variables described in Table 1 were also abstracted from medical records at enrollment. Completion rate of quality-of-life questionnaires was included as a surrogate for patient engagement in facility practice and health care. We used a modified Charlson comorbidity score previously reported for analyzing DOPPS data.14 DOPPS data included the profit status of and number of patients at each facility at the time of enrollment in the cohort. Additional community variables were obtained from the American Community Survey based on dialysis facility zip code and included percent rurality, percent of family incomes under the poverty line, percent of households led by a married couple, percent of households led by a single woman, percent of households with higher education (at least 1 member with a Bachelor’s degree or higher), percent of households with a person younger than 65 years who has a disability, percent of households with a recent immigrant (entered after 2010), percent of households with a computer, and percent of households with an active internet subscription.
Table 1.
Sociodemographic and Clinical Characteristics by Community Racial Composition Tertile, DOPPS 2010–2015
| Tertile 1 (n = 1,619) | Tertile 2 (n = 1,473) | Tertile 3 (n = 1,475) | P | |
|---|---|---|---|---|
| Black residents in zip codes of patients’ dialysis facilitiesa | 1.0% (0.0%−1.8%) | 5.5% (>1.8%−14.4%) | 34.2% (>14.4%−92.6%) | |
| Patient-Level Variables | ||||
| Age, yb | 64.1 ± 15.2 | 63.2 ± 15.0 | 60.6 ± 14.7 | <0.001 |
| Race/ethnicity | <0.001 | |||
| Non-Hispanic White | 1,069 (66.0%) | 899 (61.0%) | 419 (28.4%) | |
| Non-Hispanic Black | 57 (3.5%) | 322 (21.9%) | 860 (58.3%) | |
| Hispanic | 122 (7.5%) | 195 (13.2%) | 113 (7.7%) | |
| Asian | 124 (7.7%) | 31 (2.1%) | 26 (1.8%) | |
| Other | 247 (15.3%) | 26 (1.8%) | 57 (3.9%) | |
| Male sex | 887 (53.8%) | 818 (55.5%) | 815 (55.2%) | 0.9 |
| Vintage, y | 3.1 [1.4–5.4] | 2.9 [1.3–5.2] | 3.2 [1.6–5.7] | <0.001 |
| Vascular access (n = 4,381) | <0.001 | |||
| Fistula | 1,033 (66.4%) | 859 (61.9%) | 793 (55.2%) | |
| Graft | 257 (16.4%) | 241 (17.4%) | 373 (26.0%) | |
| Catheter | 262 (16.8%) | 273 (19.7%) | 267 (18.6%) | |
| Other | 5 (0.3%) | 15 (1.1%) | 3 (0.2%) | |
| Education (n = 1,775) | <0.001 | |||
| <12 y | 141 (18.8%) | 120 (19.7%) | 115 (27.6%) | |
| High school graduate | 265 (35.3%) | 182 (29.9%) | 122 (29.3%) | |
| Some college or college graduate | 208 (27.7%) | 176 (29.0%) | 76 (18.3%) | |
| Unknown | 137 (18.2%) | 130 (21.4%) | 103 (24.7%) | |
| Alcohol use, including suspected (n = 4,401) | 31 (2.0%) | 16 (1.2%) | 23 (1.6%) | 0.2 |
| Substance use, including suspected (n = 4,407) | 24 (1.5%) | 18 (1.3%) | 37 (2.6%) | 0.02 |
| Smoking (n = 1,344) | 0.001 | |||
| Active | 92 (15.0%) | 39 (9.8%) | 54 (16.2%) | |
| Recently stopped | 30 (4.9%) | 18 (4.6%) | 16 (4.8%) | |
| Stopped >1 y ago | 190 (30.9%) | 139 (35.1%) | 70 (21.0%) | |
| Never smoker | 303 (49.3%) | 200 (50.5%) | 193 (58.0%) | |
| Patient Comorbid Conditions | ||||
| Coronary artery disease (n = 4,417) | <0.001 | |||
| Yes | 633 (39.7%) | 601 (43.2%) | 515 (36.0%) | |
| No | 963 (60.3%) | 790 (56.8%) | 915 (64.0%) | |
| Other cardiovascular disease (n = 4,420) | <0.001 | |||
| Yes | 367 (22.9%) | 342 (24.6%) | 259 (18.1%) | |
| No | 1,234 (77.1%) | 1,049 (75.4%) | 1,169 (81.9%) | |
| Cerebrovascular disease | 0.6 | |||
| Yes | 194 (12.0%) | 191 (13.0%) | 174(11.8%) | |
| No | 1,425 (88.0%) | 1,282 (87.0%) | 1,301 (88.2%) | |
| Diabetes (n = 4,410) | 0.002 | |||
| Yes | 1,032 (63.7%) | 883 (60.0%) | 852 (57.8%) | |
| No | 587 (36.3%) | 590 (40.0%) | 623 (42.2%) | |
| Cellulitis/gangrene (n = 4,419) | 0.9 | |||
| Yes | 168 (10.5%) | 146 (10.5%) | 146 (10.2%) | |
| No | 1,433 (89.5%) | 1,244 (89.5%) | 1,282 (89.8%) | |
| History of gastrointestinal bleed (n = 4,404) | 0.7 | |||
| Yes | 72 (4.5%) | 68 (4.9%) | 60 (4.2%) | |
| No | 1,519 (95.5%) | 1,319 (95.1%) | 1,366 (95.8%) | |
| HIV status (n = 3,415) | <0.001 | |||
| Positive | 5 (0.4%) | 10 (1.0%) | 44 (3.5%) | |
| Negative | 1,108 (99.6%) | 1,019 (99.0%) | 1,229 (96.5%) | |
| Lung disease (n = 4,422) | 0.002 | |||
| Yes | 249 (15.5%) | 231 (16.6%) | 174 (12.2%) | |
| No | 1,352 (84.4%) | 1,160 (83.4%) | 1,256 (87.8%) | |
| Neurologic disorder (n = 4,420) | 0.3 | |||
| Yes | 146 (9.1%) | 148 (10.6%) | 131 (9.2%) | |
| No | 1,455 (90.9%) | 1,245 (89.4%) | 1,295 (90.8%) | |
| Cancer (n = 4,406) | 0.05 | |||
| Yes | 163 (10.2%) | 137 (9.9%) | 111 (7.8%) | |
| No | 1,429 (89.8%) | 1,251 (90.1%) | 1,315 (92.2%) | |
| Heart failure (n = 4,413) | <0.001 | |||
| Yes | 616 (38.6%) | 500 (36.0%) | 618 (43.3%) | |
| No | 982 (61.4%) | 888 (64.0%) | 809 (56.7%) | |
| Hypertension (n = 4,416) | 0.02 | |||
| Yes | 1,383 (86.6%) | 1,160 (83.6%) | 1,244 (87.0%) | |
| No | 215 (13.4%) | 227 (16.4%) | 187 (13.1%) | |
| Psychiatric disorder (n = 4,418) | <0.001 | |||
| Yes | 377 (23.3%) | 293 (19.9%) | 223 (15.1%) | |
| No | 1,242 (76.7%) | 1,180 (80.1%) | 1,252 (84.9%) | |
| Comorbidity score (Charlson) (n = 4,009) | 5.10 ± 1.69 | 5.13 ± 1.84 | 4.81 ± 1.74 | <0.001 |
| Cause of kidney disease (n = 4,401) | <0.001 | |||
| Diabetes | 767 (49.7%) | 578 (40.8%) | 581 (40.4%) | |
| Hypertension | 331 (21.4%) | 442 (31.2%) | 573 (39.8%) | |
| Glomerulonephritis | 155 (10.0%) | 151 (10.6%) | 118 (8.2%) | |
| Other | 291 (18.8%) | 247 (17.4%) | 167(11.6%) | |
| No. of kidney disease medications prescribed (n = 4,344) | <0.001 | |||
| 0 | 53 (3.5%) | 39 (2.7%) | 19 (1.4%) | |
| 1–2 | 779 (50.9%) | 584 (41.1%) | 572 (41.1%) | |
| >3 | 698 (45.6%) | 799 (56.2%) | 801 (57.5%) | |
| Types of kidney disease medications prescribed | ||||
| IV iron (n = 4,328) | 798 (53.1%) | 773 (55.6%) | 759 (55.0%) | 0.4 |
| Erythropoiesis-stimulating agents (n = 4,328) | 1,229 (82.7%) | 1,154 (82.8%) | 1,136 (81.8%) | 0.7 |
| Vitamin D (n = 4,316) | 904 (59.7%) | 987 (69.8%) | 1,026 (73.9%) | <0.001 |
| Phosphorus binders (n = 3,954) | 1,132 (85.8%) | 1,164 (87.7%) | 1,155 (88.4%) | 0.1 |
| Calcimimetics (n = 3,978) | 296 (22.2%) | 301 (22.3%) | 309 (23.9%) | 0.5 |
| Completed QoL survey | 789 (48.8%) | 707 (48.0%) | 523 (35.5%) | <0.001 |
| Albumin, g/dL (n = 4,449) | 3.76 ± 0.42 | 3.86 ± 0.39 | 3.84 ± 0.41 | <0.002 |
| BMI, kg/m2 (n = 4,304) | 28.9 ± 7.2 | 29.0 ± 7.3 | 28.5 ± 6.9 | 0.2 |
| Creatinine, mg/dL (n = 4,382) | 7.7 ± 2.7 | 8.2 ± 3.0 | 9.03 ± 3.2 | <0.001 |
| Hemoglobin, g/dL (n = 4,453) | 11.4 ± 1.21 | 11.5 ± 1.19 | 11.4 ± 1.21 | 0.5 |
| Phosphorus, mg/dL (n = 4,448) | 5.33 ± 1.62 | 5.31 ± 1.61 | 5.36 ± 1.66 | 0.7 |
| URR (n = 4,217) | 73.5 ± 7.6 | 73.6 ± 7.3 | 73.7 ± 7.2 | 0.9 |
| Dialysis duration, min/wk (n = 4,285) | 638 ± 98.2 | 652 ± 98.6 | 658 ± 92.7 | <0.001 |
| No. of missed treatments in 30 d before enrollment (n = 3,719) | 0.2 | |||
| 0 | 1,208 (89.6%) | 1,025 (88.1%) | 1,053 (87.2%) | |
| ≥1 | 140 (10.4%) | 139 (11.9%) | 154 (12.8%) | |
| No. of shortened treatments in 30 d before enrollment (n = 4,446) | <0.001 | |||
| 0 | 1,245 (79.2%) | 1,097 (77.0%) | 1,013 (70.0%) | |
| ≥1 | 326 (20.8%) | 328 (23.0%) | 437 (30.0%) | |
| Amount of fluid removedc (n = 4,320) | 3.2% ± 1.6% | 3.0% ± 1.4% | 3.2% ± 1.6% | 0.001 |
| Primary payer (n = 4,498) | <0.001 | |||
| Medicare | 1,281 (80.6%) | 1,051 (73.1%) | 1,154 (78.4%) | |
| Medicaid | 62 (3.9%) | 86 (6.0%) | 75 (5.1%) | |
| Private | 216 (13.6%) | 272 (18.9%) | 207 (14.1%) | |
| Veterans Affairs | 27 (1.7%) | 21 (1.5%) | 27 (1.8%) | |
| No insurance | 4 (0.2%) | 7 (0.5%) | 8 (0.5%) | |
| Facility-Level Variables | ||||
| For-profit status | 982 (60.6%) | 1,208 (82.0%) | 1,341 (90.9%) | <0.001 |
| Facility size (no. of patients) | 55 [33–104] | 76 [50–106] | 81 [52–114] | <0.001 |
| Community-Level Variables | ||||
| Rurality (% rural in area surrounding zip code) | 22.2% [0%−51.1%] | 2.6% [0%−22.8%] | 0% [0%−57.5%] | <0.001 |
| Poverty (% living under the poverty line in area surrounding zip code) | 11.8% [8.4%−14.4%] | 11.5% [7.5%−14.9%] | 17.1% [14.6%−22.1%] | <0.001 |
| Households led by a married couple | 63.0% [58.5%−66.1%] | 64.5% [61.0%−70.7%] | 65.2% [58.0%−68.4%] | <0.001 |
| Families led by single woman | 11.8% [9.7%−14.2%] | 11.3% [9.3%−15.0%] | 15.7% [12.7%−23.8%] | <0.001 |
| Households with higher education | 24.6% [17.5%−29.6%] | 28.2% [23.5%−39.7%] | 19.7% [14.8%−30.5%] | <0.001 |
| Households with a person < 65 y who has a disability | 11.7% [8.4%−16.0%] | 10.1% [7.3%−12.1%] | 13.3% [10.6%−16.4%] | <0.001 |
| Households with a recent immigrant | 1.7% [0.6%−5.1%] | 6.3% [2.4%−11.8%] | 4.0% [1.3%−18.9%] | <0.001 |
| Households with a computer | 85.3% [82.6%−87.8%] | 87.7% [85.4%−93.3%] | 84.1% [79.6%−88.1%] | <0.001 |
| Household with an active internet subscription | 77.6% [73.4%−80.8%] | 80.3% [77.6%−86.5%] | 73.6% [63.6%−78.5%] | <0.001 |
Note: N = 4,567. Unless otherwise indicated, values for continuous variables are given as mean ± standard deviation for normally distributed variables and median [IQR] for non-normally distributed variables; values for categorical variables given as as number (percent).
Abbreviations: BMI, body mass index; DOPPS, Dialysis Outcomes and Practice Patterns Study; HIV, human immunodeficiency virus; IQR, interquartile range; IV, intravenous; QoL, quality of life; URR, urea reduction ratio.
Given as median (range). Across all zip codes of sampled dialysis facilities, the median proportion of Black residents was 5.5% [IQR, 1.4%−21.3%].
Complete data were available from the 4,567 patients for sex, vintage, access, survey completion rates, outcome, facility characteristics, and American Community Survey-derived variables.
As percent of postdialysis weight in second session of week before enrollment.
Statistical Methods
We used STATA, version 14.0 (StataCorp LLC), for all analyses. We tested the associations of patient-, facility-, and community-level variables with the community’s percentage of Black residents, characterized as tertiles, using 1-way analysis of variance, Kruskal Wallis, or χ2/Fisher exact tests. The variables in Table 1 with excessive missing data (>1,000/4,567) that were not imputed or included in the final model were patients’ living location (nursing home vs home) and self-reported educational attainment.15 We used negative binomial regression adjusted to duration of enrollment in DOPPS (with consideration of percent who died during follow-up) to estimate the incidence rate ratio (IRR) for hospitalizations between these communities, with and without adjustment for potential confounding variables.16 We used an approach similar to linear mixed-effects modeling. Instead of using a facility-specific random effect to account for within-facility correction, clustered covariance (or clustered standard errors) were used to adjust inference following estimation of negative binomial model estimated by maximum likelihood to account for within-facility correlation in all models.17,18 The sandwich estimator was a robust model for estimating the standard error in case of model mis-specifications. Model building consisted of adding a priori-built categories to the unadjusted model in a stepwise fashion to test the impact of each category on the association of community racial composition with hospitalization rate. We grouped variables that tested as significant in bivariate testing (Table 1) and according to previously cited meaningful categories in their role as possible confounders. These categories and their associated model designation were as follows:
Model 1: individual demographic variables (age, sex, and DOPPS year of enrollment);
Model 2: model 1 plus individual clinical variables (history of diabetes, Charlson score, primary cause of kidney disease, dialysis vintage, body mass index, hemoglobin level, and serum albumin level);
Model 3: model 2 plus quality of provider/facility care variables (individual-level data for type of dialysis access, prescribed dialysis duration, number of prescribed kidney disease medications, urea reduction ratio [URR], prescription for erythropoiesis-stimulating agent, diagnosis of heart failure [surrogate for fluid overload], and type of health insurance; and facility-level data for total number of prevalent patients and profit status);
Model 4: model 3 plus individual-level patient behavior variables (dialysis adherence [number of shortened or missed dialysis sessions in the prior 30 days], interdialytic weight gain [ultrafiltration rate as percent of postdialysis weight in session 2 of the prior week of hemodialysis], substance use [yes/no], and quality-of-life questionnaire completion [yes/no]);
Model 5: model 4 plus community-level variables (percent rurality, percent of family incomes under the poverty line, percent of households led by married couple, percent of households led by a single woman, percent of households with higher education, percent of households with a person younger than 65 years who has a disability, percent of households with recent immigrant, and percent of households with active internet subscription);
Model 6: model 5 plus individual race/ethnicity.
Because USRDS data identified Black race, younger age (dichotomized at 45 years), and female sex as associated with hospitalization risk, we did 2-way effect modification testing between each of these variables and tertile of community’s percentage of Black residents with respect to hospitalization risk.19 We also tested for effect modification by diabetes status, dialysis facility profit status, Charlson score (tertiles), and dialysis facility size (dichotomized number of patients).
Bias and Missing Data
Most variables with missing data had <5% missing data except for number of missed dialysis sessions (18.6%), Charlson score (12.2%), and URR (7.7%). To account for missing data, we assumed that all missing data were at random and used a multiple imputation technique with chained predictive analytics (10 imputations).20 The observed data that the missing at random assumption was based on included facility identifier, sex, zip code-based American Community Survey characteristics (poverty and percentage of Black residents in community), diabetes status, date of enrollment in DOPPS, number of hospitalizations, and proportion with completed quality-of-life surveys. We did a sensitivity analysis to address bias introduced from missing data. We limited the study sample to those with complete data only (n = 3,096) and the imputed data set (n = 4,567). We then compared the IRR for hospitalizations between the data sets and found similar results.
Secondary and Sensitivity Analysis
Three analyses were done. A secondary analysis derived crude death rates by tertile, and with the use of Cox proportional hazards models, we calculated hazard ratios (HRs) for mortality by tertile. Sensitivity analysis 1 evaluated the effect of community racial composition on the association between individual race/ethnicity and hospitalization rate in negative binomial regression models with and without community racial composition (community’s percentage of Black residents) included as adjusters. For sensitivity analysis 2, we repeated the primary analyses of community racial composition and hospitalization rate for percent White (as a continuous variable) and percent Hispanic (as a continuous variable) separately to understand the association of other formulations of community composition (by percentage of White or Hispanic residents rather than percentage of Black residents) with hospitalization rate.
Results
Participants and Community Characteristics (Descriptive Data)
This cohort included 4,567 participants receiving hemodialysis recruited to DOPPS phases 4 and 5 (Fig 1). Mean follow-up time was 1.35 years. About half (53%) the patients were White, 27% were Black, 10% were Hispanic, and 4% were Asian (Table 1). Forty-five percent of patients were female, and 13% were 45 years or younger. Table 1 shows baseline characteristics of participants based on tertile of percentage of Black residents in zip codes of sampled dialysis facilities, with median percentages for tertiles 1, 2, and 3 of 1.0% (range, 0%–1.8%), 5.5% (range, >1.8%–14.4%), and 34.2% (range, >14.4%–92.6%), respectively. Most tertile 3 facilities were in poor urban areas, with a higher percentage of households headed by a single woman and lower percentages of households with higher education and with an active internet subscription as compared wth tertiles 1 and 2 (Table 1). These facilities were also for-profit more frequently and had a higher patient census (Table 1). Compared with patients in tertile 1 facilities, those treated in facilities in tertile 3 were younger, were more commonly Black, had a lower level of educational attainment, more commonly lived in poverty, more commonly had an arteriovenous graft or catheter rather than an arteriovenous fistula, and had a lower mean Charlson score. Fewer patients receiving treatments at tertile 3 dialysis facilities had a diagnosis of psychiatric disorder, coronary or other cardiac/cerebrovascular diseases, diabetes, or lung disease, whereas a higher proportion had hypertension, heart failure, and human immunodeficiency virus infection and reported substance use. In addition, there were clinically relevant differences in kidney failure cause: Black patients receiving dialysis in zip codes with the lowest tertile of percentage of Black residents had as their cause of kidney failure a diagnosis of glomerulonephritis more commonly and a diagnosis of hypertension less commonly than Black patients receiving dialysis in zip codes within the highest tertile of Black residents (12.3% vs 8.8% for glomerulonephritis; 33.3% vs 44.0% for hypertension; P = 0.07). Prescribed dialysis duration, URRs, and number of prescribed bundled kidney disease medications (specifically vitamin D analogues) were higher in patients receiving treatment in tertile 3 facilities. Serum creatinine and albumin levels were higher in patients treated by tertile 3 compared with tertile 1 facilities; serum hemoglobin and phosphorus levels did not differ between tertiles (Table 1). Although missed treatment and ultrafiltration rates did not differ between tertiles, more patients in tertile 3 shortened their treatment time than those in tertiles 1 and 2, and more patients completed quality-of-life questionnaires in the facilities located in the tertile 1 as compared to tertile 3 (Table 1).
Outcomes and Main Results
The overall hospitalization incidence rate among all participants was 1.19 per person year. Crude hospitalization rates (per patient-year) were 1.05 for tertile 1, 1.20 for tertile 2, and 1.37 for tertile 3. Compared with the lowest tertile, the unadjusted model IRR for hospitalizations was 1.11 (95% confidence interval [CI], 0.96–1.30) in tertile 2 and 1.28 (95% CI, 1.08–1.51) in tertile 3.
As shown in Table 2, this association remained strong after adjustment for categories of variables (models 1–5) and in the final model (model 6). When adjusting for provider quality of care, as in model 3, the IRR for hospitalizations was 1.23 (95% CI, 1.06–1.44) in tertile 2 and 1.30 (95% CI, 1.10–1.55) in tertile 3. Similarly when adjusting for patient behavior (model 4; IRRs of 1.25 [95% CI, 1.07–1.47] for tertile 2 and 1.30 [95% CI, 1.10–1.54] for tertile 3) and when adjusting for individual race/ethnicity in the full model (model 6; IRRs of 1.22 [95% CI, 1.03–1.46] for tertile 2 and 1.32 [95% CI, 1.12–1.56] for tertile 3), the association remained robust, suggesting that the association of hospitalizations and racial composition of community is independent of individual patient attributes and dialysis quality metrics.
Table 2.
IRRs of Hospitalization Count With Percentage of Black Residents in the Community
| Community Racial Composition | Unadjusted IRR (95% CI) | Model 1: Adj IRR (95% CI) | Model 2: Adj IRR (95% CI) | Model 3: Adj IRR (95% CI) | Model 4: Adj IRR (95% CI) | Model 5: Adj IRR (95% CI) | Model 6: Adj IRR (95% CI) |
|---|---|---|---|---|---|---|---|
| Tertile 1 (range: 0%−1.8%) | 1.00 (reference) | 1.00 (reference) | 1.00 (reference) | 1.00 (reference) | 1.00 (reference) | 1.00 (reference) | 1.00 (reference) |
| Tertile 2 (range: > 1.8%−14.4%) | 1.11 (0.96–1.30) | 1.12 (0.96–1.31) | 1.20 (1.04–1.40) | 1.23 (1.06–1.44) | 1.25 (1.07–1.47) | 1.21 (1.02–1.45) | 1.22 (1.031.46) |
| Tertile 3 (range: > 14.4%−92.6%) | 1.28 (1.08–1.51) | 1.29 (1.09–1.53) | 1.34 (1.14–1.58) | 1.30 (1.10–1.55) | 1.30 (1.10–1.54) | 1.27 (1.08–1.50) | 1.32 (1.121.56) |
| P for trend | 0.003 | 0.003 | <0.001 | 0.001 | 0.001 | 0.001 | <0.001 |
Note: N = 4,567. Model 1 adjusted for age and sex and Dialysis Outcomes and Practice Patterns Study year. Model 2 adjusted for model 1 plus history of diabetes, Charlson score, primary cause of kidney disease, dialysis vintage, body mass index, hemoglobin level, and albumin level. Model 3 adjusted for model 2 plus access, dialysis duration, number of kidney disease medications, urea reduction ratio, facility census, facility profit status, insurance, use of erythropoiesis-stimulating agent, and diagnosis of heart failure. Model 4 adjusted for model 3 plus patient behavior (dialysis adherence, substance use, interdialytic weight gain, and rate of questionnaire completion). Model 5 adjusted for model 4 plus community variables (percent rurality, percent of family income under the poverty line, percent of households led by married couple, percent of households led by a single woman, percent of households with higher education, percent of households with a person younger than 65 years who has a disability, percent of households with recent immigrant, and percent of households with active internet subscription) but without individual race and ethnicity. Model 6 adjusted for model 5 plus individual race and ethnicity.
Abbreviations: Adj, adjusted; CI, confidence interval; IRR, incidence rate ratio.
Other Analyses
We investigated effect modification of the association between community racial composition tertile and hospitalizations with a priori-specified patient (including sex, race, and comorbid conditions) and facility characteristics (profit status and patient census); results are summarized in Figure 2A and B. We did not find significant 2-way interaction between community racial composition tertile and age (P = 0.5), sex (P = 0.1), diabetes status (P = 0.1), race/ethnicity (P = 0.2), Charlson comorbidity index score (P = 0.1), facility profit status (P = 0.9), or facility patient census (P = 0.3). The effect of community racial composition tertile on hospitalization IRR was not found to be modified by these variables.
Figure 2.
Stratified analyses: incident rate ratios (IRRs) for hospitalization in (A) tertile 2 versus tertile 1 and (B) tertile 3 versus tertile 1 of community racial composition. Abbreviations: CI, confidence interval; DM, diabetes mellitus.
Crude mortality rates were 0.19, 0.15, and 0.14 deaths per patient-year in community racial composition tertiles 1, 2, and 3, respectively. In the secondary analysis, Cox analysis showed an HR for mortality of 0.92 (95% CI, 0.77–1.11) for tertile 3 as compared to tertile 1 and an HR of 0.85 (95% CI, 0.71–1.01) for tertile 2 as compared to tertile 1, when adjusted for age and facility clustering. Fully adjusted Cox models showed similar HRs for mortality within tertiles. In unadjusted sensitivity analysis 1, non-Hispanic White and non-Hispanic Black participants were at increased risk for hospitalization (IRRs of 1.28 [95% CI, 1.10–1.49] and 1.31 [95% CI, 1.10–1.57], respectively). The IRR became attenuated toward the null with the addition of community racial composition (percentage of Black residents in community as a continuous variable) in non-Hispanic Black participants only (1.18; 95% CI, 0.99–1.42) but not in non-Hispanic White participants (1.29; 95% CI, 1.12–1.48).
Sensitivity analysis 2 revealed that with percentage of White residents as the exposure variable, the IRR for hospitalization (for every 10 percentage points greater percentage of White residents) was 1.00 (95% CI, 0.97–1.03; P = 0.9) in unadjusted models and 1.01 (95% CI, 0.97–1.05; P = 0.5) in adjusted models. With percentage of Hispanic residents as exposure variable, for every 10 percentage points greater percentage of Hispanic residents, the IRR for hospitalization was 1.00 (95% CI, 0.96–1.04; P = 0.8) in unadjusted models and 0.97 (95% CI, 0.93–1.02, P = 0.3) in adjusted models.
Discussion
In communities defined by dialysis facility zip codes, we identified a strong association between community racial composition tertile and individual patients’ risk for hospitalization despite similar achievement of dialysis care benchmarks and even after adjustment for age, sex, comorbid conditions, and individual-level race/ethnicity. We also evaluated the effect of community attributes such as poverty, education, rurality, access to internet, and family structure on this association and found minimal attenuation. Patients receiving dialysis in communities with the highest percentage of Black residents were younger and healthier but were poorer and lived in urban communities. They used catheter access and shortened treatments more frequently. However, even after adjusting for these important variables, their hospitalization risks remained higher than for patients receiving care in communities with a lower percentage of Black residents. There seemed to be a stronger association in those younger than 45 years and in Hispanic and White patients, but the interaction was not statistically significant. Although our study does not identify those residual community-specific risks for hospitalizations, it makes certain explanations less likely with respect to this outcome. For example, the degree of comorbidity (Charlson score), low dialysis adherence behaviors, and achievement of dialysis metrics were not found to be important explanatory factors for the higher rate of hospitalizations in Black communities in the United States, unlike what was observed in other studies.21–23 For patients receiving maintenance dialysis, who rely heavily on health care institutions, providers and hospitals serving Black communities may lack sufficient training or resources to address their care barriers. Thus, health care facility access and quality should be explored as a potential contributor to this association.8,9,22–33
In our study, we showed that Black patients receiving dialysis in zip codes with the lowest percentage of Black residents had a diagnosis of glomerulonephritis more commonly and a diagnosis of hypertension less commonly as their cause of kidney failure than Black patients receiving dialysis in zip codes with the highest percentage of Black residents. The former group of patients may have been more likely to undergo kidney biopsy and have associated more intensive care before the onset of kidney failure, which also speaks to the quality of health care they likely received from nephrologists and health care systems that serve them.34 The attribution of kidney failure cause to hypertension is controversial, with many questioning its validity, and this diagnosis is used more commonly in Black than in White patients.35,36 The kidney care community is now starting to understand that some cases of kidney failure attributed to hypertension in Blacks were likely related to APOL1-mediated kidney diseases. In contrast to previous studies, we found that quality benchmarks such as URRs, number of dialysis minutes delivered, and hemoglobin and phosphorus levels were similar between facilities caring for patients across the community racial composition tertiles. Moreover, albumin and creatinine levels were higher in tertile 3, likely indicating better nutritional status in tertile 3.
Sensitivity analysis showed that as compared with the “other” racial/ethnic group, non-Hispanic White and non-Hispanic Black participants had higher hospitalization rates, consistent with prior analyses of the USRDS, but this association was attenuated when adjusting for community-level variables in Black participants only. This suggests that the attribution of hospitalization risk to Black race by the USRDS may be confounded by the risk for hospitalization inherent in residing in Black residential communities.1 Furthermore, the higher hospitalization risk of White and Hispanic residents in tertile 3 supports the role of contextual factors related to community racial composition, and not individual race/ethnicity, as a key driver of hospitalization.17 The higher health risks of Whites living in minority neighborhoods have been described in the dialysis population and the general population and may be related to socioeconomic disadvantage rather than structural barriers to care.17,37 For example, the rapid increase in the wealth gap with increasing costs of housing and education has led to a shift in this country from persons being classified as “upper and middle class” to “middle and lower class,” with many struggling and overextended to maintain their expectations and unaware they have been reclassified. In Hispanic patients, the language barrier with health care providers and institutions in patients living in poorer communities may present a barrier to patient engagement.38 Thus, the exposure to the riskstraditionally faced by residents of Black communities is recent for some groups and they may not have developed the resilience to cope through the types of social networks and other support systems that Black patients developed over many years to try to mitigate the effects of exposure to these economic and other community-level stressors that affect health. Distrust, apathy, educational disadvantage, and/or lack of self-advocacy, traditionally seen as a result of structural racism, is suggested by the finding that fewer patients treated in facilities located in majority Black communities completed surveys at their dialysis facilities.39,40
It is important to recognize the limitations of our study and interpret the results with caution. This cohort study may have residual confounding from unmeasured patient- or community-level variables. The use of community racial composition may have overlooked socioeconomic factors and their role in outcomes. Risks posed by community stressors such as noise, crime, housing problems, and access to healthy foods were not characterized, nor were shortcomings characteristic of local health care systems. Cause-specific hospitalization data were not available for the US DOPPS cohort in the years featured, which could have shed light on opportunities for future health interventions. Missing data were generally low but up to 19% for individual covariates. With this in mind, we adjusted for these variables using multiple imputation because we believed that this approach would be less biased than omitting them completely. Our imputation modeling depended on the assumption of missingness at random. A few of the community-level variables applied in the adjustment models were multicollinear. However, these variables are reserved for adjustment only and not interpreted as primary predictors of hospitalization risk. Furthermore, certain covariates were represented through surrogate measures. In the effect modification analysis, our study may also be under-powered in some of the strata to detect statistical differences by community racial composition tertile. We did not adjust for death as a competing risk because calculation of the IRR takes time at risk into account and we found that death rates did not differ by community racial composition tertile. The DOPPS database predates the 2017 introduction of hospitalization rate as a quality metric by CMS, which may affect attention to this outcome. Formal mediation analysis was not conducted to evaluate the contribution of explanatory variables to hospitalizations in putative causal pathways such as the lack of information about access to and quality of local health care systems as potential contributors to hospitalization rate. Finally, individual socioeconomic class data were not available and zip codes were used as the proxy for communities, limiting the accuracy of specific community attributes.
Our study had important strengths, such as the nationally representative participant and community characteristics with systematically collected data for important individual- and community-level sociodemographic, clinical, and facility-level covariates and the use of rigorous statistical methods.
In conclusion, this study found that patients treated in dialysis facilities located in communities with a high percentage of Black residents are at higher risk for hospitalization than patients treated in communities with a lower percentage of Black residents, despite similar quality of dialysis care and adherence, and after adjustment for individual comorbid conditions and individual- and community-level sociodemographic factors, including individual-level race/ethnicity. To eliminate health care disparities while at the same time reducing cost, the medical community should focus on addressing drivers of higher hospital/emergency department use in communities with a higher percentage of Black residents.
PLAIN-LANGUAGE SUMMARY.
There are few explanations for the variability in hospitalization rates across the United States among patients receiving hemodialysis. We studied the independent association of community racial composition with rates of hospitalization among patients receiving maintenance hemodialysis within the Dialysis Outcomes and Practice Patterns Study (DOPPS). Community racial composition was obtained from Census Bureau data for the zip codes in which the 4,567 study patients were treated. We found that patients receiving hemodialysis in communities with more Black residents were at higher risk for hospitalization after controlling for age, comorbid conditions, and adherence to hemodialysis treatments. We could not determine the underlying factors that are driving this higher hospitalization risk, which should be the focus of future studies.
Acknowledgements:
We acknowledge the help of Dr Colin Rehm for obtaining American Community Survey data and of Mr Brian Bieber for help obtaining the DOPPS database.
Support: This study was supported by a grant from the National Center for Advancing Translational Sciences, components of the National Institutes of Health (NIH), through Clinical and Translational Science Awards grant number UL1TR002556-01 awarded to Dr Golestaneh. Drs Golestaneh, Melamed, and Cavanaugh are supported by R18 DK118471-01. Dr Norris is supported by NIH grants P30AG021684 and UL1TR001881. The funders did not have a role in the study design; data collection, analysis, or reporting; or the decision to submit for publication.
Financial Disclosure: Dr Golestaneh receives salary support from the Montefiore Care Management Organization, has received honoraria from Horizon Pharmaceuticals in return for consulting services, and is a member of the Clinical Events Committee for the Spyral Pivotal Hypertension On-Medications and Spyral Pivotal Hypertension Off-Medications sponsored by Medtronic. She also received a travel grant from the Cardiorenal Society of America. Dr Karaboyas is an employee of Arbor Research Collaborative for Health, which administers the DOPPS. Global support for the ongoing DOPPS Programs is provided by a consortium of funders listed in the provided URL without restriction on publications; further information is available at www.dopps.org/AboutUs/Support.aspx. Dr Melamed receives an honorarium from the American Board of Internal Medicine for serving on the Nephrology Examination Committee. The remaining authors declare that they have no relevant financial interests.
References
- 1.Saran R, Robinson B, Abbott KC, et al. US Renal Data System 2018 Annual Data Report: epidemiology of kidney disease in the United States. Am J Kidney Dis. 2019;73(3)(suppl 1):A7–A8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Dalrymple LS, Mu Y, Nguyen DV, et al. Risk factors for infection-related hospitalization in in-center hemodialysis. Clin J Am Soc Nephrol. 2015;10(12):2170–2180. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Slinin Y, Guo H, Li S, et al. Hemodialysis patient outcomes: provider characteristics. Am J Nephrol. 2014;39(5):367–375. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Newman KL, Lynch RJ, Adams AB, Zhang R, Pastan SO, Patzer RE. Hospitalization among individuals waitlisted for kidney transplant. Transplantation. 2017;101(12):2913–2923. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Saran R, Robinson B, Abbott KC, et al. US Renal Data System 2017 Annual Data Report: epidemiology of kidney disease in the United States. Am J Kidney Dis. 2018;71(3)(suppl 1):S1–S676. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Yan G, Norris KC, Greene T, et al. Race/ethnicity, age, and risk of hospital admission and length of stay during the first year of maintenance hemodialysis. Clin J Am Soc Nephrol. 2014;9(8): 1402–1409. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7..Makuc DM, Haglund B, Ingram DD, Kleinman JC, Feldman JJ. Health Service Areas for the United States. Vital Health Stat 2. 1991;2(119):1–16. [PubMed] [Google Scholar]
- 8.Mathew AT, Rosen L, Pekmezaris R, et al. Potentially avoidable readmissions in United States hemodialysis patients. Kidney Int Rep. 2018;3(2):343–355. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Golestaneh L, Bellin E, Neugarten J, Lo Y. Avoidable visits to the emergency department (ED) and their association with sex, age and race in a cohort of low socio-economic status patients on hemodialysis in the Bronx. PLoS One. 2018;13(8): e0202697. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.United States Census Bureau. American Community Survey. 2016. https://www.census.gov/programs-surveys/acs/guidance.html. Accessed May 22, 2019.
- 11.DOPPS Practice Monitor. DPM sampling, study design, and calculation methods. 2017. https://www.dopps.org/dpm/Data_Sources_Methods.pdf. Accessed June 5, 2019.
- 12.Pisoni RL, Fuller DS, Bieber BA, Gillespie BW, Robinson BM. The DOPPS Practice Monitor for US dialysis care: trends through August 2011. Am J Kidney Dis. 2012;60(1):160–165. [DOI] [PubMed] [Google Scholar]
- 13.Robinson B, Fuller D, Zinsser D, et al. The Dialysis Outcomes and Practice Patterns Study (DOPPS) Practice Monitor: rationale and methods for an initiative to monitor the new US bundled dialysis payment system. Am J Kidney Dis. 2011;57(6):822–831. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Bikbov B, Bieber B, Andrusev A, et al. Hemodialysis practice patterns in the Russia Dialysis Outcomes and Practice Patterns Study (DOPPS), with international comparisons. Hemodial Int. 2017;21(3):393–408. [DOI] [PubMed] [Google Scholar]
- 15.Cavanaugh KL, Osborn CY, Tentori F, Rothman RL, Ikizler TA, Wallston KA. Performance of a brief survey to assess health literacy in patients receiving hemodialysis. Clin Kidney J. 2015;8(4):462–468. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Weaver CG, Ravani P, Oliver MJ, Austin PC, Quinn RR. Analyzing hospitalization data: potential limitations of Poisson regression. Nephrol Dial Transplant. 2015;30(8):1244–1249. [DOI] [PubMed] [Google Scholar]
- 17.Rodriguez RA, Sen S, Mehta K, Moody-Ayers S, Bacchetti P, O’Hare AM. Geography matters: relationships among urban residential segregation, dialysis facilities, and patient outcomes. Ann Intern Med. 2007;146(7):493–501. [DOI] [PubMed] [Google Scholar]
- 18.Becketi S Regression standard errors in clustered samples. Stata Tech Bull. 1993;3(13):1–32. [Google Scholar]
- 19.United States Renal Data System. 2017 Annual Data Report Highlights. 2017. https://www.usrds.org/adrhighlights.aspx. Accessed May 25, 2019.
- 20.Jakobsen JC, Gluud C, Wetterslev J, Winkel P. When and how should multiple imputation be used for handling missing data in randomised clinical trials - a practical guide with flowcharts. BMC Med Res Methodol. 2017;17(1):162. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Assimon MM, Flythe JE. Thirty-day hospital readmissions in the hemodialysis population: a problem well put, but half-solved. Clin J Am Soc Nephrol. 2017;12(10):1566–1568. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Flythe JE, Katsanos SL, Hu Y, Kshirsagar AV, Falk RJ, Moore CR. Predictors of 30-day hospital readmission among maintenance hemodialysis patients: a hospital’s perspective. Clin J Am Soc Nephrol. 2016;11(6):1005–1014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Mathew AT, Strippoli GF, Ruospo M, Fishbane S. Reducing hospital readmissions in patients with end-stage kidney disease. Kidney Int. 2015;88(6):1250–1260. [DOI] [PubMed] [Google Scholar]
- 24.Sehgal AR. Outcomes of renal replacement therapy among blacks and women. Am J Kidney Dis. 2000;35(4)(suppl 1): S148–S152. [DOI] [PubMed] [Google Scholar]
- 25.Sehgal AR. Impact of quality improvement efforts on race and sex disparities in hemodialysis. JAMA. 2003;289(8):996–1000. [DOI] [PubMed] [Google Scholar]
- 26.Hao H, Lovasik BP, Pastan SO, Chang HH, Chowdhury R, Patzer RE. Geographic variation and neighborhood factors are associated with low rates of pre-end-stage renal disease nephrology care. Kidney Int. 2015;88(3):614–621. [DOI] [PubMed] [Google Scholar]
- 27.Ronksley PE, Hemmelgarn BR, Manns BJ, et al. Potentially preventable hospitalization among patients with CKD and high inpatient use. Clin J Am Soc Nephrol. 2016;11(11):2022–2031. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Harding K, Mersha TB, Webb FA, Vassalotti JA, Nicholas SB. Current state and future trends to optimize the care of African Americans with end-stage renal disease. Am J Nephrol. 2017;46(2):156–164. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Green JA, Gonzaga AM, Cohen ED, Spagnoletti CL. Addressing health literacy through clear health communication: a training program for internal medicine residents. Patient Educ Couns. 2014;95(1):76–82. [DOI] [PubMed] [Google Scholar]
- 30.Green JA, Mor MK, Shields AM, et al. Associations of health literacy with dialysis adherence and health resource utilization in patients receiving maintenance hemodialysis. Am J Kidney Dis. 2013;62(1):73–80. [DOI] [PubMed] [Google Scholar]
- 31.Flythe JE, Hilbert J, Kshirsagar AV, Gilet CA. Psychosocial factors and 30-day hospital readmission among individuals receiving maintenance dialysis: a prospective study. Am J Nephrol. 2017;45(5):400–408. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Chan L, Chauhan K, Poojary P, et al. National estimates of 30-day unplanned readmissions of patients on maintenance hemodialysis. Clin J Am Soc Nephrol. 2017;12(10):1652–1662. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Fuller DS, Robinson BM. Facility practice variation to help understand the effects of public policy: insights from the Dialysis Outcomes and Practice Patterns Study (DOPPS). Clin J Am Soc Nephrol. 2017;12(1):190–199. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Norris KC, Williams SF, Rhee CM, et al. Hemodialysis disparities in African Americans: the deeply integrated concept of race in the social fabric of our society. Semin Dial. 2017;30(3): 213–223. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Weisstuch JM, Dworkin LD. Does essential hypertension cause end-stage renal disease? Kidney Int Suppl. 1992;36:S33–S37. [PubMed] [Google Scholar]
- 36.Schlessinger SD, Tankersley MR, Curtis JJ. Clinical documentation of end-stage renal disease due to hypertension. Am J Kidney Dis. 1994;23(5):655–660. [DOI] [PubMed] [Google Scholar]
- 37.Haas JS, Phillips KA, Sonneborn D, et al. Variation in access to health care for different racial/ethnic groups by the racial/ethnic composition of an individual’s county of residence. Med Care. 2004;42(7):707–714. [DOI] [PubMed] [Google Scholar]
- 38.Madden EF, Qeadan F. Dialysis hospitalization inequities by Hispanic ethnicity and immigration status. J Health Care Poor Underserved. 2017;28(4):1509–1521. [DOI] [PubMed] [Google Scholar]
- 39.Mohottige D, Boulware LE. Trust in American medicine: a call to action for health care professionals. Hastings Center Rep. 2020;50(1):27–29. [DOI] [PubMed] [Google Scholar]
- 40.Sullivan LS. Trust, risk, and race in American medicine. Hastings Center Rep. 2020;50(1):18–26. [DOI] [PubMed] [Google Scholar]