Cumulative Inequality at the End of Life?: Racial Disparities in Impairment in the Time Before Death

Abstract

Objectives

Drawing from cumulative inequality (CI) theory, the current study examined racial disparities in impairment as individuals approached death to determine whether proposed mechanisms hypothesized to fuel or diminish racial disparities at late ages were at work at the end of individualized life spans.

Method

Black–white disparities were analyzed among decedents using latent growth curves based on the data from the North Carolina Established Populations for Epidemiologic Studies of the Elderly (EPESE) (N = 1,926).

Results

Consistent with previous literature, racial inequalities in functional disparities diminish at late ages. However, significant black–white disparities emerge as older adults approach death, exponentially increasing within the 2 years immediately preceding death. Further, these disparities are not fully mediated by socioeconomic status.

Discussion

The results confirm that CI in health outcomes is observable in late life among individual life spans, suggesting the years surrounding death may be a particularly vulnerable period for health inequality. Future research should examine how advantaged statuses translate to increased access to health-related resources that aid in maintaining greater functional independence until the last stage of life.

Despite technical and medical advances, health disparities in morbidity, functioning, and mortality between blacks and whites in adulthood remain substantial in the United States (Williams & Mohammed, 2009). These inequalities in health are particularly interwoven with mortality, obscuring our understanding of disparities at late ages (Hayward & Heron, 1999). Research on racial health disparities consistently shows patterns of growing or accumulating health inequalities across early adulthood and midlife that tend to stabilize or converge in late and very late stages, sometimes resulting in reduced disparities (Brown, 2018). The underlying processes diminishing these disparities is not well understood, however, as a number of hypothesized mechanisms have been suggested but few have been clearly tested at the last stage of life.

Research examining racial inequalities in health is often situated in theories of cumulative processes, including cumulative inequality (CI) theory. Although changes in disparities have been documented at various adult ages, they have yet to be examined as individuals approach death. Although decedents are not representative of the older adult population, examining health disparities as individuals approach the end of life provides an alternate metric to understand individual or biological aging in later life and a useful counterpoint to the bulk of studies documenting observed disparities among those surviving to each chronological age (Wolf, Freedman, Ondrich, Seplaki, & Spillman, 2015). Further, the end of life is a stage noted for exponential decline in multiple domains of health and functioning (Cohen-Mansfield, Skornick-Bouchbinder, & Brill, 2018), yet empirical research on how disparities are patterned across life has generally neglected this potentially vulnerable life stage.

Utilizing a decedent subsample from the North Carolina Established Populations for Epidemiologic Studies of the Elderly (EPESE) and employing latent growth curve (LGC) models, we aimed to reconcile this gap in the literature by examining whether race disparities in disability increase, remain persistent, or decrease in late life not by chronological age, but in years leading to the end of individual lives. Our application of CI also directly tested whether the mechanisms hypothesized to fuel or diminish racial health inequalities throughout life, including the increasingly universal experience of illness at late ages and unequal access to financial resources, are associated with changing health inequality in the vulnerable time surrounding death.

Theoretical Framework

To ground our hypotheses, we utilized CI theory as a framework because it incorporates much of what has been developed in understanding health trajectories over the life course, while also couched in theoretical backgrounds surrounding the social and temporal nature of disadvantage and how it unfolds over time (Ferraro, Schafer, & Wilkinson, 2016). CI theory draws on the life course perspective, cumulative disadvantage, and stress process theories to examine how social inequality leads to differential and cumulative biological processes that are “markers of aging and predictive of senescence” (Ferraro & Shippee, 2009: 334). CI posits that social systems produce inequality in multiple domains that begin early in life, such as racial minority status and other early life markers of disadvantage such as poverty, to produce differential exposure to risks and resources as individuals age.

Due to historically embedded institutionalized discrimination in the United States, the minority status of black individuals, working alongside low socioeconomic status, are viewed as primary fundamental causes of health disparities that start in early life (Phelan and Link, 2015) and these statuses yield increasing disparities between the advantaged and disadvantaged over the life course. The premature “aging,” or health deterioration, of minorities and disadvantaged individuals is often found in empirical studies of health disparities across mid and later adulthood (Brown, 2018). Age, both chronological and biological, therefore reflects social position and lived experiences, in addition to accumulating inequality (Ferraro & Morton, 2018).

Although a number of theories related to cumulative processes have been used to situate racial disparities across the life course (Dannefer, 2003, Geronimus, Hicken, Keene, & Bound, 2006), CI is distinct as a framework because it is embedded strongly in gerontological research. It posits that along with accumulated disparities across the life course, the late and very late ages may be a time of decreasing disparities. This observed phenomenon, often referred to as age-as-leveler, is presumed to be primarily the work of selective mortality, which is itself a product of accumulating health inequality across adulthood and into later life (Dupre, 2007; Taylor, 2008). However, other theoretical and substantive literatures surrounding this issue posit other potential forces including the increasingly universal experience of health deterioration in late life, which may drive decreasing disparities (Abramson, 2015; Brown, O’Rand, & Adkins, 2012; Hoffmann, 2011).

The following section outlines the supportive and relevant literature surrounding our specific hypotheses, with connections to CI as well as previous empirical findings. Although a wealth of research has documented changing disparities across ages and life stages, to our knowledge, no study has yet to examine disparities and the major hypothesized drivers as individuals approach the end of individual lives.

Racial Disparities and Patterns of Inequality

Comprehensively, empirical studies have shown black–white disparities in health trajectories begin widening in early adulthood, only to begin tapering during mid to later life (65+) (Ferraro & Farmer, 1996; Kim & Miech, 2009) to resemble a persistent, but no longer growing, inequality (Brown, 2018; Kelley-Moore & Ferraro, 2004; Taylor, 2008). Although some research has shown an observed crossover—where blacks have better health outcomes than their white counterparts at very late ages (Johnson, 2000)—most research has shown disparities diminish to nonsignificance between age 70 and 80, consistent with an age-as-leveler argument (Brown, 2018; Kim & Miech, 2009). These findings are generally consistent with theoretical underpinnings of CI across adulthood, that social inequalities increase health disparities across adulthood. However, they are also consistent with a tertiary assertion of CI based on theory and a breadth of empirical work, that very late life may be a period of reducing inequalities at least in part driven by reduced heterogeneity in the older population as they reach late ages. Therefore, although racial disparities in functional impairment will likely be observed throughout most of adulthood, we hypothesize that black–white disparities in impairment trajectories will begin diminishing toward the end of life at late ages.

Findings that support diminishing racial disparities have stood in opposition to literature on health inequality focused specifically on the end of life instead of chronological age. Glaser and Strauss (1968) found that substantial differences exist in functional health trajectories leading up to death, determined primarily by the course of specific diseases (i.e., cancer). More recent empirical research has emphasized the steep rise in impairment in the year or two prior to death (Cohen-Mansfield et al., 2018; Gill, Gahbauer, Han, & Allore, 2010; Verbrugge, Brown, & Zajacova, 2016). Further, racial and social inequalities in impairment, morbidity, and other health-related quality of life indicators may be particularly pronounced near the end of life (Liao, McGee, Cao, & Cooper 1999; Schoeni, Freedman, & Wallace, 2001; Wolf et al., 2015). Black individuals experienced higher mortality from both illnesses thought to lead to a typological trajectory of sudden death without disablement (e.g., heart attack) and illnesses where disablement is high at death (e.g., cancer) (Kochanek, Arias, & Anderson, 2013), suggesting more research is needed to understand any observed disparities surrounding the end of life, since proximate mechanisms of disadvantage may be particularly complex.

Although much of the literature focused on how health disparities are patterned across chronological age finds evidence of leveling, the primary focus of CI is on how social systems differentially promote risks and resources across the life course. Beyond chronological age (or years from birth), research suggests that examining individual life spans as they approach the end of life (years before death) may reveal a particularly vulnerable time for health and functioning, and one characterized by substantial inequality. The end of life in older ages is often characterized by greater interactions with health and other formal care structures rooted in systems of inequality (Fiscella & Sanders, 2016), and exponentially increasing health-related costs (Fassbender, Fainsinger, Carson, & Finegan, 2009). This in turn may promote inequality in later life health outcomes in the years approaching death. As an alternative to our first hypothesis, we hypothesize that racial disparities in impairment will be either persistent or cumulative (growing) between black and white older adults as they approach the end of life.

Hypothesized Mechanisms Underlying Disparities

Although CI theory primarily references selective mortality as the underlying mechanism diminishing health disparities in late life, other potential mechanisms underlying changes in black–white health inequality across the life course have also been proposed, including age-related health deterioration and a leveling of financial and health care resources through programs life Social Security and Medicare (see Brown, 2018). One of these three mechanisms is assumed to be a true function of chronological age (social programs), while the other two are assumed to be a function of either differential (selective mortality) or universal (health deterioration) processes of aging or senescence.

Aside from selective mortality, the most discussed mechanism in the recent literature is age-related health deterioration (Abramson, 2015; Hoffmann, 2011; Yang & Lee, 2010). At late ages, the biological effects of aging may become more universal, serving to reduce observed health disparities. Taylor (2010) found that this mechanism is supported by findings in very old samples, such that the impacts of important social predictors such as education tend to diminish. Research supports the increasing universality of health deterioration as a shared experience among older adults (Abramson, 2015) and suggests that chronic illness is an appropriate indicator of the deterioration process that may fuel age-as-leveler effects (Hoffmann, 2011). Chronic conditions are a substantial part of the disablement process (Verbrugge & Jette, 1994; Wolf et al., 2015), and are implicated in mortality selection (Zajacova & Burgard, 2013). Therefore, chronic conditions—which are disproportionately experienced by disadvantaged individuals earlier in the life course (Liao et al., 1999; Yang & Lee, 2010) but are assumed to act more universally in late ages (Abramson, 2015; Hoffmann, 2011)—may serve as a proxy for age-related health deterioration, effectively suppressing disparities that may otherwise be observable. Put simply, as individuals tend to accrue health deterioration in late ages, the biology of illness becomes more predictive of functioning and death compared to markers of life course disadvantage such as race or socioeconomic position. Since controlling for these conditions in statistical models would provide a picture of racial differences, assuming comparable health deterioration across individuals, we hypothesize that racial disparities in impairment will increase rather than decrease when chronic conditions are held constant.

For a number of reasons, another important mechanism to consider in understanding racial health disparities at various life stages is socioeconomic status. First, CI theory suggests that disadvantage in one social status may spill over or interact with another in the accumulation of risk throughout the life course. The institutionalized discrimination fueling racial health disparities is thought to work substantially through reduced socioeconomic status of black individuals, although many studies find race disparities persist when SES is included as a control (Phelan and Link, 2015). Second, the end of life is a vulnerable life stage particularly marked by high health care needs and spending (Fassbender et al., 2009) such that the ability to mobilize financial and health-related resources may be particularly salient in reducing inequalities promoted by the social and health systems in the United States. In sum, we hypothesize that socioeconomic status will reduce racial disparities in impairment, but they will remain visible and significant.

Method

Data

The data source used to test these hypotheses is the North Carolina Established Populations for Epidemiologic Studies of the Elderly (EPESE). The EPESE project was sponsored by the National Institute on Aging (NIA) and was conducted at multiple sites to investigate predictors of mortality, chronic disease, impairment, and service utilization. The sample consists of individuals aged 65 years and older residing in the community at baseline. Residents were sampled from five counties in North Carolina to reflect both urban and rural populations. Respondents participated in four in-person and four telephone interviews. The baseline survey was conducted in 1986–1987, with seven in-person and telephone waves collected annually, with the final wave in 1996. The original baseline sample includes 4,162 individuals. Only 26 respondents self-reported a race other than black or white, and were dropped from our analysis. These data are especially effective for testing the above hypotheses for a number of reasons. First, these data include a large oversampling of black older adults (54% at baseline) which is crucial to any study examining racial disparities in later life where selective mortality is known to be at work, since small proportional (e.g., black) samples can lead to a lack of statistical power in determining race disparities (Taylor, Lynch, & Urena, 2018). Second, these data follow individuals to oldest old ages (100+) and have a substantial portion of decedents, making the examination of health disparities in the time before death more representative of late ages (70+), where leveling is most likely visible. Finally, the dependent variable was captured annually until the final wave making it straightforward to restructure the data for our analyses in years from waves following the baseline survey (b) such that time points originally reflected b, b+1, b+2, etc. to waves structured as preceding the year of/before death (d) such that time points for the analysis reflect d, d-1, d-2, etc.

Measures

Outcome Measures

The main dependent variable, physical impairment, was a count of Activities of Daily Living (ADL) (Katz, Ford, Moskowitz, Jackson, & Jaffe, 1963). The items measure the ability to perform seven activities: walking across a room, eating, bathing, dressing, grooming, transferring from the bed to a chair, and using the toilet. Mortality was captured by a linked date of death variable from the National Death Index in 1999 after the study was completed (see Fillenbaum, Burchett, & Blazer, 2009). This variable was used to restructure the data to examine disparities leading up to death.

Covariates

The key independent variable was race (1 = black, 0 = white). Chronic conditions tapping into age-related health deterioration included heart attack, hypertension, stroke, diabetes, and hip fracture. Dichotomous measures indicate whether the respondent reported the disease or health event over the entire survey period. Indicators of body mass index (BMI) and smoking were also included, since they are strongly linked to disparities and rising impairment rates (Seeman, Merkin, Crimmins, & Karlamangla, 2010). The respondent’s BMI was taken at baseline, with dichotomous variables capturing underweight (BMI < 18.5), overweight (BMI = 25–29.9), obese (BMI = 30–39.9), and extremely obese (BMI = 40+) (referent = normal weight). Smoking was measured at baseline as a cumulative lifetime measure of pack years, by multiplying the number of years a respondent reported smoking by the number of cigarettes (in packs per year) they reported currently or previously smoking on average.

Socioeconomic status included measures of education, income, and home ownership. Education (in years) was included from the baseline survey. Income was a continuous variable reflecting annual income from all sources in categories ranging from 1 ($0–$1,999) to 13 ($75,000 or more). The average income was computed across the survey period, and missing items (<10%) were imputed using the baseline income variable (cleaned) provided in the EPESE data. We measured assets as one dichotomous indicator of whether the respondent owned their own home, averaged across waves such that the continuous variable represents the proportion of waves the respondent reported home ownership.

Control variables include age at death measured in years and centered on its mean for all analyses. Respondent’s sex (1 = female, 0 = male) was also included. We undertook sensitivity analyses (models not shown) including variables for proxy response, widowed, and health care access with similar findings.

Analytic Strategy

To restructure our data, we restricted the sample (N = 4,162) to black and white respondents reporting (nonmissing) on functional impairment at baseline, resulting in a loss of 36 cases. The sample was also restricted to those who died between baseline (1986–1987) and the year after the survey was completed (1997), for which an age at death and some measure (nonmissing value) of ADL impairment before death could be recorded (N = 2,201). Listwise deletion of missing independent variables was used, yielding a final sample of 1,926 (12% missing). Missingness in the dependent variable was modeled with a Full Information Maximum Likelihood (FIML) estimator. The data was restructured such that each individual’s closest measure (within 1 year) before death represents the intercept of their trajectory, or the mean number of ADLs within the year of death on average. Trajectories were then estimated using annual waves as the time metric, such that each wave constitutes another wave or year before death. In this way, the models can be interpreted as disparities in later life as they progress leading up to the year of death (the intercept). Hypotheses considering mechanisms were tested using this restructured data, adding additional control variables and possible explanatory mechanisms to nested models.

LGC models in a structural equation modeling (SEM) framework (Bollen & Curran, 2006) were used to estimate intra-individual trajectories of physical impairment leading up to death. An individual’s longitudinal responses on the dependent variable y_it for respondent i at time t is modeled as a function of time. The general level 1 equation estimating the trajectories for each individual may be written as:

$$Y_{it}={\alpha }_i+{\beta }_{1i}{\lambda }_t+{\beta }_{2i}{\lambda }_t^2+{\epsilon }_{it}$$

Both linear and quadratic forms of the latent growth curve were tested, with the quadratic yielding substantially better fit. Thus, the impairment measures over time, $Y_{it}$ were modeled as a function of both linear ${\lambda }_t$ and quadratic ${\lambda }_t^2$ terms. The linear factor loadings were set to 0, −1, −2, −3, −4, −5, and −6. The negative factor loadings were chosen for ease of coefficient interpretation, as positive loadings with the same integers yielded identical estimated plots but coefficients with reversed direction. Here, the intercept represents the individual’s last reported time point before death (d) and the trajectories move backward in time (d, d-1, d-2, d-3, etc.). Because factor loadings were set to be negative, positive coefficients in the linear and quadratic slopes refer to increasing disparities with each year leading up to death. For more on alternative metrics of time in LGCs and specifying factor loadings, see Bollen and Curran (2006). The coefficients ${\alpha }_i$${\beta }_{1i}$ and ${\beta }_{2i}$ represent the intercept or mean level of impairment, the linear growth rate, and the quadratic growth rate, respectively. The second level of the model examined the variation of each individual trajectory from the mean trajectory as a function of the covariates. The first model in our analysis examined variation by race (black), controlling for age at death. The general level 2 equations are written as follows. For the latent intercept:

$${\alpha }_i={\mu }_a+{\gamma }_{\alpha 1}black+{\gamma }_{\alpha 2}age+{\zeta }_{{\alpha }_i}$$

For the linear growth rate:

$${\beta }_{1i}={\mu }_{{\beta }_1}+{\gamma }_{{\beta }_{1}1}black+{\gamma }_{{\beta }_{1}2}age+{\zeta }_{{\beta }_{1i}}$$

For the quadratic growth rate:

$${\beta }_{2i}={\mu }_{{\beta }_2}+{\gamma }_{{\beta }_{2}3}black+{\gamma }_{{\beta }_{2}4}age+{\zeta }_{{\beta }_{2i}}$$

The subsequent models in the analysis add additional covariates, therefore the equations presented above can be expanded to include the additional effect coefficients $\gamma$ corresponding to each additional covariate. Sensitivity analyses using alternative group trajectory methods (see Gill et al., 2010) yielded similar findings to those presented, but the LGC’s proved more parsimonious in presenting results.

Results

Table 1 presents descriptive statistics from our analytic sample. Black individuals made up slightly more than half of the sample and the mean age at death was 81 years old. ADL level impairment findings suggest that disablement was low on average but accumulated for both black and white respondents as they approached the end of life, with a disparity that is significant in the year before death. There were also noted racial disparities in a number of chronic conditions, where black individuals had higher rates of hypertension and diabetes, but lower rates of heart attack, cancer, and hip fracture. Black individuals were less likely to be underweight and had lower cumulative smoking but were more likely to be obese. Black individuals were also significantly more disadvantaged across the socioeconomic indicators (education, income, home ownership). Black individuals had almost three fewer years of education on average compared to whites and the difference in income categories by race suggests that white individuals reported substantially more financial resources, translating to annual incomes roughly 1.7 times higher than their black counterparts over the survey period.

Table 1.

Sample Characteristics, North Carolina Established Populations for Epidemiologic Studies of the Elderly, 1986–1996 (N = 1,926)

	Total (N = 1,926)	Blacks (N = 1,038)	Whites (N = 888)	Diff.
Black	53.89%	N/A	N/A	N/A
Female	59.87%	59.83%	59.91%	---
Age at Death (Centered, Mean = 81.38)	0.00 (7.46)	−0.26 (7.61)	0.30 (7.28)	---
Heart Attack	24.56%	21.10%	28.60%	***
Hypertension	63.97%	68.88%	58.22%	***
Stroke	20.20%	20.33%	20.05%	---
Cancer	20.82%	13.29%	29.62%	***
Diabetes	26.64%	31.50%	20.95%	***
Hip Fracture	9.40%	5.59%	13.85%	***
Underweight	5.87%	4.82%	7.09%	*
Overweight	35.25%	36.32%	34.01%	---
Obese	16.46%	21.87%	10.14%	***
Extremely Obese	1.25%	2.02%	0.34%	**
Smoking (pack-years)	18.51 (28.94)	14.01 (23.82)	23.77 (33.22)	***
Education (years)	8.01 (4.03)	6.75 (3.84)	9.47(3.74)	***
Average Income	5.31 (2.10)	4.59 (0.05)	6.15 (0.08)	***
Own Home	0.58 (0.47)	0.52 (0.01)	0.65 (0.02)	***
ADL Before Death 1 (N = 1,926)	1.86 (2.56)	1.99 (2.58)	1.71 (2.51)	**
ADL Before Death 2 (N = 1,768)	1.30 (2.13)	1.33 (2.10)	1.27 (2.16)	---
ADL Before Death 3 (N = 1,602)	1.02 (1.93)	1.05 (1.90)	0.98 (1.95)	---
ADL Before Death 4 (N = 1,410)	0.81 (1.70)	0.81 (1.66)	0.81 (1.74)	---
ADL Before Death 5 (N = 1,254)	0.57 (1.41)	0.54 (1.31)	0.60 (1.52)	---
ADL Before Death 6 (N = 1,023)	0.50 (1.27)	0.49 (1.19)	0.51 (1.37)	---
ADL Before Death 7 (N = 772)	0.37 (1.07)	0.38 (1.04)	0.37 (1.11)	---

Note: The fourth column indicates significant difference between blacks and whites using t-tests for continuous variables and chi-square tests for categorical variables. Activities of Daily Living (ADL) is a count variable measuring the ability to perform seven activities: walking across a room, eating, bathing, dressing, grooming, transferring from the bed to a chair, and using the toilet (Katz et al., 1963).

*p < .05, **p < .01, ***p < .001.

Table 2 presents the test of our first competing hypotheses concerning a diminishing or growing racial disparity as individuals approached death. Here, positive coefficients for the intercept reflect increased impairment in the year before death while positive coefficients for the slopes reflect growing disparities in years leading to death. On average, white individuals died with a mean of 1.64 ADL items and black individuals died with almost one-third (0.28) of an ADL limitation more. Further, this significant disparity increased in the years leading up to death in an exponential fashion (represented by positive and significant linear and quadratic terms), consistent with a pattern of cumulative disadvantage in the time before death for blacks. Figure 1 presents the estimated average trajectories for white and black respondents for the mean age at death. Here, the predicted values are plotted as individuals approach death such that the intercept is represented as the last time point. In late ages (mid-70’s on average) there was no visible disparity in impairment by race. However, the 2–3 years surrounding death show support for CI, such that black individuals faced increasing disablement in the time surrounding death.

Table 2.

Trajectories Before Death, North Carolina Established Populations for Epidemiologic Studies of the Elderly, 1986–1996 (N = 1,926)

Variable	Intercept	Linear	Quadratic
Black	0.277*	0.138**	0.015*
Age at Death	0.072***	0.013***	0.000
Mean	1.638***	0.418*	0.037
Variance	4.231***	0.439***	0.007***
Chi-square (df)	156.431***	(27)
CFI	0.974
TLI	0.966
RMSEA	0.050
BIC	34,171.611

Note: CFI = Comparative Fit Index; BIC = Bayesian information criterion; RMSEA = Root Mean Square Error of Approximation; TLI = Tucker-Lewis Index. Trajectories are estimated using the decedent sample, controlling on age at death.

*p < .05, **p < .01, ***p < .001.

Figure 1.

Predicted ADL disability trajectories for whites and blacks, North Carolina Established Populations for Epidemiologic Studies of the Elderly, 1986–1996 (N = 1,926). Note: Wave 7 represents year furthest from death, Wave 1 represents year before death.

Table 3 presents tests of the other mechanisms thought responsible for age-as-leveler and CI findings, namely age-related health deterioration and socioeconomic status. Model 1 included gender as a control, with the race coefficients changing very little. Model 2 focused on the role of health deterioration (chronic conditions, BMI, and smoking) in leveling the race disparity in impairment trajectories in the time before death. Consistent with expectations suggesting the increasingly universal experience of health deterioration in very late life, the inclusion of covariates in Model 2 increased the magnitude of the intercept and slope coefficients by 42, 41, and 47%, respectively. This suggests that holding health deterioration over the survey period constant, the racial disparity in impairment would be even more sizeable in the year before death. Specifically, the most disabling conditions were stroke and hip fracture, and these were especially disabling in the year before death. The prevalence of stroke in this sample was roughly 20% for both blacks and whites alike. However, there was a significant disparity in hip fractures, with whites more than twice as likely to report hip fracture. Although not significant in Table 1, blacks had higher prevalence of overweight in our sample, which was associated with a decrease in the impairment intercept and linear slope in Model 2. In sum, although black individuals had a significantly higher prevalence of most diseases in this sample, whites reported similar or higher prevalence of the most disabling diseases before death at late ages, which may underlie the increased magnitude of the racial disparity once conditions were included in the model. Black individuals also had lower prevalence of risky health profiles at these ages, including cumulative smoking and underweight.

Table 3.

The Effect of Cumulative Inequality on Disability Trajectories at the End of Life, North Carolina Established Populations for Epidemiologic Studies of the Elderly, 1986–1996 (N = 1,926)

	Model 1			Model 2			Model 3
Variable	Int.	Lin.	Quad.	Int.	Lin.	Quad.	Int.	Lin.	Quad.
Black	0.274*	0.138**	0.015*	0.388***	0.195**	0.022**	0.257*	0.156*	0.017
Age at Death	0.066***	0.013**	0.000	0.062***	0.011**	0.000	0.060***	0.010*	0.000
Female	0.458***	0.043	−0.001	0.313**	0.025	−0.001	0.195	0.012	0.000
Heart Attack				0.013	0.116†	0.023*	−0.009	0.109†	0.022*
Hypertension				−0.161	−0.074	−0.009	−0.160	−0.072	−0.009
Stroke				1.625***	0.277***	0.015	1.620***	0.275***	0.014
Cancer				0.070	0.082	0.012	0.095	0.094	0.014
Diabetes				0.344**	0.057	0.003	0.332**	0.055	0.003
Hip Fracture				1.621***	0.091***	0.025†	1.597***	0.375***	0.025†
Under Weight				0.328	0.021	0.002	0.295	0.014	0.001
Over Weight				−0.293*	−0.128*	−0.015†	−0.301†	−0.129*	−0.015†
Obese				−0.165	−0.069	−0.007	−0.178	−0.075	−0.008
Very Obese				0.477	−0.235	−0.044	0.344	−0.247	−0.044
Smoking				−0.004†	−0.001	0.000	−0.003†	−0.001	0.000
Education							0.006	−0.005	−0.001
Income							−0.080*	−0.018	−0.001
Own Home							−0.221†	−0.004	0.003
Mean	1.366***	0.392***	0.038***	1.079***	0.352***	0.036**	1.731***	0.514***	0.055**
Var.	4.184***	0.440***	0.007***	3.431***	0.400***	0.007***	3.391***	0.396***	0.006***
Chi-square (df)	159.008***	(31)		235.038***	(58)		246.769***	(67)
CFI	0.974			0.970			0.970
TLI	0.965			0.952			0.952
RMSEA	0.046			0.033			0.031
BIC	34,170.37			34,082.79			34,126.61

Note: CFI = Comparative Fit Index; BIC = Bayesian information criterion; RMSEA = Root Mean Square Error of Approximation; TLI = Tucker-Lewis Index.

^† p < .10, *p < .05, **p < .01, ***p < .001.

Model 3 introduced socioeconomic resource variables to examine financial mechanisms of increasing or decreasing inequality in the time before death. The inclusion of these variables decreased the magnitude of the race disparity (34, 20, and 23%, respectively). However, the disparity for the intercept and linear slope remained significant, suggesting a significant racial disparity remained in the time before death. A reduction in model fit suggests the SES measures overall do not have strong explanatory power in Model 3. Surprisingly, only income was significant for the intercept. This suggests that financial resources that can be mobilized in the last years of life may be particularly important in impairment trajectories. The SES findings, along with sensitivity analyses alternately examining supplemental health insurance (models not shown), suggest that the availability of financial resources, and especially for health-related needs, is both salient and unequally distributed in the last stage of life. It is also notable that education was not significant in the time leading to death, suggesting support for a leveling of early life achievement effects in the last years of life. This finding has been replicated in other studies examining advanced late life (Taylor, 2010).

Discussion

Longitudinal research has predominantly examined later life health at some (often arbitrary) time point or chronological age reference, following individuals forward until death, attrition, or study completion. Our approach was to use individual mortality as the point of reference and to follow individuals as they aged up to this point. This alternate approach is important since individuals, and especially those in disadvantaged groups, are likely to experience chronological age very differently.

Previous research has often found that health disparities increase between black and white individuals across adulthood. However, late life is a time where observed disparities often diminish. The hypothesized mechanisms fueling leveling or increasing disparities are not well understood or supported by empirical research focused on the end of life. In examining racial disparities among decedents as individuals approach and experience the last stage of life, we found no evidence of a significant disparity more than 5 years prior to death, which is likely due to selective mortality at earlier ages. However, significant health disparities were visible and significant in the last few years of life. These findings reconcile conflicting literatures on diminishing disparities in later ages and substantial disparities at the very end of life. They also suggest future research should conceptualize the end of life as a particularly vulnerable period. Informing the CI framework (Ferraro & Shippee, 2009), our findings show that disadvantaged statuses translate to growing health inequality at particular life stages even while potentially being unobserved at specific ages. As CI currently speaks to age-as-lever effects, our findings suggest we should not simply caution researchers regarding selective forces in cohorts or samples, but rather emphasize the processes of aging as individualized and not merely chronological. Our findings suggest that even when selective forces and universalizing mechanisms have led to observable leveling at late ages, substantial inequality remains unexplained in the few years preceding death. Further, reorienting individual health trajectories by years leading to death among decedents is a useful alternative in understanding health disparities by individual rather than chronological age. We also found some support for the leveling mechanism of health deterioration (Abramson, 2015; Hoffmann, 2011) as measured by chronic disease and health-related variables. Net of health profiles, the racial disparity in impairment increased, primarily by hip fracture. Additionally, our early life socioeconomic marker (education) diminished in its effects for health approaching the end of life and only income in the years leading to death was protective for impairment and reduced the racial disparity. Alternate analyses examining supplemental health insurance alone showed similar results, suggesting financial resources directly targeted to health needs (access to medication, devices, and home modification) are particularly important in reducing impairments in the last stage of life. This is consistent with research suggesting the end of life is a costly period with high interaction with economic and health care structures that may promote inequality. We note that in sensitivity analyses, income was the only variable with a significant interaction with race, suggesting diminished returns from financial resources among black individuals. Future research should further examine modifiable factors that may reduce disparities at this vulnerable period of life.

After accounting for health deterioration and socioeconomic status, we found that disparities in impairment remained significant in the last years of life, with black individuals experiencing roughly ¼ of an ADL impairment more than their white counterparts. This suggests that quality of care and access to mobility and self-care related resources should be examined in future research, especially in relation to the vulnerable time surrounding the end of life. Although we believe these findings speak to a longstanding but conflicted literature on health disparities over the life course, there are limitations worth noting. First, our analytic sample included individuals dying during the decade covered in the NC EPESE. Although roughly half the baseline sample was included, and our results were consistent with other studies of decedents and those imputing values for survivors (Wolf et al. 2015), it is important to note that our sample was not representative of the entire population. We believe this sample was uniquely suited for our research questions, however, as the yearly data collection and matched mortality data made the estimation of trajectories in years leading up to death more comparable to studies based on chronological age. Further, a strength of our data was the large proportion of black respondents, allowing us to account for substantial variability in education and income. This oversampling and variability within a minority group provided clearer tests of disparities and mechanisms where percentages of minorities often dwindle at very late ages. Our decedent sample primarily represents those dying at late ages; samples better representing decedents in mid and early late life may produce alternate findings.

Finally, the last wave (1996) had a longer follow-up period than the others (3 years), but sensitivity analyses suggest this did not alter the findings. Although our findings only speak to disparities at the end of life in late ages, it is likely our results were conservative given the leveling that was likely to occur prior to our observation window. Examining the end of life in early and middle adulthood would likely yield different findings, as impairment, illness, and the most common causes of death are different at other life stages.

Despite these limitations, this study has made an original contribution to the research literature by examining health disparities in late life, focusing on individual trajectories as they approach death. Our findings inform a broad literature on health disparities across the life course and the fifth axiom of CI theory, particularly in suggesting the time surrounding death may be uniquely characterized by accumulating inequality even in the context of late ages. Our findings also have policy implications, since health resources and interventions at late life stages may partially alleviate growing health disparities between white and black older adults as they navigate the last few years of life. Notably, our findings suggest advantaged statuses translate to resources that aid in maintaining greater functional independence up until the very end of life. Accordingly, policymakers must strive to make greater and more equitable investments in health-related resources that play a prominent role in reducing health disparities at the last stage of life.

Funding

The research on which this article is based was performed pursuant to Contract Number N01-AG-1-2102 with the National Institute on Aging, in support of the Established Populations for Epidemiologic Studies of the Elderly (Duke University). The first author was supported by National Institute on Aging (grant R00AG030471). The second author was supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. 2016-1449440. Any opinions, findings, and conclusions or recommendations expressed in this study are those of the authors and do not necessarily reflect the view of the National Science Foundation.

Conflict of Interest

None reported.

Author Contributions

M. G. Taylor designed the study, performed analyses, and wrote the paper. S. Min and K. Reid wrote the background sections of the paper and assisted in revisions.