Race, community disadvantage, and cognitive decline: Findings from KHANDLE and STAR

Abstract

INTRODUCTION

Community disadvantage is associated with late‐life cognition. Few studies examine its contribution to racial disparities in cognition/cognitive change.

METHODS

Inverse probability weighted models estimated expected mean differences in cognition/cognitive change attributed to residing in less advantaged communities, defined as cohort top quintile of Area Deprivation Indices (ADI): childhood 66–100; adulthood ADI 5‐99). Interactions by race tested.

RESULTS

More Black participants resided in less advantaged communities. Semantic memory would be lower if all participants had resided in less advantaged childhood (b = ‐0.16, 95% confidence interval [CI] = ‐0.30, ‐0.03) or adulthood (b = ‐0.14, 95% CI = ‐0.22, ‐0.04) communities. Race interactions indicated that, among Black participants, less advantaged childhood communities were associated with higher verbal episodic memory (interaction p‐value = 0.007) and less advantaged adulthood communities were associated with lower semantic memory (interaction p‐value = 0.002).

DISCUSSION

Examining racial differences in levels of community advantage and late‐life cognitive decline is a critical step toward unpacking community effects on cognitive disparities.

1. BACKGROUND

Cognitive disparities are produced over a lifetime of accumulated advantage and disadvantage. ¹ , ² Prior research has demonstrated that individual socioeconomic status (SES) in childhood and midlife—especially educational attainment—is an important modifiable risk factor for racial disparities in late‐life cognition and dementia risk. ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ More recently, community SES has been linked with late‐life cognitive and brain health outcomes. ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ Examining race differences in lifecourse community SES and cognitive aging or dementia risk could help target interventions that reduce disparities. ²⁶ , ²⁷

Structural racism underpins racialized health inequities through multiple mechanisms, including residential segregation. ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ Residential segregation has systematically deprived minoritized communities of individual and community social and material resources that have been directly associated with lowering dementia risk, including clean air, access to healthy food, and access to quality health care. ³⁶ , ³⁷ , ³⁸ Lifecourse community SES may therefore be an important explanatory factor for racial disparities in late‐life cognition. A handful of studies have considered lifecourse community exposures for later‐life outcomes using additive approaches, which precludes testing for sensitive periods. ¹⁹ , ³² , ³³ , ³⁹ Disentangling the effects of individual SES from community SES at different stages of the lifecourse may point to the causal pathways that shape susceptibility/resistance to the embodiment of contextual exposures (i.e., the social exposome) that produces disparities. ⁴⁰ , ⁴¹ , ⁴² For example, living in a disadvantaged community in childhood may be more strongly associated with semantic memory, as this domain is strongly influenced by early life social environments. ⁴³ By contrast, community disadvantage at any life stage may operate as a chronic stressor, impairing cognition across multiple domains. ⁴⁴ , ⁴⁵

This study adds to the literature by examining racial differences in the relationship between childhood and adulthood community SES and late‐life cognition and cognitive decline. We use causal inference methods to estimate the effects of community SES while accounting for time‐varying individual SES. Using a lifecourse perspective, we hypothesize stronger effects of low community SES in childhood (relative to adulthood) on late‐life cognitive outcomes, indicative of an early life sensitive period for exposure. ⁴⁶ We further hypothesize that, because of structural racism, ³⁶ Black participants will be overrepresented among those with a history of low community SES, and that both prevalence and effect (i.e., a significant interaction) of community SES will help to explain Black/White disparities in late‐life cognition. ⁴ , ⁴⁷

RESEARCH IN CONTEXT

1. Systematic review: We reviewed studies in PubMed. Many have found associations of community‐level socioeconomic status (e.g., Area Deprivation Index) with late‐life cognition, cognitive decline, and brain health. Most studies to date have been conducted in predominately White populations. Few have investigated differences between Black and White populations.

2. Interpretation: Associations between residing in a less advantaged community in childhood and adulthood with late‐life cognition vary by race. Our findings suggest lifelong harmful effects structural racism, even among a relatively advantaged cohort of Black and White participants.

3. Future directions: This study highlights the importance of considering how relationships between community‐level exposures and individual health relationships may vary by population. Future research should examine race‐specific pathways/mediators between community disadvantage and cognitive aging outcomes to identify group‐specific public health intervention targets.

2. METHODS

The Kaiser Healthy Aging and Diverse Life Experiences (KHANDLE) study and the Study of Healthy Aging in African Americans (STAR) are harmonized longitudinal cohorts of long‐term Kaiser Permanente Northern California members living in the San Francisco Bay area and Sacramento Valley. Participants were eligible for KHANDLE if they were age 65 or older on January 1, 2017, spoke English or Spanish, and participated in voluntary multiphasic health check‐ups (MHC) from 1964 to 1985. Stratified random sampling by race/ethnicity and educational attainment was used to recruit approximately equal proportions of Asian, Black, Latino, and White participants. Participants eligible for STAR self‐identified as Black or African American, were age 50 or older on January 1, 2018, participated in MHC, and spoke English. Stratified random sampling by age and educational attainment was used to recruit approximately equal proportions of Black participants ages 50–64 and 65 and older. Exclusion criteria for both studies included electronic medical record diagnosis of dementia or other neurodegenerative disease, or the presence of severe health conditions that would impede participation in study interviews. Both studies were institutional review board (IRB) approved, and all participants provided written informed consent.

2.1. Measures

Cognition. Verbal episodic memory, executive function, and semantic memory were derived from the Spanish and English Neuropsychological Assessment Scales (SENAS). ⁴⁸ , ⁴⁹ Administration procedures, development, and psychometric characteristics have been described in detail elsewhere. ⁴⁸ , ⁴⁹ SENAS was administered in either English or Spanish, based on the participant's preference. Until March 2020, SENAS was administered in person; after March 2020, SENAS was administered by telephone under a separate protocol during the coronavirus disease 2019 (COVID‐19) pandemic. The phone SENAS battery did not assess semantic memory, which requires visual stimuli. Each cognitive domain was z‐standardized using the Wave 1 mean and standard deviation from the combined KHANDLE and STAR cohorts.

Community Disadvantage. At the baseline interview, KHANDLE and STAR participants provided residential addresses from birth and ages 5, 10, 18, 30, 40, and 65. These addresses were geocoded to the smallest geographic unit (i.e., county, census block group) published by the U.S. Census for the period when the participant lived at that address. In general, available data allowed us to map early life residences to the county, and adulthood residences to the block group. The respective geographic identifiers were assigned an Area Deprivation Index as available through the Neighborhood Atlas and from the University of Wisconsin Center for Health Disparities Research. ⁵⁰ ADI scores are derived from an index of 17 indicators published by the U.S. decennial census and American Community Survey and converted to a national percentile rank. For each geo‐analytic time frame (e.g., census decade), an ADI of 1 represents the 1% of most advantaged geographic units in the United States, and an ADI of 100 represents the most disadvantaged 1% of geographic units in the United States. The details of ADI development have been described elsewhere. ⁵¹ , ⁵² Prior work suggests the relationship between ADI and health outcomes tends to be non‐linear; many studies use the national top quintile (ADI > 80) to define neighborhood disadvantage. ⁵³ , ⁵⁴ In this cohort, we found evidence of non‐linearity at the tails of the distribution (Figures S1‐S3). We examined ADI based on two sets of cutoffs. In primary analyses, childhood community disadvantage was defined as ADI > 80. We could not examine effects of adulthood ADI using this cutoff, as only five cohort participants resided in a community with ADI > 80 in adulthood. In secondary analyses, we defined the least advantaged communities as the top quintile of the cohort's ADI distribution: ADI > 65 in childhood; ADI > 4 in adulthood. Secondary models provide an opportunity to explore the effects of relative community‐level advantage or disadvantage in our cohort of long‐term northern California residents—an area with vast disparities across neighborhoods, but of relative affluence when compared nationally (Figure 1). For both sets of analyses, birth address was used for childhood ADI because it encapsulates the earliest lifecourse exposure. When birth residence was missing, we used ADI from the participant's residence at age 5 (n = 63) or age 10 (n = 29). Adulthood ADI was based on age 30 residence. If age 30 residence was missing, we used their residence at age 40 (n = 147), age 65 (n = 20), or age 18 (n = 14).

FIGURE 1.

Distribution of Area Deprivation Indices in childhood and adulthood, stratified by race.

Childhood and Adulthood Socioeconomic Status. Childhood SES was based on parental educational attainment, perceived childhood financial status, food security, and parental homeownership. Maternal and paternal years of education were included as separate continuous covariates. Childhood financial status was based on self‐reported family finances from birth to age 16 as “pretty well off financially,” “about average,” or “poor.” Participants who reported ever (vs. never) having to “skip a meal or go hungry because there was not enough money to buy food” were classified as food insecure. Parental homeownership was retrospectively reported and dichotomized as homeowner versus renter/other. Individual SES in adulthood was determined by the participant's educational attainment. Participants reported years of education completed if less than a high school diploma, or the highest credential/degree completed. We combined responses into a single continuous variable, reflecting years of completed education (range 0–20; 16 = Bachelor's degree).

Covariates. Demographic covariates were age at wave 1, gender (male/female), and race (non‐Hispanic Black/non‐Hispanic White).

2.2. Analytic sample

Because a large portion of Asian and Latino KHANDLE participants are immigrants whose childhood addresses could not be mapped to ADIs, we restricted analyses to Black and White participants who were born or residing within the United States by age 10. Our final analytic sample (n = 1480) was comprised of 1079 Black and 401 White participants. Participants were excluded for missing residential history at all ages in childhood or adulthood (n = 86; 5%), childhood SES variables (n = 35, 2%), educational attainment (n = 1, < 0.015%), late‐life income (n = 92, 5%), or wave 1 cognitive domains (n = 11, 0.01%).

2.3. Statistical analysis

Race‐stratified participant characteristics were examined overall and among sub‐samples from the cohort‐defined least advantaged communities (cohort top quintile) in childhood or adulthood. We implemented mixed models to estimate minimally adjusted associations between community advantage/disadvantage and domain‐specific cognition and cognitive change that did not adjust for individual SES. Next, we estimated marginal structural models using stabilized inverse probability weights (IPWs) to account for time‐varying individual SES. ⁵⁵ , ⁵⁶ This approach allows for a more accurate estimation of the true effect of community‐level SES in the presence of time‐varying individual SES, which falls on the causal pathway as both a mediator and a confounder. This may be of critical import in this cohort, which has experienced upward social mobility, as reflected by the relatively high affluence of the communities they live compared to the communities they grew up. IPWs were calculated using logistic regression to estimate the inverse probability of residing in a disadvantaged community in childhood given the participant's childhood SES, and of residing in a disadvantaged community in adulthood given the participant's SES in childhood and adulthood, and childhood residence in a disadvantaged community. In this approach, IPWs up‐weight individuals who, given their history of covariates, were unlikely to reside in a disadvantaged community. This created a pseudo‐population in which community disadvantage in childhood and adulthood was no longer associated with individual SES, allowing for more accurate estimation of the effect of community level disadvantage. IPWs estimated marginal structural models adjusted for age at wave 1 interview, gender, and race. The estimated coefficients of marginal structural models are the average treatment effects, E [Ya1 = 1]‐E [Ya1 = 0], interpreted as the expected mean difference in cognitive function or rate of cognitive change if everyone in the study had lived in a disadvantaged community compared with no one having lived in a disadvantaged community. We included race*community disadvantage interaction terms to test if the expected mean difference of baseline cognition and cognitive change attributed to community disadvantage varies by race. Sensitivity analyses assessed variation in effects due to different ages eligibility criteria in KHANDLE (age 65+) and STAR (age 50+). We refit weighted models using the cohort‐defined cutoffs for least advantaged communities, restricted to participants ages ≥65 at wave 1. All models were adjusted for baseline age, gender, and race. Cohort‐derived model constraints adjusted for practice effects in longitudinal analyses. ⁵⁷ All analyses were completed in Stata 17 (StataCorp, College Station, TX).

3. RESULTS

When using nationally‐ranked ADI percentiles, participants in this study lived in moderately advantaged communities in childhood and highly advantaged communities in adulthood (Figure 1). Black participants resided in neighborhoods with a mean ADI of 29 (standard deviation [SD] = 33) in childhood and 6 (SD = 10) in adulthood (Table 1). White participants resided in neighborhoods with a mean ADI of 19 (SD = 29) in childhood and 4 (SD = 4) in adulthood. In this study, 152 Black and 37 White participants resided in a disadvantaged community in childhood, defined as ADI > 80. Participants from the least advantaged quintile of childhood communities (ADI range 66‐100) were older (mean = 79.7, SD = 7.9 vs. mean = 72.1 SD = 8.6), more likely to be Black (84% vs. 73%), had lower childhood SES across all measures (e.g., 51% reported family was poor vs. 34% overall), and were more likely to earn less than $55,000 (46% vs. 34% overall) at their wave 1 interview. We observed no substantial demographic differences among those residing in the least advantaged adulthood communities, as this group encompassed all but the most advantaged 5% of communities nationally (ADI range = 5–99; Figure 1).

TABLE 1.

Characteristics of KHANDLE and STAR participants stratified by race and least advantaged community residence, defined using the cohort's top quintile of Area Deprivation Index scores.

	Overall (n = 1)			Childhood ADI 66‐100 (cohort least advantaged quintile)		Adulthood ADI 5‐99 (cohort least advantaged quintile)
	Black (n = 1079)	White (n = 401)	Black (n = 249; 23%)	White (n = 47; 12%)	Black (n = 197; 18%)	White (n = 68; 17%)
Age (years)	70.7 (8.6)	76.0 (7.2)	78.3 (7.6)	86.7 (5.3)	69.8 (10.6)	77.8 (6.2)
Female, n (%)	746 (69)	225 (56)	157 (63)	26 (55)	122 (62)	40 (59)
ADI in childhood, mean (SD)	28.6 (33.4), range 1–100	18.9 (28.6), range 1–97	82.5 (9.6), range 66–100	88.4 (9.2), range 68–97	31.4 (34.6), range 1–95	21.9 (33.4), range 1–97
ADI in adulthood, mean (SD)	5.6 (10.0), range 1–99	4.0 (4.3), range 1–58	5.5 (9.1), range 1–92	4.0 (3.9), range 1–24	17.5 (19.2), range 5–99	8.8 (8.5), range 5–58
Childhood geographic region, n (%)
West	546 (51)	233 (58)	25 (10)	27 (58)	99 (50)	34 (50)
South	153 (14)	69 (17)	58 (23)	9 (19)	29 (15)	13 (19)
Midwest	200 (18)	71 (17)	96 (39)	9 (19)	35 (18)	15 (22)
Northeast	180 (17)	28 (7)	70 (28)	2 (4)	34 (17)	6 (9)
Childhood SES
Maternal education, mean (SD)	9.7 (5.0)	11.2 (4.7)	7.1 (5.1)	8.2 (5.9)	9.5 (5.4)	10.9 (4.5)
Missing maternal education, n (%)	139 (13)	36 (9)	56 (23)	12 (26)	32 (16)	6 (9)
Paternal education, mean (SD)	7.6 (5.8)	10.8 (5.7)	4.9 (4.9)	7.9 (6.8)	7.2 (6.0)	9.5 (6.1)
Missing paternal education, n (%)	276 (26)	58 (15)	81 (33)	16 (34)	58 (29)	13 (19)
Family finances
Well off, n (%)	87 (8)	40 (10)	11 (4)	2 (4)	20 (10)	5 (7)
Average, n (%)	589 (55)	261 (65)	104 (42)	28 (60)	98 (50)	45 (66)
Poor/It varied, n (%)	403 (37)	100 (25)	134 (54)	17 (36)	79 (40)	18 (27)
Childhood family homeowners, n (%)	640 (59)	279 (70)	115 (46)	27 (57)	113 (57)	49 (72)
Food secure, n (%)	968 (90)	372 (93)	216 (87)	43 (92)	177 (90)	62 (91)
Adulthood SES
Education, mean (SD)	14.5 (2.4)	15.5 (2.6)	14.0 (2.6)	15.2 (3.1)	14.6 (2.5)	15.8 (2.5)
Income
<54,999, n (%)	393 (36)	116 (29)	116 (47)	21 (45)	76 (39)	23 (34)
55,000–99,999, n (%)	409 (38)	152 (38)	88 (35)	15 (32)	68 (34)	25 (37)
≥1,00,000, n (%)	277 (26)	133 (33)	45 (18)	11 (23)	53 (27)	20 (29)

Abbreviations: ADI, Area Deprivation Index; KHANDLE, Kaiser Healthy Aging and Diverse Life Experiences study; SES, socioeconomic status; SD, standard deviation; STAR, Study of Health Aging in African Americans.

3.1. Childhood community disadvantage (ADI > 80)

Models without individual SES. In minimally adjusted mixed effects models, average baseline verbal episodic memory scores (b = 0.08, 95% confidence interval [CI] = −0.05, 0.21) were higher among those who lived in a disadvantaged community in childhood and executive function scores (b = −0.10, 95% CI = −0.23, 0.03) were lower among those who lived in a disadvantaged childhood community, though confidence intervals for both estimates included the null (Table 2). Average baseline semantic memory scores were significantly lower among those who lived in a disadvantaged community in childhood (b = −0.18, 95% CI = −0.27, −0.09). Childhood community disadvantage was not associated with rate of change for verbal episodic memory (b = −0.01, 95% CI = −0.06, 0.05) or executive function (b = −0.03, 95% CI = −0.07, 0.01).

TABLE 2.

Mixed effects and marginal structural models estimate the effects of childhood community disadvantage, defined as having an Area Deprivation Index score > 80 (national top quintile)

		Verbal episodic memory	Executive function	Semantic memory
A	Childhood neighborhood disadvantaged	0.08 (−0.05, 0.21)	−0.10 (−0.23, 0.03)	−0.18 (−0.27, −0.09)
	Years	−0.04 (−0.06, −0.02)	−0.02 (−0.03, −0.003)	NA
	Years*childhood neighborhood disadvantage	−0.01 (−0.06, 0.05)	−0.03 (−0.07, 0.01)	NA
B	Childhood neighborhood disadvantaged	0.16 (−0.02, 0.33)	−0.05 (−0.25, 0.16)	−0.04 (−0.15, 0.07)
	Years	−0.04 (−0.06, −0.02)	−0.02 (−0.03, −0.001)	NA
	Years*childhood Neighborhood disadvantage	−0.04 (−0.13, 0.06)	−0.04 (−0.09, 0.03)	NA

Note: A: Mixed effects models, adjusted for age at baseline cognitive assessment, gender, and race/ethnicity.

B: Marginal structural models are stabilized inverse probability weighted on time‐varying individual socioeconomic status and adjusted for age at baseline cognitive assessment, gender, and race/ethnicity.

Models accounting for individual SES. In stabilized inverse probability weighted models, we estimated higher mean baseline verbal episodic memory (b = 0.16 95% CI = −0.02, 0.33) and lower mean baseline executive function (b = −0.05, 95% CI = −0.25, 0.16) and semantic memory (b = −0.04, −0.15, 0.07) had all participants resided in a disadvantage childhood community (ADI > 80) versus no one residing in a disadvantaged childhood community (Table 2). We observed no difference in mean annual rate of verbal episodic memory change (b = −0.04, 95% CI = −0.13, 0.06) or executive function change (b = −0.04, 95% CI = −0.09, 0.03) associated with disadvantaged childhood communities.

3.2. Least advantaged childhood (ADI > 65) and adulthood communities (ADI > 4)

Models without childhood SES. When we examined relative community advantage/disadvantage based on cohort ADI distributions, we found trends toward higher mean baseline verbal episodic memory among those who resided in the least advantaged cohort quintile of communities in childhood (b = 0.04; 95% CI = −0.08, 0.16) and lower mean baseline verbal episodic memory among those who resided in the least advantaged cohort quintile of communities in adulthood (b = −0.04; 95% CI = −0.15, 0.07; Table 3). Associations of residing in less advantaged communities in childhood or adulthood with mean change in verbal episodic memory were close to null (childhood: b = −0.003, 95% CI = −0.05, 0.04; adulthood: b = 0.02, 95% CI = −0.03, 0.06). Mean executive function baseline scores were lower among those who resided in the least advantaged childhood communities (b = −0.15; 95% CI = −0.27, −0.03); this trend was less pronounced among those residing in the least advantaged adulthood communities (b = −0.04; 95% CI = −0.15, 0.07). Residing in less advantaged childhood or adulthood communities was not associated with mean change in executive function (childhood: b = −0.02, 95% CI = −0.05, 0.01; adulthood: b = −0.01; 95% CI = −0.04, 0.02). Baseline mean semantic memory was significantly lower among those from the least advantaged communities in childhood (b = −0.30, 95% CI = −0.40, −0.19) and adulthood (b = −0.13, 95% CI = −0.22, −0.03).

TABLE 3.

Marginal structural models estimating the effects of residing in a less advantaged community, defined as the cohort top quintile of Area Deprivation Index scores in childhood (ADI > 65) and adulthood (ADI > 4)

		Verbal episodic memory	Executive function	Semantic memory
A	Least advantaged childhood community quintile (ADI 66‐100)	0.04 (−0.08, 0.16)	−0.15 (−0.27, −0.03)	−0.30 (−0.40, −0.19)
	Least advantaged adulthood community quintile (ADI 5‐99)	−0.04 (−0.15, 0.07)	−0.03 (−0.13, 0.08)	−0.13 (−0.22, −0.03)
	Years between cognitive assessments	−0.05 (−0.07, −0.03)	−0.01 (−0.03, 0.003)	NA
	Years*least advantaged childhood community quintile	−0.003 (−0.05, 0.04)	−0.02 (−0.05, 0.01)	NA
	Years*least advantaged adulthood community quintile	0.02 (−0.03, 0.06)	−0.01 (−0.04, 0.02)	NA
B	Least advantaged childhood community Quintile (ADI 66‐100)	0.06 (−0.09, 0.21)	−0.05 (−0.25, 0.16)	−0.16 (−0.30, −0.03)
	Least advantaged adulthood community quintile (ADI 5‐99)	−0.08 (−0.20, 0.05)	−0.09 (−0.23, 0.05)	−0.14 (−0.22, −0.04)
	Years between cognitive assessments	−0.06 (−0.09, −0.04)	−0.01 (−0.03, 0.01)	NA
	Years*least advantaged childhood community quintile	0.01 (−0.06, 0.09)	−0.02 (−0.09, 0.05)	NA
	Years*least advantaged adulthood community quintile	0.05 (−0.01, 0.11)	−0.004 (−0.04, 0.03)	NA

Note: A: Mixed effects models, adjusted for age at baseline cognitive assessment, gender, and race/ethnicity.

B: Marginal structural models are stabilized inverse probability weighted on time‐varying individual socioeconomic status and adjusted for age at baseline cognitive assessment, gender, and race/ethnicity.

Models accounting for childhood and adulthood SES. When we accounted for SES in childhood and adulthood, associations of least advantaged communities and mean baseline verbal episodic memory were potentiated compared to models that did not account for individual SES (childhood: b = 0.06, 95% CI = −0.09, 0.21; adulthood: b = −0.08, 95% CI = −0.20, 0.05; Table 3). We also observed a trend of slower mean verbal episodic memory decline had all participants resided in the least advantaged communities in adulthood (b = 0.05, 95% CI = −0.01, 0.11), though not childhood (b = 0.01, 95% CI = −0.06, 0.09). For executive function, accounting for individual SES attenuated the negative effect of residing in a less advantaged childhood community (b = −0.05, 95% CI = −0.25, 0.16) and potentiated the effect of residing in a less advantaged adulthood community (baseline b = −0.09, 95% CI = −0.23, 0.05). We observed no effect of least advantaged childhood or adulthood communities on mean executive function change (childhood: b = −0.02, 95% CI = −0.09, 0.05; adulthood: b = −0.004, 95% CI = −0.04, 0.03). After accounting for individual SES, the relationship between least advantaged childhood community and mean baseline semantic memory was attenuated (b = −0.16, 95% CI = −0.30, −0.03), but associations between least advantaged adulthood communities and semantic memory were unchanged (b = −0.14, 95% CI = −0.22, −0.04). In sensitivity analyses that restricted participants to ages 65 and older, we observed a similar pattern of results across all domains (Table S1).

3.3. Race differences in effect of least advantaged community residence

We observed a significant interaction by race in the effect of residing in a disadvantaged childhood community (ADI > 80) on verbal episodic memory (p = 0.007). Stratified analyses indicated that disadvantaged communities were associated with higher verbal episodic memory among Black participants (marginal mean = 0.17; 95% CI = 0.02, 0.32) and lower verbal episodic memory among White participants (marginal mean = −0.15; 95% CI = = −0.44, 0.13; Figure S4).

The effect of residing in the least advantaged childhood communities on executive function varied by race in minimally adjusted mixed models (p = 0.05), but not marginal structural models (p = 0.12). Race by less advantaged adulthood community interactions were significant for late‐life semantic memory in minimally adjusted (p = 0.004) and marginal structural models (p = 0.002). Stratified analyses indicate that among Black participants, living in the least advantaged quintile of communities in adulthood was associated with approximately a quarter standard deviation lower (marginal mean = −0.24; 95% CI = −0.36, −0.12) baseline semantic memory after accounting for individual SES and childhood community advantage/disadvantage (Figure S5). Conversely, if all participants had resided in the most advantaged 80 percent of adulthood communities, we would expect mean semantic memory to be 1/10th standard deviation higher overall (marginal mean = 0.11; 95% CI = 0.01, 0.20), and about a quarter standard deviation higher among Black participants specifically (marginal mean = 0.24; 95% CI = 0.12, 0.35).

4. DISCUSSION

In these cohorts, a larger proportion of Black than White participants resided in disadvantaged neighborhoods in childhood and adulthood, reflecting historical and contemporary patterns of structural racism. Childhood community disadvantage (ADI > 80) was associated with higher mean baseline verbal episodic memory, and lower mean baseline executive function and semantic memory. Lowering the ADI cutoff to examine the cohort's least advantaged quintile of communities in childhood (ADI > 65) attenuated toward the null the positive association with mean baseline verbal episodic memory and negative association with mean baseline executive function but potentiated the negative association with mean baseline semantic memory. In adulthood, residing in the less advantaged quintile of communities (ADI > 4) was associated with lower baseline semantic memory.

Accounting for individual SES attenuated the effect of childhood community contexts with executive function and potentiated the effect of adulthood community contexts with executive function and verbal episodic memory. Associations of childhood and adulthood community contexts with semantic memory were independent of individual SES. In racially stratified analyses, residing in a less advantaged community in adulthood was associated with lower baseline semantic memory for Black participants, but not White participants (Figure S4).

Our findings corroborate those of Meyer et al., (2018), who found the association of community SES and baseline executive function was attenuated after accounting for individual SES, but remained significant for semantic memory. ²³ This may be because semantic memory, a form of experiential knowledge about the world, is influenced by early life social environments and late‐life cognitive engagement. ⁴³ , ⁵⁸ In addition, inflammatory markers of stress from discrimination have been associated with poorer late‐life semantic memory. ⁴⁵ , ⁵⁹ , ⁶⁰ Overall, our findings demonstrate that adulthood community contexts are associated with worse late‐life cognition, and that both childhood and adulthood community disadvantage may contribute to racial disparities in late‐life semantic memory through both a higher prevalence of exposure and a stronger effect. ⁴⁷ Intervening on community disadvantage across the lifecourse may therefore reduce late‐life cognitive disparities.

Our finding of higher verbal episodic memory among those from disadvantaged childhood communities is best understood through race‐stratified analyses, which reveals this association is only present among Black participants (Figure S5). One explanation is a pattern of selective migration to northern California. Among those in this study who are from a childhood community with ADI > 80, 57% of White participants resided in the West in childhood, compared with only 8% of Black participants. Most Black participants with childhood community disadvantage migrated to California from the Midwest (42%), South (25%), or Northeast (25%). Black Americans who emigrated from the South during the Great Migration were more likely to be better educated and with greater access to skilled employment, which may differentiate this cohort's Black participants from other Black Americans. ⁶¹ In other words, this cohort's Black participants may have higher mean baseline verbal episodic memory scores in spite of residing in disadvantaged childhood communities. ³⁶

We acknowledge several limitations. Our cohort participants tended to live in relatively advantaged communities, when compared nationally, limiting power to detect effects—especially for adulthood community disadvantage. While there are substantial disparities in community disadvantage in the San Francisco Bay and Sacramento Valley, these disparities are minimized with national ADI percentiles. We attempted to limit this bias by examining our models with multiple ADI cutoffs. Nonetheless, we remained underpowered to test race interactions with high precision. Another limitation is in our broad definition of community. Census block groups were not designated for all U.S. areas until 1990; childhood ADIs in this cohort were based on county‐level geographical units.

Among study strengths, this is the first study to use historical ADIs calculated by the University of Wisconsin Center for Health Disparities Research, showing the long‐arm effect of community disadvantage for late‐life cognition—even among a relatively affluent cohort. This among the few studies to examine community disadvantage from both childhood and adulthood, and to apply appropriate methods for disentangling the effects of time‐varying individual socioeconomic status from those of time‐varying community exposure. It is also one of the first studies to examine how racial heterogeneity in community disadvantage is associated with late‐life cognition and cognitive decline. This study contributes to our understanding of how structural racism, which has differentially shaped residential history and community disadvantage, contributes to health disparities and resilience. Additionally, our findings add to the literature on relative advantage/disadvantage at a contextual level. Historical ADIs, which were percentile ranked by census decade, were linked to participant residences based on a set of ages (e.g., birth, 5, 10) that, in KHANDLE and STAR cohorts, comprised ADI rankings from 1910 to 1970. Because ADI is not anchored to an absolute threshold of deprivation and standards of living increased for all Americans across these decades, our findings highlight the importance of considering the impact of relative community disadvantage on lifecourse health and disparities.

Future studies should continue to examine heterogeneity in exposure and effect of community disadvantage and late‐life cognitive aging outcomes in racially/ethnically and geographically diverse populations. Exploring a community‐level gradient in health effects – a concept few studies have applied ⁶² —would further our understanding of the connections between residential discrimination and health. Additionally, future work is planned to examine lifestyle factors (e.g., vascular risk factors, health behaviors) as mediators of the relationships between community contexts and late‐life cognition; this could better clarify the pathways for the social exposome to affect cognitive health. Nonetheless, community context measures, such as the ADI, are policy actionable. Strategically targeting health promoting resources to communities and neighborhoods with greater disadvantage can reduce the contribution of social and structural determinants to disparities in dementia risk. ⁶³

CONFLICT OF INTEREST STATEMENT

Rachel L. Peterson, Rebecca Pejak, Kristen M. George, Paola Gilsanz, Michelle Ko, Oanh L. Meyer, Elizabeth Rose Mayeda, Amy Kind, Rachel A. Whitmer report no conflict of interest. Author disclosures are available in the supporting information.

CONSENT STATEMENT

KHANDLE and STAR studies are approved by the Kaiser Permanente Northern California Division of Research Institutional Review Board. All participants provided written informed consent.

Supporting information

Supporting Information

Supporting Information

Peterson RL, Pejak R, George KM, et al. Race, community disadvantage, and cognitive decline: Findings from KHANDLE and STAR. Alzheimer's Dement. 2024;20:904–913. 10.1002/alz.13511