Relationship of neighborhood social determinants of health on racial/ethnic mortality disparities in US veterans—Mediation and moderating effects

Abstract Objective To examine mediation and moderation of racial/ethnic all‐cause mortality disparities among Veteran Health Administration (VHA)‐users by neighborhood deprivation and residential segregation. Data sources Electronic medical records for 10/2008‐9/2009 VHA‐users linked to National Death Index, 2000 Area Deprivation Index, and 2006‐2009 US Census. Study design Racial/ethnic groups included American Indian/Alaskan Native (AI/AN), Asian, non‐Hispanic black, Hispanic, Native Hawaiian/Other Pacific Islander, and non‐Hispanic white (reference). We measured neighborhood deprivation by Area Deprivation Index, calculated segregation for non‐Hispanic black, Hispanic, and AI/AN using the Isolation Index, evaluated mediation using inverse odds‐weighted Cox regression models and moderation using Cox regression models testing for neighborhood*race/ethnicity interactions. Principal findings Mortality disparities existed for AI/ANs (HR = 1.07, 95%CI:1.01‐1.10) but no other groups after covariate adjustment. Neighborhood deprivation and Hispanic segregation neither mediated nor moderated AI/AN disparities. Non‐Hispanic black segregation both mediated and moderated AI/AN disparities. The AI/AN vs. non‐Hispanic white disparity was attenuated for AI/ANs living in neighborhoods with greater non‐Hispanic black segregation (P = .047). Black segregation's mediating effect was limited to VHA‐users living in counties with low black segregation. AI/AN segregation also mediated AI/AN mortality disparities in counties that included or were near AI/AN reservations. Conclusions Neighborhood characteristics, particularly black and AI/AN residential segregation, may contribute to AI/AN mortality disparities among VHA‐users, particularly in communities that were rural, had greater black segregation, or were located on or near AI/AN reservations. This suggests the importance of neighborhood social determinants of health on racial/ethnic mortality disparities. Living near reservations may allow AI/AN VHA‐users to maintain cultural and tribal ties, while also providing them with access to economic and other resources. Future research should explore the experiences of AI/ANs living in black communities and underlying mechanisms to identify targets for intervention.


| INTRODUC TI ON
Health care is an important-but not sole-determinant of health outcomes. 1 Even within health care systems that strive to provide equal access to all patients, racial/ethnic disparities exist. [2][3][4][5][6] Health care systems alone cannot expect to eliminate racial/ethnic health disparities.
Eliminating disparities will require examining social determinants of health: social and contextual factors outside of health care systems that affect health.
The Veterans Health Administration (VHA) is an example of a health system that has made strides in addressing racial/ethnic disparities.
Within VHA, non-Hispanic blacks (hereafter, "blacks") had similar or lower mortality from prostate cancer, lung cancer, and chronic kidney disease relative to non-Hispanic whites (hereafter, "whites"). 2,7,8 However, some VHA racial/ethnic disparities still persist, including heart disease mortality among blacks, all-cause mortality among American Indian/Alaskan Natives (AI/ANs), and hypertension control in both of these groups. 2,4,5 The neighborhoods in which people live are an important social determinant of health. Methodological advances have allowed researchers to disentangle the neighborhood's compositional effects (ie, individuals with worse health cluster in certain neighborhoods, for example, sicker people live in lower income communities) from its contextual effects on health (ie, group properties of the neighborhood influence health, for example, neighborhood poverty leads to worse health). 9 Studies in the US general population have identified neighborhood contextual effects, reinforcing that where we live does, indeed, affect our health. 10 While neighborhoods encompass complex social and physical environments, identifying specific neighborhood characteristics associated with health disparities may guide policy and practice, such as community-based partnerships, initiatives to improve healthy food access, and comprehensive neighborhood revitalization efforts. [11][12][13][14] Two neighborhood characteristics associated with health disparities in the general population are neighborhood socioeconomic status (SES) and racial/ethnic segregation. 10,[15][16][17] Lower SES neighborhoods typically have fewer health promoting resources (eg, healthy food options, recreational spaces), more crime, and subpar housing, which negatively affect health. Furthermore, lower SES and racial/ethnic minority communities have fewer health care resources. 18,19 Racial/ethnic disparities may arise from racial/ethnic groups living in different neighborhoods that expose them to different contexts and risk factors. In the US general population, racial/ethnic minority groups at higher risk for mortality, such as blacks, are more likely to live in segregated and more deprived neighborhoods. 20 However, segregation can also have positive health effects for racial/ethnic minority groups by buffering residents from racism and other stressors, attracting minority-owned businesses that provide culturally concordant services (eg, traditional food outlets), providing spaces like churches for groups to congregate and organize, and fostering social connections that promote transmission of health-related information (eg, awareness of health care service availability). 10,21,22 Segregation effects on health may vary by race/ethnicity. Segregation of different racial/ethnic groups arose through different processes and historical forces. Black segregation formed through discriminatory practices and policies (eg, residential redlining concentrated blacks into low SES, disinvested communities). 20 AI/AN segregation occurred through European and American colonization that led to forced and traumatic relocation of AI/ANs onto tribal reservations. 23 Many reservations are located in rural areas with high poverty and unemployment and limited health care resources . 23,24 However, residing on reservations may also promote health by strengthening tribal and cultural ties. 25 Hispanic segregation has occurred through also providing them with access to economic and other resources. Future research should explore the experiences of AI/ANs living in black communities and underlying mechanisms to identify targets for intervention. the creation of ethnic enclaves. Despite being located in disproportionately low SES areas, 26 Hispanic enclaves may promote health by preserving cultural practices and promoting social cohesion. 27 It is unknown whether neighborhood deprivation and segregation account for observed disparities among racial/ethnic minority VHA-users. Understanding whether such relationships exist in this population can yield important insights for the VHA and the broader health care system. A key difference between VHA-users and the US general population is that VHA-users have access to a health care system that provides comprehensive physical and mental health services and has initiatives to improve access (eg, transportation and extended clinic hours). 28 Given that unequal access to health care contributes to racial/ethnic disparities, 29 examining drivers of racial/ ethnic disparities among VHA-users provides a unique opportunity to understand how social determinants contribute to disparities beyond health care access. 30 VHA-users, compared to the US general population, have lower incomes and poorer health status; 31,32 therefore, with our analysis, we can examine neighborhood influences within a population with medical and social vulnerabilities.
This study has specific implications to VHA, given its broad mission to improve the health of all Veterans. Compared to other health care systems, VHA may be better positioned to address social determinants of health, as they are already targeting some determinants including housing and employment. For example, VHA's permanent supportive program is one of the largest programs in the nation and has been credited with substantially reducing Veteran homelessness nationally. 33 VHA also has employment training programs for those with service-connected disability and serious mental illness. 34 It is unknown, though, whether these efforts to improve Veteran health and well-being buffer Veterans from some of the deleterious effects of neighborhood-level social determinants of health.
In this study, we examined the role of two neighborhood-level social determinants of health on racial/ethnic all-cause mortality disparities among a cohort of Veteran VHA-users-neighborhood deprivation and residential segregation. Our first aim explored racial/ethnic differences in all-cause mortality from 2008 to 2016 in a national cohort of VHA-users.
Our second aim examined whether neighborhood deprivation a) mediated or b) moderated racial/ethnic all-cause mortality disparities among VHAusers. Our third aim similarly examined whether residential segregation a) mediated or b) moderated these relationships. We hypothesized that blacks and AI/ANs would experience disparities in all-cause mortality, that both neighborhood deprivation and segregation would mediate the racial/ethnic all-cause mortality disparities, and that racial/ethnic disparities would be greater in neighborhoods with greater deprivation and segregation and attenuated in less deprived and segregated neighborhoods. 23

| Data and sample
Our analytic sample consisted of a national cohort of Veterans

| Dependent variables
We examined time to all-cause mortality, calculated as the difference in years between date-of-death and qualifying date-of-initial FY2009 health care use. Deaths were classified as either observed on the date of death or administratively censored at the end of the mortality ascertainment period.

| Independent variable
Our main independent variable was a categorical indicator of patient race/ethnicity. We created the race/ethnicity variable using a validated algorithm that combines multiple data sources and years of data in a hierarchy ordered by data quality of the source data files, with self-identified race/ethnicity considered the highest. 36,37 We classified all individuals with Hispanic ethnicity as Hispanic and all others by race. Our analysis included the following racial/ethnic groups: non-Hispanic AI/AN (hereafter, "AI/AN"), non-Hispanic Asian (hereafter, "Asian"), black, Hispanic, non-Hispanic Native Hawaiian/Other Pacific Islander (hereafter, "NH/OPI"), and white, as well as multirace and unknown race (results not presented for latter two groups).

| Mediators and moderators
We examined two neighborhood characteristics: neighborhood deprivation and neighborhood segregation. Neighborhood deprivation was captured through the ADI. 35 This validated index was derived from 17 measures of income, education, employment, and housing quality from the 2000 American Community Survey (ACS) data. 38,39 Higher ADI values indicate greater neighborhood deprivation.
While numerous measures of neighborhood segregation have been developed to measure residential segregation, 40 our analysis used the isolation index, which captures the degree to which a member of one racial/ethnic group is exposed only to individuals of that same group. 40 The isolation index has several advantages over other segregation measures for our analysis: it better captures the underlying processes through which segregation can negatively affect health (eg, concentrated disadvantage) and is one of the most commonly used segregation indices. 17,41 The isolation index is calculated where n is the number of census tracts in the county, x i is the comparison group population in census tract i, X is the sum of all x i (total comparison group population in the county), and t i is the total population of census tract i. The isolation index ranges from 0 to 1, where higher isolation indices indicate greater racial/ethnic group isolation. We calculated separate isolation indices for blacks, Hispanics, and AI/ANs.

| Other variables
We included the following covariates in our analysis: age (categori- Index of Comorbidity, based on a weighted count of smoking status and seven chronic medical conditions associated with increased mortality (prior myocardial infarction, cancer, lung disease, congestive heart failure, diabetes, pneumonia, and stroke). 42 We identified medical and mental health diagnoses using the ICD-9-CM outpatient and inpatient diagnosis codes from FY2009. Individual-level SES was based on VHA's enrollment priority group. Veterans that did not have a service-connected disability, but that had an income below VHA's threshold for requiring copayment for care were classified as having low SES, while those above the threshold were classified as high SES. For Veterans with a military service-connected disability, their priority group did not reflect their income; thus, they are classified as indeterminate SES. Rurality indicator categories were based upon the Goldsmith Modification of the Office of Management and Budget definition of urban, rural, and highly rural land areas. 43

| Statistical analysis
We calculated descriptive statistics of race/ethnicity stratified means and proportions of the VHA population's demographics, health characteristics, and neighborhood characteristics, and crude race/ethnicitystratified mortality rates (number of deaths/100 000 population).

| Racial/ethnic all-cause mortality differences
We used three Cox regression models to calculate all-cause mortality hazard ratios comparing each racial/ethnic minority group to whites (reference group). Model 1 adjusted for age and sex. Model 2 further adjusted for individual SES and urban/rural status, which are important factors when considering neighborhood-level social determinants of health. Model 3 further adjusted for comorbidity and mental health comorbidity, which are on the causal pathway of mortality disparities. All models included clustered standard errors at the VHA facility level to account for within-VHA facility correlation.

| Mediation analysis
Mediation analysis involves estimating the total effect of race/ethnicity on mortality, and decomposing this effect into its direct (unmediated) and indirect (mediated) effect components. We tested mediation by neighborhood deprivation and segregation separately in the following way. First, we estimated the total effect of race/ ethnicity on mortality for each mediator through Cox regression while adjusting for potential confounders and the neighborhood mediator of interest. For each total effects model showing evidence of a racial/ethnic disparity relative to whites at P < .05, we subsequently estimated the direct and indirect effects of race/ethnicity with Tchetgen Tchetgen's inverse odds weighting approach (IOW) 44 as described by Nguyen 2015's practical guidance. 45 IOWs were derived by fitting separate polytomous regression models estimating the relationship between race/ethnicity and each neighborhood characteristic mediators with covariate adjustment, and using the resulting coefficients to calculate an IOW for each observation in the nonwhite racial/ethnic groups ("treatment" groups), while each whites observation was given an IOW of 1 ("control" group). We then fitted inverse odds-weighted Cox regression models to estimate direct effects of race/ethnicity on mortality while controlling for the same demographics and comorbidities from the total effects model.
Next, we calculated the indirect effects of race/ethnicity on mortality by subtracting the direct effect coefficient from the total effect coefficient for racial/ethnic group disparities of interest, and bootstrapped estimates to obtain standard errors and 95% confidence intervals. We considered indirect effects statistically significant at P < .05 as evidence of mediation. Finally, we transformed the total, direct and indirect effect coefficients into hazard ratios and transformed the indirect effect coefficients into proportions of the total effect explained by the neighborhood characteristic mediator of interest (indirect effect/total effect).

| Moderation analysis
We examined whether neighborhood deprivation or neighborhood segregation modified racial/ethnic mortality differences. We modeled the neighborhood-level modifiers in separate Cox regression models that included a product term between the moderator of interest and race/ ethnicity, while controlling for all covariates described above in model 3's specifications. If the interaction term was statistically significant at P < .05, we considered the neighborhood characteristic to be a mediator.

| Sensitivity analyses
We conducted two sensitivity analyses. First, urban-rural differences may be important to consider because AI/ANs are more likely to reside

| Sensitivity analyses
Stratifying the black segregation mediation analysis by rurality, revealed that black segregation trended toward mediating AI/AN

| Stratified mediation analyses
Based on our initial findings that black segregation mediated AI/AN disparities and attenuated AI/AN disparity in more black-segregated communities, we conducted supplementary mediation analysis of black segregation, stratified by high (>33 percent black isolation) and low (≤33 percent black isolation) black segregation counties. We determined high and low segregation based on the level of black isolation where AI/AN and white mortality hazard ratios were equivalent. We found that black segregation mediated AI/AN disparities among Veterans living in counties with low black segregation (indirect effect HR = 1.04, 95%CI:1.01-1.06, P-value < .01), but not in high black-segregated counties (Table S4).

| D ISCUSS I ON
We built upon prior work on racial/ethnic mortality disparities among VHA-users 2,4 to characterize racial/ethnic mortality differences and then explore the role of two neighborhood characteristics-SES and segregation-in explaining residual disparities within this population with access to health care. Overall, we found allcause mortality disparities among AI/ANs only, relative to whites, after accounting for individual-level social determinants and health differences, whereas we found mortality advantages for most other racial/ethnic minority groups. This AI/AN disparity appeared to be both partially mediated and moderated by black segregation and mediated by AI/AN segregation, but not by neighborhood deprivation or Hispanic segregation.
Studies among VHA-users have identified some conditions and disease area where blacks experience similar or better mortality outcomes than whites. 2,4,7,8 However, we found that black VHA-users still experienced higher mortality after accounting for age and sex differences. However, with further adjustment for individual SES and rurality, blacks had a mortality advantage relative to whites.
Individual SES and rurality are both important social determinants related to mortality. 47,48 That black mortality disparities no longer existed after we accounted for these social determinants suggests that they may contribute to black-white mortality differences within this population with health care access.
There is a continual and pressing need to address AI/AN mortality disparities, which are also well-documented in the US general population, as this population has endured a legacy of historic conflict, trauma, and injustice. 24 European and American colonizers forcibly removed AI/ANs from their native lands and deliberately sought to destroy their cultural practices, spiritual beliefs, and family systems. 23,24 The effects of these racist policies and practices endure today, resulting in high levels of discrimination, poverty and emotional and physical trauma among AI/ANs, 23 and AI/AN reservations located in rural areas with few economic opportunities and resources. 23 In this sample with health care access, AI/AN-white mortality disparities existed, even after accounting for individual SES, rurality, and medical and mental health differences, which prompted us to explore the role of neighborhood-level factors. To our knowledge, this study is the first to examine neighborhood-level mediators of AI/AN all-cause mortality disparities.
We found that black residential segregation both moderated and partially mediated AI/AN VHA-user mortality disparities.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.