Differences in Social Determinants of Health Underlie Racial/Ethnic Disparities in Psychological Health and Well-Being: Study of 11,143 Older Adults

Abstract

Objective:

To determine the impact of selected social determinants of health (SDoH) on psychological health and well-being (defined as depression, cognition, self-rated health) among Black and Hispanic/Latinx adults relative to White adults aged 51 to 89.

Methods:

We measured disparities in depressive symptomatology, cognition, and self-rated health among 2,306 Non-Hispanic/Latinx Black, 1,593 Hispanic/Latinx, and 7,244 Non-Hispanic/Latinx White adults from the Health and Retirement Study (n=11,143). Blinder-Oaxaca decomposition was used to examine whether differences in selected SDoH explained a larger share of the disparities than age, sex, measures of health, health behaviors, and healthcare utilization. Selected SDoH included education, parents’ education, number of years worked, marital status, veteran status, geographic residence, nativity status, income, and insurance coverage.

Results:

Black and Hispanic/Latinx adults reported worse depressive symptomatology, cognition, and self-rated health compared to White adults. Selected SDoH were associated with a larger proportion of the Black-White disparities in depressive symptomatology (51%), cognition (39%), and self-rated health (37%) than did age, sex, measures of health, health behaviors, and healthcare utilization. SDoH were associated with a larger proportion of the Hispanic/Latinx-White disparity in cognition (76%) and self-rated health (75%), but age and physical health correlated with the disparity in depressive symptomatology (28%). Education, parents’ education, years worked, income, and insurance parity were SDoH that associated with the disparities.

Conclusions:

Differences in SDoH underlie racial/ethnic disparities in depression, cognition, and self-rated health among older adults. Education, income, number of years worked, and insurance parity are key SDoH.

Introduction

The tenets of social epidemiology suggest that where and how we live, work, play, and age affect our health and well-being(1). While some social determinants of health (SDoH) are modifiable (e.g., health behaviors) or related to our lived environment (e.g., affordability of and access to healthcare), others are difficult to change in older age (e.g., childhood education quality, nativity status, occupational history, income). It is well-established that Black and Hispanic/Latinx adults in the United States (U.S.) experience poorer health outcomes and have lower healthcare utilization compared to White adults(2, 3). Disparities in one domain of health may intersect with and precipitate disparities in another domain(2), and while some health disparities compound over time (i.e., cumulative disadvantage)(4), others are triggered by a decline in health or are persistent across the lifespan(2). The origins of most health disparities are complex and difficult to disentangle because overt and covert (structural) racism(5) and discrimination have shaped SDoH among racially/ethnically marginalized adults in the U.S.

Black adults are more likely to have multiple chronic diseases than White adults, while Hispanic/Latinx adults are less likely to have multimorbidity than both Black and White adults(6). In analyses from the Health and Retirement Study (HRS), Black adults developed multimorbidity at an earlier age and both Black and Hispanic/Latinx adults were more likely to experience severe and persistent functional disability than White adults(2, 7, 8). In addition to poorer physical health, marginalized older adults often report worse mental health, though Black and Hispanic/Latinx older adults are less likely to be screened for or diagnosed with depression than White older adults(9). This may be due to disparities in mental healthcare initiation and adequacy(10) or beliefs surrounding mental illness, stigma, and preferences for treatment(11). Depressive symptomatology is thought to be greater among Black and Hispanic/Latinx adults compared to White adults aged 50–75; however, this disparity narrows substantially between the ages 76 and 90(12). In addition to depression, Black older adults may be more socially disconnected, have greater perceived isolation, and smaller social networks(13). Some studies suggest that the rate of loneliness is higher among Hispanic/Latinx adults(14), while others do not(15). Deficits in cognition among marginalized adults have been tied to poor early-life educational quality and literacy in addition to other health, psychosocial, and socioeconomic factors(16, 17). While these findings paint a picture of poorer health and well-being among Black and Hispanic/Latinx adults, the Healthy Immigrant Paradox(18) suggests that immigrants and migrants may appear healthier than their native-born counterparts and differences may exist by ethnic subgroup.

Understanding the effect of SDoH on psychological health and well-being (PHWB) by race/ethnicity will be crucial for the development of primordial prevention strategies (e.g., policies targeting socioenvironmental risk factors at an early age to prevent downstream disadvantage and morbidity). Several SDoH are known to affect PHWB (operationalized as depressive symptomatology, cognition, and self-rated health). Education and income indirectly impact health and well-being through health behaviors, diet, environmental exposures, and access to healthcare(19–21) in addition to direct effects such as cognitive reserve(22) and allostatic load(23). Beyond socioeconomic status and healthcare access, other SDoH such as marital status (possible mechanisms: e.g., depressed affect, loneliness, social isolation)(24), veteran status (e.g., allostatic load, post-traumatic stress disorder, substance abuse)(25), nativity status (e.g., acculturation, perceived discrimination)(20, 26), and geographic place of residence (e.g., “Southern” diet in the U.S., access to quality education, healthcare, and jobs)(27, 28) are known to impact health and well-being. What is less understood is whether SDoH explain a larger proportion of the Black-White and Hispanic/Latinx-White disparities in PHWB compared to other well-known correlates of health.

Using data from the HRS, we posit that selected SDoH (i.e., education, parents’ education, number of years worked (NYW), marital status, veteran status, geographic residence, nativity status, income, and health insurance coverage) will explain a larger share of the disparities in depressive symptomatology, cognition, and self-rated health among Black and Hispanic/Latinx adults than age, sex, measures of health, health behaviors, and healthcare utilization relative to their White peers. By identifying SDoH that primarily drive racial/ethnic disparities in PHWB, government agencies and healthcare systems can better allocate resources toward interventions and policies that will appreciably reduce health inequities.

Methods

See the data supplement accompanying the online version of this article for methodological references, detailed descriptions of the study variables, and model specifications.

Participants

The HRS is a publicly-available, prospective cohort study of adults aged 51+ in the U.S. starting in 1992 with assessments occurring every 2 years, funded by the National Institute on Aging (U01AG009740). Black adults were oversampled from the Southern portion of the U.S. and Hispanic/Latinx adults were oversampled from the Western portion of the U.S. We utilized the RAND HRS Longitudinal File because it contained imputed variables (e.g., cognition, wealth/income), which were crucial for modeling. Data came from Wave 13 (2016) of the HRS.

Few adults aged 90 years or older identified as Black or Hispanic/Latinx, thus adults in the Baby Boomer and Silent Generation cohorts were selected (born between 1928–1964). After excluding participants due to their birthyear, those identifying as “Other” race and not Hispanic/Latinx, or those with missingness on the Center for Epidemiological Studies-Depression (CES-D) 8 Scale or Modified Telephone Interview for Cognitive Status (TICS-M) were excluded (Supplemental Figure 1). A final sample size of 11,143 adults was retained as our analytic sample (2,306 Non-Hispanic/Latinx Black, 1,593 Hispanic/Latinx, and 7,244 Non-Hispanic/Latinx White adults). Adults identifying as Black, “Other”, or White and Hispanic/Latinx were categorized as Hispanic/Latinx.

Outcomes and Measures

PHWB was operationalized as depressive symptomatology, cognition, and self-rated health. Depressive symptomatology was measured by the CES-D 8 [Range 0–8]. Global cognition was measured with the total cognition score from the TICS-M [Range 0–30]. Participants were asked to rate their general health from 1 [Excellent] to 5 [Poor].

Selected Social Determinants of Health

Selected SDoH included education (years), mother’s education (years), father’s education (years), NYW (years), marital status (dummycoded as married, married with an absent spouse, or partnered[1], vs. separated, divorced, widowed, or never married[0]), veteran status (yes[1] vs. no[0]), Southern U.S. resident (dummycoded as South[1] vs. or Northeast, Midwest, West, Other U.S. resident[0]), nativity status (foreign-born[1] vs. native-born[0]), annual household income in natural-logged U.S. dollars, employer-sponsored health insurance coverage (yes[1] vs. no[0]), Medicare coverage (yes[1] vs. no[0]), and Medicaid coverage (yes[1] vs. no[0]).

Statistical Analysis

Participant characteristics were reported with descriptive statistics, and unadjusted differences between racial/ethnic groups were examined with ANOVA or chi-square tests. Effect sizes compared Black vs. White and Hispanic/Latinx vs. White adults using Cohen’s d for both continuous and categorical variables (Supplemental Method Material).

In order to determine whether SDoH explained much of the Black-White and Hispanic/Latinx-White disparities in depressive symptomatology, cognition, and self-rated health, two-fold Blinder-Oaxaca decomposition was used with reference coefficients recovered from a pooled regression model that included group membership in the model, as recommended by Jann (2008)(29). Blinder-Oaxaca decomposition estimates the magnitude of the disparity between Black and White participants and then Hispanic/Latinx and White participants. Next, the disparity is decomposed into a characteristics effect (i.e., variance explained by differences in the characteristics of each group) and a coefficients effect (i.e., variance explained by differences in beta coefficients when linear regressions are conducted for each group).

As an example, assume that Black participants were younger than White participants in the HRS. The characteristics effect would determine what proportion of the disparity in cognition was explained by the younger age of the Black participants. The coefficients effect would determine what proportion of the disparity in cognition was driven by differences in the beta coefficients for age. Because no simple explanation exists for why a beta coefficient differs by race/ethnicity, the coefficients effect is often referred to as the “unexplained” effect. In labor economics, the coefficients effect often represents discrimination. However, in our study, the coefficients effect could represent inequivalent effects of the covariates (e.g., one year of education does not have an equivalent effect across groups), testing biases by group (e.g., questions in the CES-D may be culturally-specific), or error variance due to unobserved constructs. In this paper, the results focus on the characteristics effects.

In order to improve interpretability of the results, estimates were converted to shares, which represent the proportion of the disparity that the covariate explains. A positive share for age would suggest that if Black and White adults were the same age, the disparity in cognition would decrease. In contrast, a negative share would indicate that if Black and White adults were the same age, the disparity in cognition would increase. Shares that exceed 100% indicate that the marginalized group would be better off than White adults if characteristics were equivalent. Rather than reporting p-values, 95% confidence intervals were calculated. Share intervals that contained 0% were considered null effects.

Results

See Table 1 for a comparison of Black, Hispanic/Latinx, and White participants. Notably, most differences between Black or Hispanic/Latinx adults and White adults were small (d<0.20) or small-to-medium (0.20≤d<0.50) in effect size. Black adults differed from White adults with a medium (0.50≤d<0.80) or large effect size (d≥0.80) on age, income, and cognition.

Table 1.

Comparisons of racial/ethnic groups

Variables	Race/Ethnicity						Statistical Significance and Effect Size Estimates
Social Determinants of Healtd	White N = 7,244		Black N = 2,306		Hispanic / Latinx N = 1,593		p	Black vs. White Cohen’s d	Hispanic / Latinx vs. White Cohen’s d
	M / %	(SD) / N	M / %	(SD) / N	M / %	(SD) / N
Education (Years)	13.49	(2.53)	12.59	(2.85)	10.07	(4.46)	<.001	0.35**	1.15****
Mother’s Education (Years)	11.03	(3.03)	9.92	(3.58)	6.03	(4.55)	<.001	0.35**	1.50****
Father’s Education (Years)	10.65	(3.63)	8.98	(3.98)	6.20	(4.82)	<.001	0.45**	1.16****
Number of Years Worked	34.82	(16.45)	26.54	(17.73)	21.06	(16.88)	<.001	0.49**	0.83****
Marital Status							<.001
Married or Partnered	63%	4,534	42%	968	59%	943		0.37**	0.06*
Divorced/Separated	13%	917	26%	601	18%	287		0.32**	0.12*
Widowed	21%	1,517	19%	447	16%	252		0.04*	0.10*
Never Married	4%	270	12%	285	7%	107		0.32**	0.12*
Veteran Status	20%	1,478	15%	340	8%	127	<.001	0.12*	0.24**
U.S. Geographic Location							<.001
Northeast	14%	1,002	15%	357	14%	221		0.04*	0.00*
Midwest	28%	1,995	17%	385	5%	84		0.22**	0.41**
South	40%	2,915	60%	1,384	40%	629		0.35**	0.02*
West	18%	1,325	8%	179	40%	642		0.24**	0.41**
Other	<1%	4	0%	0	1%	14		0.02*	0.14*
Nativity Status (Foreign-Born)	5%	326	7%	154	61%	968	<.001	0.08*	1.54****
Annual Household Income: ln(USD)	$10.80	(0.99)	$10.26	(1.16)	$10.13	(1.07)	<.001	0.52***	0.66***
Annual Household Income: median USD	$47,673		$28,975		$25,200
Covered by Current or Previous Employer	21%	1,525	24%	550	18%	275	<.001	0.06*	0.06*
Covered by Any Government Program	84%	6,007	73%	1,665	67%	1,045	<.001	0.22****	0.32**
Medicare	81%	5,829	65%	1,471	58%	902	<.001	0.35**	0.43**
Medicaid	6%	451	19%	434	26%	398	<.001	0.39**	0.52***
Veterans Administration	7%	502	7%	170	4%	55	<.001	0.02*	0.10*
Demographics	M / %	(SD) / N	M / %	(SD) / N	M / %	(SD) / N	p	d	d
Age (Years)	71.43	(9.79)	65.99	(10.46)	65.48	(10.48)	<.001	0.55***	0.60***
Age Group							<.001	0.41**	0.39**
Silent Generation (1928–1945)	76%	4,298	15%	837	10%	544
Baby Boomers (1946–1964)	54%	2,946	27%	1,469	19%	1,049
Sex (% Male)	43%	3,115	40%	918	44%	708	.007	0.06*	0.02*
Spanish Survey Language Preference	<1%	13	0%	0	50%	797	<.001	0.04*	1.76****
Physical Health	M / %	(SD) / N	M / %	(SD) / N	M / %	(SD) / N	p	d	d
Self-Rated Health (1 [Excellent] - 5 [Poor])	2.82	(1.03)	3.10	(1.04)	3.24	(1.07)	<.001	0.27**	0.40**
Body Mass Index	28.23	(5.91)	30.19	(6.60)	29.23	(5.76)	<.001	0.32**	0.17*
# Impaired ADLs	0.26	(0.77)	0.44	(0.99)	0.46	(1.07)	<.001	0.21**	0.23**
# Impaired IADLs	0.23	(0.67)	0.34	(0.85)	0.40	(0.96)	<.001	0.15*	0.23**
# Chronic Conditions (of 8)	2.43	(1.52)	2.37	(1.53)	2.09	(1.55)	<.001	0.04*	0.22**
Hypertension	62%	4,450	76%	1,740	61%	974	<.001	0.26**	0.00*
Diabetes Mellitus	24%	1,722	34%	785	37%	583	<.001	0.20**	0.22**
Cancer/Malignancy^	20%	1,462	13%	310	9%	144	<.001	0.14*	0.22**
Chronic Lung Disease	12%	895	10%	225	7%	109	<.001	0.06*	0.14*
Cardiovascular Disease	30%	2,183	22%	506	18%	289	<.001	0.16*	0.20**
Stroke or TIA	9%	685	10%	234	7%	111	.002	0.02*	0.06*
Psychiatric Disorder	20%	1,455	15%	341	21%	327	<.001	0.12*	0.00*
Arthritis/Rheumatism	65%	4,720	58%	1,326	49%	785	<.001	0.14*	0.26**
Ever Diagnosed with a Sleep Disorder	16%	1,187	16%	359	14%	223	.05	0.02*	0.06*
Mental Health & Cognition	M / %	(SD) / N	M / %	(SD) / N	M / %	(SD) / N	p	d	d
CES-D 8 Total Score	1.29	(1.86)	1.73	(1.97)	1.89	(2.24)	<.001	0.23**	0.31**
Clinical Depression (CES-D 8 Score ≥ 3)	18%	1,331	26%	590	28%	452	<.001	0.16*	0.20**
Felt Lonely (% Yes)	15%	1,121	20%	453	22%	352	<.001	0.10*	0.14*
Total Cognition (TICS-M)	22.93	(4.69)	20.12	(5.22)	19.87	(5.06)	<.001	0.58***	0.64***
Mental Status Summary Score	13.10	(2.17)	11.52	(2.77)	11.33	(2.72)	<.001	0.68***	0.78***
Total Word Recall	9.83	(3.34)	8.61	(3.32)	8.53	(3.24)	<.001	0.37**	0.39**
Immediate Word Recall	5.39	(1.64)	4.99	(1.63)	4.75	(1.62)	<.001	0.24**	0.39**
Delayed Word Recall	4.44	(1.92)	3.61	(1.96)	3.78	(1.85)	<.001	0.43**	0.34**
Ever Diagnosed with ADRD	2%	164	3%	61	3%	54	.03	0.02*	0.06*
Health Behaviors & Healthcare Utilization (Prior 2 Years)	M / %	(SD) / N	M / %	(SD) / N	M / %	(SD) / N	p	d	d
Ever Smoked Cigarettes	56%	4,049	54%	1,247	50%	798	<.001	0.04*	0.08*
# Days/Week Alcohol Consumption	1.40	(2.20)	0.81	(1.57)	0.71	(1.48)	<.001	0.28**	0.33**
Any Physical Activity ≥ Once/Week†	72%	5,169	64%	1,463	68%	1,071	<.001	0.14*	0.06*
Hospital Stay (% Yes)	28%	2,031	28%	628	21%	329	<.001	0.02*	0.12*
# of Hospital Stays	0.54	(1.72)	0.61	(2.53)	0.47	(2.50)	.12	0.03*	0.04*
Nursing Home Stay (% Yes)	4%	310	3%	71	2%	27	<.001	0.06*	0.10*
# of Nursing Home Stays	0.06	(0.61)	0.05	(0.42)	0.02	(0.18)	.008	0.03*	0.08*
Doctor Visit (% Yes)	95%	6,845	89%	2,040	76%	1,200	<.001	0.20**	0.54***
# of Doctor Visits	10.61	(17.15)	9.35	(23.17)	6.92	(16.50)	<.001	0.07*	0.22**
Dental Visit (% Yes)	69%	4,988	51%	1,173	57%	900	<.001	0.32**	0.20**
Utilized Home Health Care (% Yes)	9%	625	10%	239	7%	103	<.001	0.06*	0.06*
Received Outpatient Surgery (% Yes)	24%	1,735	15%	346	13%	211	<.001	0.18*	0.20**
Regular Use of Prescription Drugs (% Yes)	86%	6,248	82%	1,890	76%	1,200	<.001	0.10*	0.22**

Note. Denominators used to calculate proportions excluded those with missingness.

^{‘^’}excludes skin cancer. p-values come from Analysis of Variance (ANOVA) for continuous variables and chi-square tests for categorical variables.

^‘*’= small effect size (d < 0.20),

^‘**’= small-to-medium effect size (0.20 ≤ d < 0.50)

^‘***’= medium effect size (0.50 ≤ d < 0.80),

^‘****’= large effect size (d ≥ 0.80). Effect sizes are given in the absolute value of Cohen’s d (negative values were not reported for ease of interpretation).

^‘†’= Physical activity included vigorous, moderate, or light activities that occurred at least once per week. Silent Generation and Baby Boomer adults were summed by the row and not by the column. Marital status is truncated in this table for ease of interpretation, but all levels were analyzed with the chi-square tests in Table 1. Household income reflects total income for the last calendar year and is the sum of the participant and spouse earnings, pensions and annuities, Social Security Income and Social Security Disability, Social Security retirement, unemployment and workers compensation, other government transfers, household capital income, and other income. Reported or imputed household income of $0 was set to missing (n = 176) in order for natural-log transformation to occur. Total Cognition [Range 0–35] (Mental Status + Total Word Recall), Mental Status [Range 0–15] (serial 7’s (0–5 points)), backward counting from 20 (0–2 points), object (0–2 points), date (0–4 points), and U.S. President/Vice President naming tasks (0–2 points)), Total Word Recall [Range 0–20] (Immediate + Delayed Word Recall), Immediate Word Recall [Range 0–10] (count of correctly identified words from a 10-word list of nouns), and Delayed Word Recall [Range 0–10] (count of correctly identified words from a 10-word list of nouns after 5 minutes during which the participant answered other survey questions).

Abbreviations: ADL = Activities of Daily Living. ADRD = Alzheimer’s Disease and Related Dementias. CES-D = Center for Epidemiological Studies-Depression. IADL = Instrumental Activities of Daily Living. TICS-M = Telephone Interview for Cognitive Status – Modified. “ln” = natural logarithm.

Hispanic/Latinx adults differed from White adults with a medium or large effect size on age, education, parents’ education, NYW, nativity status, income, Medicaid coverage, doctor visits, and cognition. Excluded participants were similar to the included participants (Supplemental Table 1).

Decomposition of Black-White Disparities

If Black and White adults had similar characteristics, 138% of the disparity in depressive symptomatology, 14% of the disparity in cognition, and 85% of the disparity in self-rated health would be attenuated (Table 2). That is, the majority of the disparities in depressive symptomatology and self-rated health were associated with differences in the characteristics of each group, while differences in cognition were largely unexplained and, thus, due to other variables not captured by this study. A share of 138% for depressive symptomatology suggests that if the characteristics between Black and White adults were equivalent, Black adults would have fewer depressive symptoms than White adults.

Table 2.

Decomposition of disparities in depression, cognition, and self-rated health between Black and White adults aged 51+

	Depression (CES-D 8)		Cognition (TICS-M)		Self-Rated Health (1 [EX] - 5 [PR])
	Average Score		Average Score		Average Score
Non-Hispanic/Latinx Black	1.568		20.759		3.029
Non-Hispanic/Latinx White	1.210		23.238		2.774
Unadjusted Disparity	0.358		2.479		0.254
Aggregate Decomposition of Disparity	Estimate	Share (%)	Estimate	Share (%)	Estimate	Share (%)
Characteristics Effect (Explained)	0.495	138%	0.345	14%	0.217	85%
Coefficients Effect (Unexplained)	−0.137	−38%	2.134	86%	0.038	15%
Subaggregate Characteristics Effects	Estimate	Share (%)	Estimate	Share (%)	Estimate	Share (%)
Social Determinants of Health		51%		39%		37%
Education (Years)	0.007	2%	0.275	11%***	0.012	5%***
Mother’s Education (Years)	0.001	0%	0.020	1%	0.011	4%
Father’s Education (Years)	0.008	2%	0.064	3%	0.009	4%
Number of Years Worked	0.043	12%***	0.145	6%***	0.018	7%***
Married (Yes)	0.079	22%***	−0.049	−2%	−0.009	−3%
Veteran Status (Yes)	0.000	0%	0.006	0%	0.001	0%
Southern U.S. Resident (Geography)	−0.003	−1%	0.018	1%	−0.001	0%
Nativity Status (Foreign-Born)	0.002	1%	0.001	0%	0.000	0%
Annual Income (natural-log transformed USD)	0.022	6%	0.224	9%***	0.040	16%***
Employer Sponsored Health Insurance (Yes)	−0.007	−2%	0.000	0%	−0.002	−1%
Medicaid Reimbursed (Yes)	0.014	4%	0.097	4%***	0.003	1%
Medicare Reimbursed (Yes)	0.015	4%	0.155	6%***	0.011	4%
Demographics		44%		−31%		8%
Age	0.152	42%***	−0.740	−30%***	0.022	9%***
Sex (Male)	0.006	2%	−0.037	−1%	−0.002	−1%
Physical Health		28%		1%		7%
Self-Rated Health (1 [Excellent] - 5 [Poor])	0.103	29%***	0.053	2%***	-	-
# Impaired ADLs (0–5)	0.030	8%***	0.010	0%	0.015	6%***
# Impaired IADLs (0–5)	0.016	5%***	0.053	2%***	0.004	2%
# Chronic Conditions (0–8)	−0.011	−3%	−0.003	0%	−0.014	−5%
Body Mass Index	−0.038	−11%***	−0.079	−3%***	0.014	5%***
Sleep Disorder Diagnosis (Yes)	−0.001	0%	0.001	0%	−0.001	0%
Mental Health & Cognition		12%		1%		25%
CES-D 7 Score (No loneliness item)	-	-	0.012	1%	0.041	16%***
Felt Lonely (Yes)	-	-	0.015	1%	0.000	0%
Total Cognition (TICS-M)	0.045	12%***	-	-	0.024	9%***
ADRD Diagnosis (Yes)	0.000	0%	0.001	0%	0.000	0%
Health Behaviors & Healthcare Utilization		3%		4%		7%
Ever Smoked Cigarettes (Yes)	−0.001	0%	−0.004	0%	−0.002	−1%
# Days/Week Alcohol Consumption	−0.005	−1%	0.024	1%	0.013	5%***
Physical Activity ≥ Once/Week	0.011	3%***	0.039	2%***	0.014	5%***
Doctor Visit (Prior 2 Years)	0.007	2%	0.049	2%***	−0.002	−1%
Hospitalization (Prior 2 Years)	0.000	0%	−0.006	0%	−0.005	−2%

Note.

^‘***’indicates that the 95% confidence interval does not contain 0%. All values rounded to 0.000 have no directionality labeled. Proportions are calculated with unrounded estimates (10 digits). The first 3 rows offer the unadjusted disparity for the corresponding racial/ethnic group compared to White adults. The aggregate decomposition of disparity gives the characteristics and coefficients effects after adjusting for all variables seen in each table. Subaggregate characteristics effects report the amount of the disparity that each variable explains (estimate) as well as the value represented as a percentage (share). Physical activity included vigorous, moderate, or light activities that occurred at least once per week. Married included married with a present spouse, married with an absent spouse, and partnered. Household income reflects total income for the last calendar year and is the sum of the participant and spouse earnings, pensions and annuities, Social Security Income and Social Security Disability, Social Security retirement, unemployment and workers compensation, other government transfers, household capital income, and other income.

Abbreviations: ADL = Activities of Daily Living. ADRD = Alzheimer’s Disease and Related Dementias. CES-D = Center for Epidemiological Studies-Depression. IADL = Instrumental Activities of Daily Living. TICS-M = Telephone Interview for Cognitive Status – Modified. USD = U.S. Dollars.

Effects of Social Determinants of Health

Selected SDoH accounted for the largest share of the disparity in depressive symptomatology (51%), cognition (39%), and self-rated health (37%) among Black adults (Figure 1). Marital status (Share: 22%, 95% CI: [16%,28%]) and NYW (12%,[6%,18%]) explained a significant amount of the disparity in depressive symptomatology. Education (11%,[8%,14%]; 5%,[2%,8%]), income (9%,[7%,11%]; 16%,[11%,21%]), and NYW (6%,[4%,8%]; 7%,[3%,12%]) explained a significant amount of the disparities in cognition and self-rated health, respectively. Additionally, Medicare (6%,[4%,9%]) and Medicaid (4%,[2%,6%]) coverage were associated with the disparity in cognition.

Figure 1.

Percentage of Black-White unadjusted health disparities that would reduce if participant characteristics were equivalent across groups by domain

Decomposition of Hispanic/Latinx-White Disparities

If Hispanic/Latinx and White adults had similar characteristics, 105% of the disparity in depressive symptomatology, 64% of the disparity in cognition, and 92% of the disparity in self-rated health would be attenuated (Table 3). That is, the majority of the disparities in all outcomes were associated with differences in the characteristics of each group. A share of 105% for depressive symptomatology suggests that if the characteristics between Hispanic/Latinx and White adults were equivalent, Hispanic/Latinx adults would have fewer depressive symptoms than White adults.

Table 3.

Decomposition of disparities in depression, cognition, and self-rated health between Hispanic/Latinx and White adults aged 51+

	Depression (CES-D 8)		Cognition (TICS-M)		Self-Rated Health (1 [EX] - 5 [PR])
	Average Score		Average Score		Average Score
Hispanic/Latinx	1.704		20.497		3.174
Non-Hispanic/Latinx White	1.210		23.238		2.774
Unadjusted Disparity	0.494		2.742		0.400
Aggregate Decomposition of Disparity	Estimate	Share (%)	Estimate	Share (%)	Estimate	Share (%)
Characteristics Effect (Explained)	0.519	105%	1.751	64%	0.366	92%
Coefficients Effect (Unexplained)	−0.026	−5%	0.990	36%	0.034	8%
Subaggregate Characteristics Effects	Estimate	Share (%)	Estimate	Share (%)	Estimate	Share (%)
Social Determinants of Health		28%		76%		75%
Education (Years)	0.012	2%	1.110	40%***	0.063	16%***
Mother’s Education (Years)	−0.029	−6%	0.032	1%	0.056	14%***
Father’s Education (Years)	0.024	5%	0.142	5%	0.039	10%***
Number of Years Worked	0.058	12%***	0.272	10%***	0.036	9%***
Married (Yes)	0.015	3%	−0.004	0%	−0.002	0%
Veteran Status (Yes)	−0.002	0%	−0.004	0%	0.005	1%
Southern U.S. Resident (Geography)	0.000	0%	−0.001	0%	0.000	0%
Nativity Status (Foreign-Born)	−0.020	−4%	−0.067	−2%	0.022	6%
Annual Income (natural-log transformed USD)	0.028	6%	0.304	11%***	0.059	15%***
Employer Sponsored Health Insurance (Yes)	0.002	0%	0.001	0%	0.001	0%
Medicaid Reimbursed (Yes)	0.028	6%	0.110	4%***	−0.003	−1%
Medicare Reimbursed (Yes)	0.022	4%	0.196	7%***	0.022	6%***
Demographics		32%		−26%		5%
Age	0.162	33%***	−0.743	−27%***	0.019	5%
Sex (Male)	−0.006	−1%	0.026	1%	0.002	0%
Physical Health		35%		4%		−7%
Self-Rated Health (1 [Excellent] - 5 [Poor])	0.162	33%***	0.076	3%***	-	-
# Impaired ADLs (0–5)	0.046	9%***	−0.002	0%	0.023	6%***
# Impaired IADLs (0–5)	0.045	9%***	0.110	4%***	0.008	2%
# Chronic Conditions (0–8)	−0.052	−11%***	−0.027	−1%	−0.060	−15%***
Body Mass Index	−0.023	−5%***	−0.041	−1%***	0.007	2%
Sleep Disorder Diagnosis (Yes)	−0.007	−1%	0.004	0%	−0.004	−1%
Mental Health & Cognition		9%		3%		19%
CES-D 7 Score (No loneliness item)	-	-	0.015	1%	0.054	14%***
Felt Lonely (Yes)	-	-	0.029	1%	−0.002	0%
Total Cognition (TICS-M)	0.043	9%***	-	-	0.024	6%***
ADRD Diagnosis (Yes)	0.000	0%	0.041	2%	0.000	0%
Health Behaviors & Healthcare Utilization		2%		6%		−1%
Ever Smoked Cigarettes (Yes)	−0.003	−1%	−0.004	0%	−0.002	−1%
# Days/Week Alcohol Consumption	−0.002	0%	0.035	1%	0.015	4%***
Physical Activity ≥ Once/Week	0.005	1%	0.016	1%	0.006	2%
Doctor Visit (Prior 2 Years)	0.010	2%	0.145	5%***	−0.005	−1%
Hospitalization (Prior 2 Years)	0.000	0%	−0.022	−1%	−0.017	−4%***

Note.

^‘***’indicates that the 95% confidence interval does not contain 0%. All values rounded to 0.000 have no directionality labeled. Proportions are calculated with unrounded estimates (10 digits). The first 3 rows offer the unadjusted disparity for the corresponding racial/ethnic group compared to White adults. The aggregate decomposition of disparity gives the characteristics and coefficients effects after adjusting for all variables seen in each table. Subaggregate characteristics effects report the amount of the disparity that each variable explains (estimate) as well as the value represented as a percentage (share). Physical activity included vigorous, moderate, or light activities that occurred at least once per week. Married included married with a present spouse, married with an absent spouse, and partnered. Household income reflects total income for the last calendar year and is the sum of the participant and spouse earnings, pensions and annuities, Social Security Income and Social Security Disability, Social Security retirement, unemployment and workers compensation, other government transfers, household capital income, and other income.

Abbreviations: ADL = Activities of Daily Living. ADRD = Alzheimer’s Disease and Related Dementias. CES-D = Center for Epidemiological Studies-Depression. IADL = Instrumental Activities of Daily Living. TICS-M = Telephone Interview for Cognitive Status – Modified. USD = U.S. Dollars.

Effects of Social Determinants of Health

Selected SDoH accounted for the largest share of the disparity in cognition (76%) and self-rated health (75%), but not for depressive symptomatology (28%) among Hispanic/Latinx adults (Figure 2). Interestingly, NYW (12%,[5%,19%]) was the only SDoH associated with depressive symptomatology among Hispanic/Latinx adults. As with Black adults, differences in education (40%,[34%,47%]), income (11%,[8%,14%]), NYW (10%,[6%,13%]), Medicare coverage (7%,[4%,11%]), and Medicaid coverage (4%,[1%,7%]) explained the largest proportion of the disparity in cognition. Additionally, Hispanic/Latinx adults’ poorer self-rated health was associated with level of education (16%,[9%,23%]), NYW (9%,[5%,13%]), and income (15%,[11%,19%]). Unlike Black adults, Hispanic/Latinx adults’ poorer self-rated health was also related to their mother’s education (14%,[3%,25%]), father’s education (10%,[2%,18%]), and whether they had Medicare coverage (6%,[2%,9%]).

Figure 2.

Percentage of Hispanic/Latinx-White unadjusted health disparities that would reduce if participant characteristics were equivalent across groups by domain

Sensitivity Analyses

Sensitivity analyses were conducted to determine whether the Hispanic/Latinx-White comparison differed by nativity status. No appreciable differences were found. Three-fold decomposition was conducted to consider the interaction between the characteristics and coefficients effects. All findings were consistent and/or more pronounced.

Discussion

Our hypotheses were fully supported for the Black-White disparities in depressive symptomatology, cognition, and self-rated health, with SDoH explaining the largest share (51%, 39%, and 37%, respectively). For the Hispanic/Latinx-White comparison, the findings were less straightforward; SDoH did not account for the largest share in depressive symptomatology (age and physical health primarily accounted for the disparity), but they were associated with the largest proportion of the disparities in cognition (76%) and self-rated health (75%). These effects were larger than what was reported for the Black-White comparison (39% and 37%), suggesting that disparities in cognition and self-rated health are more likely to be associated with differences in selected SDoH for Hispanic/Latinx adults than for Black adults. Because many SDoH are difficult to change in older age, primordial prevention strategies are needed to target health and healthcare access in early childhood and carried throughout adulthood (e.g., policies aimed at educational quality, income inequality, insurance parity, and workers at-risk of leaving the workforce).

In addition to the anticipated and well-documented effects of education and income inequity(20–22), NYW was associated with disparities in depressive symptomatology, cognition, and self-rated health among Black and Hispanic/Latinx adults. NYW may index several factors such as age, ability to find and maintain employment, and disability status, all of which may correlate with PHWB. Although Black and Hispanic/Latinx adults were 6 years younger than the White adults on average, they worked for 8 and 13 fewer years than White adults, respectively. This suggests that inequities in NYW were not explained by differences in age alone. It may be that physical functioning, ability to find and maintain adequate employment to meet financial needs(30), and familial caregiving responsibilities(31) contributed to the lower NYW among marginalized adults.

Given that most Black and Hispanic/Latinx older adults should be covered by Medicare in the U.S., it is disconcerting that lower healthcare coverage and utilization was found among Black and Hispanic/Latinx adults. Though overall coverage and utilization rates improved among marginalized adults as a result of the Affordable Care Act(3), Black adults saw the smallest benefit(19). Several explanations for poorer healthcare coverage and utilization may exist, including de facto segregation of healthcare facilities and historical traumas. While sanctioned forms of hospital segregation were eliminated during the 1960s, structural racism remains, partly due to housing segregation and other socioeconomic pressures(32). Beyond place impacting access, healthcare coverage and utilization are also affected by distrust of medical systems by communities of color due to historical traumas (e.g., Tuskegee syphilis experiment among Black men, forced sterilization of Latinas), age (i.e., beneficiary status), and willingness of employers to sponsor health insurance for blue-collar or unskilled workers – of which Black and Hispanic/Latinx adults disproportionately represent in the U.S. workforce(30). Furthermore, immigrants must be a lawful permanent resident in the U.S. to receive Medicare, suggesting that nativity status may have indirectly impacted these constructs through insurance parity.

One area of health that deserves greater attention is cognition given the end of legal racial segregation and the perpetuation of structural racism(5), despite recent societal trends of improved access to education and healthcare. Our study found that the Black-White disparity in cognition was more difficult to explain (86% unexplained) than the Hispanic/Latinx-White disparity in cognition (36% unexplained). Participants’, mothers’, and fathers’ education in years were associated with the disparity to a greater extent among Hispanic/Latinx adults than Black adults. This phenomenon suggests that 1 year of education is not equivalent among Black, Hispanic/Latinx, and White adults, and that improving educational attainment (years alone) may not ameliorate cognitive disparities among Black adults in later life. The quality of education that Black older adults received in the Jim Crow Era may have impacted the comparability of years of education. Additionally, other factors such as reading level(16), ageism, racism, xenophobia, and test bias are relevant factors when measuring cognition among marginalized populations(33, 34), but were not accounted for in our study. Beyond education and discrimination, income and insurance parity correlated with cognition among Black and Hispanic/Latinx adults, meaning that healthcare access and utilization may influence disparities. Language fluency and acculturation may also affect cognitive performance, particularly on measures that utilize word recall and recognition if participants were assessed in a non-native language. Moreover, telephone and internet assessments may not generalize as well to older adults from lower socioeconomic backgrounds. Understanding how SDoH map onto cognitive trajectories will be increasingly important, given the large and growing proportion of marginalized adults diagnosed with Alzheimer’s disease and related dementias(35).

Depressive symptomatology and self-rated health are constructs intimately related to PHWB. The Black-White mental health paradox suggests that Black adults have lower or similar rates of mental health disorders to White adults despite poorer physical health and being subjected to greater adversities(36). In our study and others that use the HRS(37), Black and Hispanic/Latinx adults endorsed greater depressive symptomatology and rated their health as poorer than White adults. Although the CES-D 8 is a commonly-used measure to identify clinically-significant depression, symptomatology cannot be directly compared to psychiatric disorders. For example, Black adults may endorse greater symptomatology than White adults, but may not meet the clinical threshold for major depressive disorder, as criteria may be culturally-, racially-, or ethnically-specific (Supplemental Discussion & References). Interestingly, marital status was associated with the disparity in depressive symptomatology for Black adults, whereas age and physical health was associated with the disparity in Hispanic/Latinx adults. Being married is protective of depression(38), and literature suggests that socioeconomic status, physical health, and marital status may attenuate depression among Black and White adults(39). Though our study does show a relationship between physical health and depressive symptomatology, only a modest effect of income was reported. Rather, our study suggests that NYW more strongly associates with the Black-White and Hispanic/Latinx-White disparity in depressive symptomatology than income. This provides further evidence that NYW remains a unique SDoH that operates independently from age and socioeconomic status, such that finding stable employment throughout one’s life may have a similar effect on PHWB whether at $20,000 or $200,000 per annum. Further, others propose that chronic stress exposure and stress appraisal may explain the Black-White disparity in depressive symptomatology in the HRS(37). These constructs are likely along the causal pathway between SDoH and depressive symptomatology and should be investigated further.

Disparities in self-rated health were associated with differences in the characteristics of Black (85% explained) and Hispanic/Latinx (92% explained) adults. Inequities in selected SDoH accounted for a greater proportion of the disparities in self-rated health than age, sex, measures of health, health behaviors, and healthcare utilization. SDoH such as low educational attainment and income may impact self-rated health through poor neighborhood safety and low physical activity(40). Our study provides nuance to these findings, as self-rated health was not exclusively related to education and income. It was also associated with NYW (Black and Hispanic/Latinx) and parents’ educational levels (Hispanic/Latinx). Moreover, in-line with cumulative disadvantage theory(4), we found that inequities in cognition and depressive symptomatology were strongly associated with the disparity in self-rated health, and that inequities in self-rated health also associated with the disparities in depressive symptomatology among Black and Hispanic/Latinx adults.

This study has several limitations. Comparing Black and Hispanic/Latinx adults to White adults is arbitrary. Clinicians should refrain from viewing White adults as the gold standard of health and should work alongside their patients to develop culturally sensitive and appropriate treatment goals. Blinder-Oaxaca decomposition and the theory that developed it assumes that one group is marginalized, and that this marginalization can be explained through observed and unobserved effects. When using Blinder-Oaxaca decomposition to study disparities between two marginalized groups, it loses much of the meaning. Though we could not use Blinder-Oaxaca decomposition to adequately measure disparities between marginalized communities (i.e., Black vs. Hispanic/Latinx), we believe that this topic deserves further inquiry.

Our findings should be understood in the context of some level of selection bias and over-representation of specific ethnic subgroups. The majority of the Hispanic/Latinx participants were oversampled from the Western portion of the U.S. and are disproportionately of Mexican ancestry. The Black participants were oversampled from the Southern portion of the U.S., where – historically – education, income, and healthcare access is particularly inequitable.

This study is cross-sectional and thus cannot be used to assert causality. Additionally, our definition of PHWB was limited to a single scale of depressive symptomatology, a single item of self-rated health with five possible responses, and a scale to measure cognitive functioning. Finally, using splines in future analyses may better control for non-linear effects of age. See Supplemental Discussion & References for additional material.

In conclusion, we found strong evidence that selected SDoH accounted for a larger proportion of the Black-White disparities in depressive symptomatology, cognition, and self-rated health than each of the other four domain shares (i.e., demographics, physical health, mental health and cognition, and health behaviors and healthcare utilization). Conversely, selected SDoH were associated with a larger proportion of the Hispanic/Latinx-White disparity in cognition and self-rated health, but other factors such as age and physical health were related to the disparity in depressive symptomatology.