Abstract
Background End-of-life care is a prominent consideration in patients on maintenance dialysis, especially when death appears imminent and quality of life is poor. To date, examination of race- and ethnicity-associated disparities in end-of-life care for patients with ESRD has largely been restricted to comparisons of white and black patients.
Methods We performed a retrospective national study using United States Renal Data System files to determine whether end-of-life care in United States patients on dialysis is subject to racial or ethnic disparity. The primary outcome was a composite of discontinuation of dialysis and death in a nonhospital or hospice setting.
Results Among 1,098,384 patients on dialysis dying between 2000 and 2014, the primary outcome was less likely in patients from any minority group compared with the non-Hispanic white population (10.9% versus 22.6%, P<0.001, respectively). We also observed similar significant disparities between any minority group and non-Hispanic whites for dialysis discontinuation (16.7% versus 31.2%), as well as hospice (10.3% versus 18.1%) and nonhospital death (34.4% versus 46.4%). After extensive covariate adjustment, the primary outcome was less likely in the combined minority group than in the non-Hispanic white population (adjusted odds ratio, 0.55; 95% confidence interval, 0.55 to 0.56; P<0.001). Individual minority groups (non-Hispanic Asian, non-Hispanic black, non-Hispanic Native American, and Hispanic) were significantly less likely than non-Hispanic whites to experience the primary outcome. This disparity was especially pronounced for non-Hispanic Native American and Hispanic subgroups.
Conclusions There appear to be substantial race- and ethnicity-based disparities in end-of-life care practices for United States patients receiving dialysis.
End-of-life care is an increasingly prominent consideration, especially in situations where death appears imminent and quality of life is poor. Research and commentary on end-of-life issues have accelerated rapidly. For example, a PubMed search in March of 2018 with the term “end-of-life” revealed that annual citation rates increased over one hundred–fold between 1990 and 2017, from 19 to 2243.1 Unexplained race-based disparities have long been a feature of the United States dialysis population, with black patients experiencing higher incidence rates of ESKD treated with RRT, longer survival, and lower transplantation rates then age-matched, white counterparts.2 Given its emerging importance, it is surprising that examination of race- and ethnicity-associated disparities in end-of-life care in patients with ESKD has largely been restricted to comparisons of white and black patients. This is even more surprising when one considers demographic changes in the general population, and the observation that ESKD rates are meaningfully higher in patients of Hispanic ethnicity than in their white contemporaries.2 Furthermore, it is unknown whether potential disparities differ by mode of health insurance and regional income inequalities. Given these knowledge gaps, we performed a retrospective national study to determine whether end-of-life care in United States dialysis exhibits disparities across multiple races and ethnicities.
Methods
Objectives
Among patients dying on maintenance dialysis between January 1, 2000 and June 30, 2014, the objectives of this study were to quantify associations between race-ethnicity and: (1) the composite outcome of a) discontinuation of dialysis and b) death in either a nonhospital or hospice setting, the primary outcome; and (2) individual components of the composite outcome (discontinuation of dialysis, hospice care, death in a nonhospital setting). In addition, we wished to determine whether associations between race-ethnicity and end-of-life care outcomes differed by: (3) mode of health care insurance and (4) income dispersion, as measured by the Gini index.
Subjects and Measurements
We used United States Renal Data System (USRDS) standard analytic files (“saf”) to study patients on maintenance dialysis who died between January 1, 2000 and June 30, 2014. We used the saf.Death file to determine date, cause, and place of death; discontinuation of dialysis; and hospice care. Demographic factors were obtained from the saf.Patients file; mirroring strategies used in the National Health and Nutrition Examination Survey, “Race” and “Hispanic” variables were used to define race-ethnicity; race was selected when ethnicity was non-Hispanic, and ethnicity when the latter was Hispanic.3 The saf.Patients file was also used to determine date of birth, sex, type of kidney disease, and duration of RRT. Last mode of dialysis and dialysis unit characteristics were determined from the saf.Rxhist and saf.Facility files, respectively. The saf.Payhist file was used to determine insurance status. In the majority subgroup (74.6%) with Medicare parts A and B insurance, Medicare hospitalization files (saf.Hosp) and International Classification of Diseases (Ninth Revision) Clinical Modification codes were used to capture the presence of common medical conditions seen in the last 4 weeks of life. The saf.Residenc file was used to determine the state and county of residence at the time of death; linkage by county to census and Department of Agriculture Economic Research Service files allowed us to determine income dispersion (the Gini index) and rural-urban continuum codes, respectively.4,5
Statistical Analyses
We used the chi-squared test and logistic regression, respectively, for comparisons of categoric variables and calculation of odds ratios. Four adjustment strategies were used for estimating race- and ethnicity-related odds ratios for end-of-life care parameters: model 1—no adjustment; model 2—adjustment for age and sex; model 3—model 2 plus adjustment for Gini index of county-level income dispersion, rural-urban continuum code, type of kidney disease, years of RRT, mode of dialysis, prior transplant, type of insurance, and dialysis unit characteristics; model 4 (applied to subgroup with Medicare parts A and B)—model 3 plus adjustment for number of hospitalization-identified illnesses in the last 28 days of life. Adjustment models were repeated in the following subgroups: those with Medicare Parts A and B health insurance, those insured with group health organizations, those with county-level Gini index below the national median, and those with Gini index above the national median. SAS Version 9.4 (Cary, NC) was used for statistical analyses.6
Results
Of the study population, 27.6% were classified as non-Hispanic black, 1.0% as non-Hispanic Native American, 3.2% as non-Hispanic Asian, and 11.7% as Hispanic (Table 1). The primary cause of death was cardiovascular causes in 39.6%, infection in 10.9%, malignancy in 2.8%, and uremia/dialysis withdrawal in 7.4% (Table 1). A total of 63.2% died in hospital, 19.6% at home, and 7.4% in nursing homes; 14.7% received hospice care before death and 24.9% of the study population discontinued dialysis before death, predominantly because of failure to thrive (34.8% of withdrawals) and medical complications (24.9%; Table 1).
Characteristics at death of patients on dialysis, compared by race-ethnicity (n=1,098,384)
The primary outcome—a composite of (1) discontinuation of dialysis and (2) death in a nonhospital or hospice setting—was less likely in patients from minority groups (10.9%) than in the white population (22.6%, P value <0.001; Table 2). Corresponding values for dialysis discontinuation, hospice, and nonhospital death were 16.7% versus 31.2%, 10.3% versus 18.1%, and 34.4% versus 46.4%, respectively (P value <0.001 for each comparison; Table 2). Within individual minority groups, primary outcome estimates were arrayed as follows: non-Hispanic black (9.8%) <non-Hispanic Asian (11.2%) <Hispanic (13.0%) <non-Hispanic Native American (14.2%); this pattern was repeated for each component of the primary outcome, with the exception of hospice care, where the pattern was non-Hispanic black (9.5%) <non-Hispanic Asian (10.6%) <non-Hispanic Native American (11.1%) <Hispanic (12.1%) (Table 2).
Unadjusted associations of primary outcome (withdrawal of dialysis and death in either a nonhospital or hospice setting), and of component outcomes in patients on dialysis dying between 2000 and 2014 (n=1,098,384)
When adjustment was made for age, sex, era, Gini index of county-level income dispersion, rural-urban continuum code, type of kidney disease, years of RRT, mode of dialysis, prior transplant, type of insurance, and dialysis unit characteristics (model 3), the likelihood of the composite primary outcome was lower among patients from any minority group (adjusted odds ratio [AOR] versus white, 0.55; 95% confidence interval [95% CI], 0.55 to 0.56; P value <0.001; Table 3); within individual minority groups, model 3 AOR values (versus white, P value <0.001 for each) were similarly low for non-Hispanic Asian (AOR, 0.49; 95% CI, 0.48 to 0.45) and non-Hispanic black (AOR, 0.48; 95% CI, 0.47 to 0.49) subgroups and higher, although <1, for non-Hispanic Native American (AOR, 0.72; 95% CI, 0.68 to 0.76) and Hispanic (AOR, 0.73; 95% CI, 0.72 to 0.74) subgroups (Table 3). Model 3 minority-associated AORs for the primary outcome were <1 (P value <0.001) in all subgroups examined (Figure 1). Findings were similar when outcome models were repeated in the subgroups defined by type of health insurance (Table 4) and in subgroups defined by median county-level Gini index of income dispersion (Tables 4 and 5).
Odds ratios of primary outcome (withdrawal of dialysis and death in either a nonhospital or hospice setting), and of component outcomes, according to minority status in patients on dialysis dying between 2000 and 2014 (n=1,098,384)
Subgroup analyses; parameter estimates are odds ratios (with 95% CIs and non-Hispanic white as reference category) from logistic regression models for the primary outcome (withdrawal of dialysis and death in either a nonhospital or hospice setting). Odds ratios for race and ethnicity were similar in all subgroups examined. Ethnicity is non-Hispanic, unless otherwise stated. Model 3 adjustment strategy was used: adjustment for age, sex, era, Gini index of county-level income dispersion, rural-urban continuum code, type of kidney disease, years of RRT, mode of dialysis, prior transplant, type of insurance, and dialysis unit characteristics. “Conditions” refers to patients with Medicare parts A and B with hospitalizations for the following in the last 28 days of life: myocardial infarction, cardiac failure, stroke, malignancy, pneumonia, and septicemia. HD, hemodialysis; HMO, health maintenance organization; HTN, hypertension; mill., million (population); PD, peritoneal dialysis; RRT, renal replacement therapy.
Odds ratios of primary outcome (withdrawal of dialysis and death in either a non-hospital or hospice setting), and of component outcomes, according to minority status in patients on dialysis dying between 2000 and 2014, analyzed separately in subgroups defined by insurance provider
Odds ratios of primary outcome (withdrawal of dialysis and death in either a nonhospital or hospice setting), and of component outcomes, according to minority status in patients on dialysis dying between 2000 and 2014, analyzed separately in subgroups defined by the median county-level Gini index of 0.43
STROBE statement—checklist of items that should be included in reports of observational studies
Discussion
We observed that patients of minority race or ethnicity were less likely to discontinue dialysis, less likely to receive hospice care, and more likely to die in hospital than their non-Hispanic white counterparts. Across the breadth of end-of-life outcomes studied here, patterns were broadly similar for patients of non-Hispanic black and Asian race-ethnicity, and broadly similar for patients of non-Hispanic Native American race-ethnicity and Hispanic ethnicity. These disparities could not be explained by differences in age, demography, local income dispersion, urban-rural configuration, dialysis facility type, mode of insurance coverage, and recent illness profiles, and were evident in a wide range of subgroups.
The proportions of patients on dialysis dying in nonhospital or hospice settings in this study appeared low when compared with populations without ESKD. For example, among Medicare-insured participants in the Health and Retirement Survey who died between 1998 and 2012, the average age at death was 83 years and 29.3% died in hospital (versus 63.2% in patients on dialysis in our study) and 37.3% received hospice care (versus 14.7%).7 In nondialysis settings racial and ethnic disparities have also been described in many domains of end-of-life care, including access to palliative care, symptom alleviation, and communication.8–10 Compared with white individuals, those from minority populations have a greater likelihood of death in a hospital setting, and those of black race or Hispanic ethnicity are at higher risks of being hospitalized and receiving intensive care in the last 6 months of life.11 Several studies also suggest that individuals from black, Hispanic, and Asian minority populations are less well informed about advance directives and less likely to complete them.12–14 Other studies have shown lower rates of hospice use for older adults from minority populations, across a wide variety of care settings, major diagnoses, and geographic regions.15–19
Few studies have specifically focused on the nexus of race, ethnicity, dialysis withdrawal, hospice care, and death outside of conventional hospital settings. This being said, unexplained disparities have been observed in related domains of care. For example, one study reported that Hispanic patients on dialysis were more likely to undergo intensive medical procedures such as ventilation, tracheostomy, feeding enterostomy, and cardiopulmonary resuscitation in the last 6 months of life.20 Other studies comparing black to white patients showed that in-hospital death, dialysis discontinuation, and hospice referral differed substantially between the two racial populations examined, as well as across regions of the United States.21–23
Our study differs from these informative studies in a number of ways. One difference from preexisting research in this area is that, because the analysis relied on the Death Notification Form, as opposed to Medicare claims, we were able to describe patterns of end-of-life care for groups without Medicare Parts A and B coverage. In this regard, it was notable that adjusted race-and-ethnicity–associated odds ratios for our primary outcome were similar in the subgroups with insurance provided by Medicare Parts A and B and health maintenance organizations. Another potentially novel aspect of this study was the examination of racial-ethnic disparities within regions of different income dispersion. Regarding regional income dispersion, it was notable that adjusted race-and-ethnicity–associated odds ratios for our primary outcome were similar in all levels examined.
In particular, having focused on several minority groups, we found that non-Hispanic white patients on dialysis were the outlier with regard to end-of-life care. Although our study cannot accurately assess the role of patient-related and non–patient-related factors in our findings, the observation that end-of-life care differed between nonminority and the combined minority population, as well as between individual minority groups, tempts one to speculate that the causes of the disparities seen in this study may be multifaceted; for example, if end-of-life care was entirely decided by an entity other than the individual patient, and this entity uniformly treated individuals from the majority population in one way, and individuals from minority populations in another way, one would not expect to see differences between minority groups.
This study has limitations, including its retrospective, registry-based design and the use of reimbursement claims to define comorbidity close to death. Another limitation of our study is the fact that the variables contained in the USRDS Death Notification Form file have not been formally validated. Confronted with substantial racial and ethnic biases at the level of health delivery systems, careful, prospective confirmatory studies are needed to characterize attitudes and belief systems regarding end-of-life issues, both among health care recipients and health care providers.
Despite its limitations, our study may have useful features. It is nationally representative and the large sample sizes help to generate precise association estimates, within the overall population and within multiple subgroups. Given that ongoing illness is likely to influence end-of-life care, the ability to capture comorbid illness present may be advantageous, because it tends to counter the hypothesis that racial and ethnic disparities in end-of-life care are explicable by differences in illness profiles.
Disclosures
None.
Acknowledgments
R.N.F.: (1) Substantial contributions to conception and design. Substantial contributions to acquisition of data. Substantial contributions to analysis and interpretation of data. (2) Substantial contributions to drafting the article. Substantial contributions to revising it critically for important intellectual content. (3) Final approval of the version to be published. (4) Agreement to be accountable for all aspects of the work. D.J.S.: (1) Substantial contributions to conception and design. Substantial contributions to acquisition of data. Substantial contributions to analysis and interpretation of data. (2) Substantial contributions to drafting the article. Substantial contributions to revising it critically for important intellectual content. (3) Final approval of the version to be published. (4) Agreement to be accountable for all aspects of the work. P.D.: (1) Substantial contributions to conception and design. (2) Substantial contributions to revising it critically for important intellectual content. (3) Final approval of the version to be published. (4) Agreement to be accountable for all aspects of the work. A.I.: (1) Substantial contributions to analysis and interpretation of data. (2) Substantial contributions to revising it critically for important intellectual content. (3) Final approval of the version to be published. (4) Agreement to be accountable for all aspects of the work. S.R.: (1) Substantial contributions to conception and design. (2) Substantial contributions to drafting the article. Substantial contributions to revising it critically for important intellectual content. (3) Final approval of the version to be published. (4) Agreement to be accountable for all aspects of the work.
Footnotes
Published online ahead of print. Publication date available at www.jasn.org.
- Copyright © 2018 by the American Society of Nephrology