Racial/Ethnic Disparities in Alzheimer’s Disease Risk: Role of Exposure to Ambient Fine Particles

Abstract

Background

Whether racial/ethnic disparities in Alzheimer’s disease (AD) risk may be explained by ambient fine particles (PM_2.5) has not been studied.

Method

We conducted a prospective, population-based study on a cohort of Black (n = 481) and White (n = 6 004) older women (aged 65–79) without dementia at enrollment (1995–1998). Cox models accounting for competing risk were used to estimate the hazard ratio (HR) for racial/ethnic disparities in AD (1996–2010) defined by Diagnostic and Statistical Manual of Mental Disorders, 4th edition and the association with time-varying annual average PM_2.5 (1999–2010) estimated by spatiotemporal model.

Results

Over an average follow-up of 8.3 (±3.5) years with 158 incident cases (21 in Black women), the racial disparities in AD risk (range of adjusted HR_{Black women} = 1.85–2.41) observed in various models could not be explained by geographic region, age, socioeconomic characteristics, lifestyle factors, cardiovascular risk factors, and hormone therapy assignment. Estimated PM_2.5 exposure was higher in Black (14.38 ± 2.21 µg/m³) than in White (12.55 ± 2.76 µg/m³) women, and further adjustment for the association between PM_2.5 and AD (adjusted HR_PM2.5 = 1.18–1.28) slightly reduced the racial disparities by 2%–6% (HR_{Black women} = 1.81–2.26). The observed association between PM_2.5 and AD risk was ~2 times greater in Black (HR_PM2.5 = 2.10–2.60) than in White (HR_PM2.5 = 1.07–1.15) women (range of interaction ps: <.01–.01). We found similar results after further adjusting for social engagement (social strain, social support, social activity, living alone), stressful life events, Women’s Health Initiative’s clinic sites, and neighborhood socioeconomic characteristics.

Conclusions

PM_2.5 may contribute to racial/ethnic disparities in AD risk and its associated increase in AD risk was stronger among Black women.

Air pollution, Dementia, Epidemiology, Incidence

Significant differences in Alzheimer’s disease (AD) incidence across racial/ethnic groups exist, with Black individuals approximately 2–3 times more likely to develop AD compared to White (1). In the United States, Black women had the highest prevalence (15.1%) of AD and related dementias (ADRDs) among nearly 5 million people (aged 65 years and older) diagnosed in 2014 (2). Several well-established AD risk factors (eg, genetics, sociodemographics, vascular conditions) could not fully explain these racial disparities (1). In the study on “Critical Perspectives on Racial and Ethnic Differences in Health in Late Life,” the National Research Council noted that “little work has been published addressing ethnic differences in the environmental risk factors that…explain ethnic group differences in frequency of AD” (3).

Fine particulate matter (PM with aerodynamic diameter <2.5 μm [PM_2.5]) may be one such environmental risk factor potentially contributing to the racial disparities in AD risk. Neurotoxicological data in animal models have demonstrated associations between inhaled exposures to particles and increased early markers of neurodegenerative disease (accumulation of amyloid-β, phosphorylation of tau), as well as structural and functional changes in the brain (4–12). Neuroimaging studies in humans have provided evidence for associations between PM_2.5 and increased amyloid-β plaques (13) and smaller gray matter volumes in brain areas vulnerable to AD neuropathologies (14,15). Converging with these data, several cohort studies have reported associations between PM_2.5 and increased risk of ADRD (16). Although Black individuals in the United States are more likely to reside near PM-emitting facilities and have significantly greater exposures (17), whether PM_2.5 may contribute to the racial/ethnic disparities in AD risk has been overlooked. Extant epidemiologic evidence suggests minority populations may be more susceptible to the adverse health effects of PM_2.5 (18). One recent study found that the attributable burden of dementia death associated with PM_2.5 was greater among Black individuals (19). However, no studies have explored whether the adverse effects of PM_2.5 on ADRD incidence differentially affect Black populations.

We conducted a longitudinal study to address these knowledge gaps. First, we investigated whether the putative increase in AD risk among Black compared to White women may be explained by between-individual differences in estimated exposure to ambient PM_2.5. Second, we examined whether race/ethnicity imparts differential susceptibility to the putative PM_2.5 neurotoxicity by testing the hypothesis that the association between PM_2.5 and AD risk is stronger in Black than in White women.

Method

Study Design and Population

Participants were drawn from the Women’s Health Initiative (WHI) Memory Study (WHIMS) (20)—an ancillary study to the WHI hormone therapy (WHI-HT) trials designed to test the efficacy of hormone therapy on dementia. Community-dwelling postmenopausal women (N = 7 479) without dementia during 1995–1998 were recruited (65–80 years old at enrollment) from more than 40 study sites across 24 U.S. states and Washington, DC. During baseline screenings, all participants chose the racial or ethnic group they identified with most. As part of the annual contact in 2003–2004, personal information was updated with race/ethnicity questions used in the 2000 U.S. census. For this study, we included women who self-identified as African American or Black (Black) or non-Hispanic White (White). Written informed consent was obtained from all participants as part of WHI-HT and WHIMS studies.

Incident AD Ascertainment

Our primary outcome was clinically defined AD, as determined using WHIMS protocols (20). Centrally trained, masked, and certified technicians administered the Modified Mini-Mental State (3MS) (21) examination to screen for cognitive impairment at baseline and annually. Women scoring below an education-adjusted cut point underwent further extensive neuropsychological testing (including a modified Consortium to Establish a Registry for Alzheimer’s Disease battery (22)) and were evaluated by a board-certified physician-specialist with experience in diagnosing dementia. Suspected dementia cases underwent further clinical workup, including a cranial computerized axial tomography scan and laboratory blood tests to rule out possible reversible causes of cognitive decline and dementia. Starting in 2008, screening of global cognitive impairment was based on the Telephone Interview for Cognitive Status—modified (23), which is highly correlated with 3MS (0.89) and has excellent sensitivity (0.87)/specificity (0.89) to differentiate dementia cases from normal controls. Data were then transmitted to a central adjudication committee for final confirmation of dementia. The experts reviewed all the data and classified the WHIMS participant as having no dementia, mild cognitive impairment, or probable dementia. For those incident cases of probable dementia, their clinical phenotypes were then further classified (eg, vascular, AD, mixed type, unknown, other dementia-related classifications) based on Diagnostic and Statistical Manual of Mental Disorders, 4th edition, which has substantial interrater reliability and good validity for AD diagnosis (compared with postmortem confirmation) (24). Our study focused on cases with primary AD diagnosis defined as cases of incident probable dementia that were further classified as AD only. In order to reduce the bias in ascertaining dementia cases among minority groups, including Black individuals who tended to have lower neuropsychological testing scores, normative data by race wherever available were provided to the adjudicators who were blinded to residential locations and estimated air pollution exposure levels.

Air Pollution Exposure Assessment

Detailed procedures for air pollution estimation have been reported elsewhere (25) and summarized in Supplementary Methods. Briefly, residential addresses prospectively collected at each clinic visit and updated at least biannually were geocoded to obtain the latitude and longitude of each location using standardized protocols (26). We generated individual-level PM_2.5 estimates by constructing spatiotemporal models that are a space–time function using the Bayesian Maximum Entropy (BME) method. The BME data fusion method integrates daily observed PM_2.5 obtained from the U.S. Environmental Protection Agency (EPA) Air Quality System nationwide monitoring data, yielding daily PM_2.5 observations at a total of 2 001 monitoring station locations across 48 states along with the District of Columbia. These monitoring data were combined with the output of chemical transport models (CTMs), which characterize the local emission sources, meteorology, chemical transformations, and transport of pollutants (25). In the first step of the BME data fusion, a Regionalized Air quality Model Performance (RAMP) analysis (25) was conducted to perform a localized bias correction of CTM outputs. The CTM outputs were obtained at an hourly time scale and at the finest spatial grid resolution available, which varied from 36 km for the nationwide domain down to 4 km in some highly populated areas (25,27). These resolutions provided a fine scale description of PM_2.5 across the nationwide domain. Next, we performed a BME data integration to obtain geostatistical estimates of PM_2.5, using EPA observations as hard data and the RAMP corrected CTM output as soft data. This BME estimation resulted in maps of PM_2.5 that combined the precision of observations and the fine scale resolution of CTM outputs. The statistically validated BME model (average Pearson’s R² = .70) was then applied to each geocoded residential location to estimate daily ambient PM_2.5 concentrations in 1999–2010. The daily estimates were aggregated to the 1-year moving average exposure, accounting for residential mobility and length of stay at each residence. We focused on annual average, as did previous studies showing an association with increased dementia mortality (19), hospitalization (28,29), and incidence (30), considering their underlying pathogenic processes may be accelerated by late-life exposures to PM_2.5 (31).

Statistical Analysis

Cause-specific proportional hazard models accounting for death as the competing event were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for racial disparities (Black vs White women) in AD risk, comparing models with and without adjustment for time-varying PM_2.5 exposure. Follow-up time was calculated as days from WHI-HT inception (32) to the date of the cognitive assessment that triggered the referral for additional testing subsequently leading to classification of primary AD, death, or the date of last cognitive assessment (through December 31, 2010), whichever came first. In Cox models, cases of incident probable dementia that were not classified as exclusively AD-related (eg, vascular, mixed type, unknown, other dementia-related classifications) were excluded, while women with mild cognitive impairment remained as they were still considered at risk for developing AD during the follow-up. The proportional hazard assumption was supported by the proportionality test. Covariates considered in our analyses (details are given in Supplementary Methods) included age, socioeconomic characteristics (education, family income, employment status), lifestyle factors (smoking status, alcohol use, physical activity), clinical characteristics (cardiovascular diseases [CVDs] , hypertension, history of depression, body mass index [BMI], diabetes mellitus, postmenopausal hormone treatment), and WHI-HT intervention assignment. To evaluate the possibility of differential susceptibility to PM_2.5, a product term of race/ethnicity and PM_2.5 was included in models to examine whether the putative association between PM_2.5 and AD risk differed by race/ethnicity (Black vs White women).

The robustness of our findings were tested with several sensitivity analyses: further adjusting for census track-level socioeconomic characteristics of residential neighborhood (details are given in Supplementary Methods); adjusting for WHI clinic sites to account for possible spatial confounding by other unmeasured covariates; further adjusting for measures of social engagement (social strain, social support, social activity, living alone) (details are given in Supplementary Methods); further adjusting for a measure of stressful life events (details are given in Supplementary Methods); and defining the outcome as cases of incident probable dementia that were further classified as either AD or mixed type (ie, etiologies of both AD and vascular dementia).

All reported p values are 2-sided and values <.05 indicated statistical significance. Analyses were performed using SAS software version 9.4 (SAS Institute).

Results

Compared to women missing data (n = 687), those with complete data were younger, more likely to be White, residing in the Northeast, and physically active, had higher household incomes and educational attainment, had lower annual average PM_2.5 exposure, and were less likely to have depression or diabetes (Supplementary Table 1). During follow-up (mean 8.3 years), 158 (137 White and 21 Black) incident AD cases were identified, with incidence rates much higher in Black than White women (4.88 vs 2.41 cases per 1 000 person-years). Annual average PM_2.5 exposure was higher in Black (mean: 14.38; median: 14.38; range: 7.32–21.19 µg/m³) compared to White (mean: 12.55; median: 12.13; range: 3.86–25.54 µg/m³) women. Compared to White women, Black women were slightly younger and more likely to reside in the South and currently smoke, had lower household incomes and educational attainment, but were less likely to be drinking alcohol or physically active. Black women were more likely to have a higher BMI and history of diabetes and hypertension. They were also more likely to live alone, experience more social strain and activity, report greater stressful life events, but had less social support (Table 1). Compared to their counterparts, women exposed to higher levels of PM_2.5 were older, were more likely to reside in the South and Midwest, had lower household incomes, were less likely to drink, were less physically active, were more likely to have a history of hypercholesterolemia and hypertension, and were more socially active (Table 1).

Table 1.

Population Characteristics at Baseline by Annual Average PM_2.5 and Stratified by Race/Ethnicity in the Biracial WHIMS Cohort, 1996–2010 (N = 6 485)

		Annual Average PM2.5a		Racial/Ethnic Groupb
Characteristic	N	(N = 6 485)	P Valuec	Blacks (n = 481)	Whites (n = 6 004)	P Valued
U.S. geographic region			<.001			<.001
Northeast	1 795	12.1 (1.7)		102 (21.2)	1 693 (28.2)
South	1 363	13.5 (2.1)		212 (44.1)	1 151 (19.2)
Midwest	1 642	13.0 (2.3)		110 (22.9)	1 532 (25.5)
West	1 685	12.3 (4.1)		57 (11.9)	1 628 (27.1)
Age			<.001			.003
65–69	3 026	12.6 (2.8)		260 (54.1)	2 766 (46.1)
70–74	2 334	12.6 (2.9)		149 (31.0)	2 185 (36.4)
≥75	1 125	13.0 (2.7)		72 (15.0)	1 053 (17.5)
Income			<.001			<.001
<$10 000	323	12.9 (2.8)		73 (15.2)	250 (4.2)
$10 000–$34 999	3 142	12.6 (2.8)		230 (47.8)	2 912 (48.5)
$35 000–$74 999	2 149	12.8 (2.7)		128 (26.6)	2 021 (33.7)
≥$75 000	640	13.0 (2.7)		42 (8.7)	598 (10.0)
Missing	231	12.4 (2.5)		8 (1.7)	223 (3.7)
Education			.08			<.001
<High school	424	12.9 (2.7)		72 (15.1)	352 (5.9)
High school/GED	1 419	12.6 (2.7)		73 (15.3)	1 346 (22.5)
>High school	4 624	12.7 (2.8)		332 (69.6)	4 292 (71.7)
Employment status			.06			.36
Currently employed	1 170	12.8 (2.7)		79 (16.5)	1 091 (18.2)
Not working	669	12.9 (2.8)		44 (9.2)	625 (10.4)
Retired	4 626	12.6 (2.8)		357 (74.4)	4 269 (71.3)
Alcohol intake			<.001			<.001
Nondrinker	776	13.2 (2.8)		92 (19.4)	684 (11.5)
Past drinker	1 234	12.8 (2.8)		155 (32.7)	1 079 (18.1)
<1 drink per day	3 617	12.6 (2.7)		215 (45.4)	3 402 (57.1)
>1 drink per day	803	12.5 (2.7)		12 (2.5)	791 (13.3)
Smoking status			.28			.004
Never smoked	3 389	12.7 (2.8)		228 (48.6)	3 161 (53.3)
Past smoker	2 579	12.6 (2.8)		193 (41.2)	2 386 (40.2)
Current smoker	429	12.8 (2.6)		48 (10.2)	381 (6.4)
Moderate or strenuous activities ≥ 20 min			<.001			<.001
No activity	3 748	12.8 (2.8)		343 (71.6)	3 405 (56.8)
Some activity	309	12.9 (2.7)		20 (4.2)	289 (4.8)
2–4 episodes per week	1 281	12.5 (2.8)		67 (14.0)	1 214 (20.3)
>4 episodes per week	1 131	12.4 (2.8)		49 (10.2)	1 082 (18.1)
Body mass index, kg/m2			.73			<.001
<25	1 862	12.7 (2.7)		78 (16.4)	1 784 (29.9)
25–29	2 333	12.7 (2.8)		152 (31.9)	2 181 (36.5)
≥30	2 254	12.7 (2.8)		247 (51.8)	2 007 (33.6)
Cardiovascular disease			.11			.13
No	5 281	12.7 (2.8)		377 (80.0)	4 904 (82.8)
Yes	1 111	12.8 (2.8)		94 (20.0)	1 017 (17.2)
History of depression			.27			.27
No	5 845	12.7 (2.7)		422 (90.8)	5 423 (92.2)
Yes	502	12.8 (2.9)		43 (9.2)	459 (7.8)
Diabetes treated ever (pills or shots)			.72			<.001
No	6 095	12.7 (2.8)		413 (86.2)	5 682 (94.8)
Yes	380	12.7 (2.7)		66 (13.8)	314 (5.2)
Hypercholesterolemia requiring pills ever			.02			.08
No	5 245	12.6 (2.8)		371 (78.9)	4 874 (82.2)
Yes	1 155	12.8 (2.8)		99 (21.1)	1 056 (17.8)
Hormone therapy use ever			.18			.24
No	3 541	12.7 (2.6)		275 (57.2)	3 266 (54.4)
Yes	2 942	12.6 (3.0)		206 (42.8)	2 736 (45.6)
Hypertension ever			.01			<.001
No	3 915	12.6 (2.7)		194 (41.5)	3 721 (62.6)
Yes	2 497	12.8 (2.8)		273 (58.5)	2 224 (37.4)
WHI-HT intervention assignment			.79			.61
Control	3 309	12.7 (2.8)		240 (49.9)	3 069 (51.1)
Intervention	3 176	12.7 (2.7)		241 (50.1)	2 935 (48.9)
Social straine			.23			<.001
Low	4 162	12.6 (2.8)		216 (48.5)	3 946 (67.4)
High	2 136	12.7 (2.7)		229 (51.5)	1 907 (32.6)
Social supportf			.52			.002
Low	3 296	12.7 (2.8)		264 (59.3)	3 032 (51.9)
High	2 993	12.7 (2.8)		181 (40.7)	2 812 (48.1)
Social activityg			<.001			<.001
Low	3 262	12.5 (2.8)		174 (36.6)	3 088 (51.6)
High	3 202	12.8 (2.7)		301 (63.4)	2 901 (48.4)
Living aloneh			.10			.003
No	4 319	12.6 (2.7)		284 (61.1)	4 035 (67.7)
Yes	2 105	12.8 (2.8)		181 (38.9)	1 924 (32.3)
Stressful life eventsi			.13			<.001
Low	3 542	12.6 (2.8)		183 (40.3)	3 359 (57.3)
High	2 774	12.7 (2.7)		271 (59.7)	2 503 (42.7)

Notes: GED = General Educational Development tests; PM_2.5 = particulate matter with aerodynamic diameter <2.5 μm; WHI-HT = Women’s Health Initiative hormone therapy; WHIMS = Women’s Health Initiative Memory Study.

^aData are expressed as mean (SD) of participants.

^bData are expressed as number (percentage) of participants.

^c  p Value testing statistical difference across subcategories using analysis of variance tests.

^d  p Value testing the distribution of racial/ethnic groups across subcategories using chi-squared tests.

^eHow many of an individual’s social ties are characterized by strain.

^fAvailability of support, affection, and positive social interactions.

^gFrequency of engagement in social clubs/organizations and religious services/church activities.

^hWhether the participant lives alone.

ⁱStressful life events over the previous year.

In Table 2, we present the results of cause-specific Cox models examining racial/ethnic disparities in AD risk (Black vs White women, HR_{Black women}) before and after adjusting for PM_2.5 exposure. In crude analyses, AD risk was greater in Black compared to White women (HR_{Black women}: 2.10 [95% CI: 1.33–3.34], without PM_2.5-adjustment). Adjustment for age, region, socioeconomic factors, and lifestyle factors modestly reduced the racial disparities in AD (HR_{Black women}: 1.85 [95% CI: 1.14–3.01]), with income making a major contribution to this observed difference. Additional adjustment for history of depression, hormone therapy use, WHI-HT assignment, and cardiovascular risk profiles (Models 4 and 5, without PM_2.5-adjustment) did not further reduce the racial disparities. Adjustment for PM_2.5 exposure slightly reduced the racial disparities in AD risk by 3%–6% (Models 1–5, with PM_2.5-adjustment). In the model accounting for the full list of covariates (Model 5, with PM_2.5-adjustment), each interquartile range (IQR) increment (IQR: 3.73 µg/m³) in PM_2.5 was associated with a 28% increase in AD risk (HR_PM2.5: 1.28 [95% CI: 1.02–1.61]). These findings were robust in our sensitivity analyses (Supplementary Tables 2 and 3) further adjusting for residential neighborhood socioeconomic characteristics, WHI clinic site, measures of social engagement, or stressful life events. Additionally, these racial disparities were sustained when defining the outcome variable as AD or mixed type dementia, though the elevated AD risk associated with PM_2.5 exposure did not reach statistical significance (Supplementary Table 4).

Table 2.

Hazard Ratios of Alzheimer’s Disease Incidence in Black vs White Women Before and After Adjusting for Annual Average PM_2.5 Exposure

		Without PM2.5-Adjustment			With PM2.5-Adjustment
	N (cases)	HRBlack Women	95% CI	P Value	HRBlack Women	95% CI	P Value	HRPM2.5	95% CI	P Value
Crude	6 485 (158)	2.10	1.33, 3.34	.002	1.97	1.21, 3.22	.006	1.24	1.02, 1.51	.03
Adjusted
Model 1a	6 485 (158)	2.03	1.28, 3.21	.003	1.97	1.21, 3.21	.007	1.20	0.97, 1.50	.10
Model 2b	6 450 (156)	1.89	1.18, 3.02	.008	1.83	1.09, 3.06	.02	1.20	0.96, 1.50	.11
Model 3c	6 319 (150)	1.85	1.14, 3.01	.01	1.81	1.06, 3.07	.03	1.18	0.93, 1.48	.17
Model 4d	5 995 (136)	2.32	1.43, 3.76	<.001	2.22	1.30, 3.81	.004	1.24	0.98, 1.58	.07
Model 5e	5 798 (130)	2.41	1.44, 4.04	<.001	2.26	1.29, 3.96	.005	1.28	1.02, 1.61	.04

Notes: HR_{Black Women} = hazard ratio for the risk of incident Alzheimer’s disease of Black compared to White women; HR_PM2.5 = hazard ratio for the risk of incident Alzheimer’s disease associated with each interquartile range increment (3.73 µg/m³) of annual average PM_2.5 exposure; PM_2.5 = particulate matter with aerodynamic diameter <2.5 μm.

^aAdjusting for age and region.

^bModel 1 + socioeconomic factors (income, education, and employment status).

^cModel 2 + lifestyle factors (alcohol use, smoking, and physical activity).

^dModel 3 + history of depression, hormone therapy use before baseline, and Women’s Health Initiative’s hormone therapy intervention assignment.

^eModel 4 + cardiovascular risk profiles (body mass index, hypercholesterolemia, diabetes, hypertension, and cardiovascular disease).

In Figure 1, we present the association between PM_2.5 and AD risk, stratified by race/ethnicity. The association of increased AD risk with PM_2.5 exposure was 2 times stronger (range of interaction Ps: .002–.01) in Black women (range of HR_PM2.5: 2.10–2.60; ps < .01) than in White women (HR_PM2.5: 1.07–1.15; ps = .19–.58). These racial disparities in the PM_2.5–AD association were sustained in our sensitivity analyses where we defined the outcome variable as AD or mixed type dementia (Supplementary Table 5). An ad hoc analysis was conducted to examine whether cardiovascular risk factors or stressful life events more commonly found in Black populations could explain this observed racial difference in the PM_2.5 effect. After further adjusting for the interactions with these cardiovascular risk factors (Supplementary Table 6) and stressful life events (Supplementary Table 7), the positive PM_2.5–AD association was still 2–2.5 times stronger in Black than in White women (Supplementary Tables 8 and 9).

Figure 1.

Hazard ratios of Alzheimer’s disease incidence associated with annual average PM_2.5 exposure, stratified by race/ethnicity.

Notes: HR_PM2.5 = hazard ratio for the risk of incident Alzheimer’s disease associated with each interquartile range increment (3.73 µg/m³) of annual average PM_2.5 exposure; PM_2.5 = particulate matter with aerodynamic diameter <2.5 μm. HRs presented in the figure for all participants are the main effects reported in Table 2.

^a  p = .009 for interaction. ^bAdjusting for age and region, p = .004 for interaction. ^cModel 1 + socioeconomic factors (income, education, and employment status), p = .01 for interaction. ^dModel 2 + lifestyle factors (alcohol use, smoking, and physical activity), p = .010 for interaction. ^eModel 3 + history of depression, hormone therapy use before baseline, and Women’s Health Initiative’s hormone therapy intervention assignment, p = .008 for interaction. ^fModel 4 + cardiovascular risk profiles (body mass index, hypercholesterolemia, diabetes, hypertension, and cardiovascular disease), p = .002 for interaction.

Discussion

In this geographically diverse prospective cohort of community-dwelling older women, increased exposure to ambient PM_2.5 partially contributed to Black–White disparities in the risk for clinically defined AD. We also found Black women were ~2 times more likely to suffer from the adverse PM_2.5 effect on increased AD risk. These associations could not be explained by sociodemographics, lifestyle factors, measures of social engagement, stressful life events, and CVD and related comorbidities, and were robust after accounting for spatial confounding by neighborhood socioeconomic characteristics and WHI clinic site. These novel results provide the first direct evidence linking environmental pollutants with racial/ethnic disparities in AD risk. These findings may offer new impetus for studying environmental neurotoxins that disproportionately affect Black and other minority groups to help “decrease disparities in Alzheimer’s for ethnic and racial minority populations,” as mandated by the National Alzheimer’s Project Act (33).

Our data contribute to the current literature investigating why Black individuals had elevated incidence of AD. Although it has been suggested that genetics, lifestyle factors, and CVD and related comorbidities may explain the greater AD risk among Black individuals (1), several population-based U.S. cohort studies (34–43) still showed a 27%–140% higher AD risk in Black than in White individuals after adjusting for these conventional risk factors (Supplementary Table 10). A meta-analysis yielded a combined rate ratio of 1.64 (95% CI: 1.35–2.00), suggesting Black individuals are 64% more likely to develop AD than White (44). Compared to these earlier studies which mostly used localized populations, our observation of racial disparities in AD incidence in this nationwide, geographically diverse sample has greater generalizability. Previous reports of racial disparities in AD based on geographically diverse cohorts (36,40,42) did not control for regional difference or spatial confounding. In our study, the increased AD risk in Black women was adjusted for geographic region (adjusted models, Table 2) and was not sensitive to further adjustment for other spatial variations (Supplementary Table 2; Model 7), further supporting the internal validity of our findings. Additionally, previous studies did not explore the role of social factors, although their possible contribution to racial disparities in AD risk has been suggested (45). We found that the AD risk was 152% higher in Black older women after adjusting for the full set of conventional risk factors (Table 2) and persisted in sensitivity analyses further adjusting for measures of social engagement and stressful life events (Supplementary Table 3). The pervasive evidence of racial disparities in ADRD not explained by behavioral and biological factors points to the need to investigate the role of racism as a structural cause of accelerated brain aging (46). Environmental neurotoxins, including PM_2.5, are an example of environmental racism contributing to racial disparities in brain aging. We call for new research that moves us closer to a better understanding of the importance of environmental racism and its impact on the pathogenetic processes of ADRD.

Our study illustrates the importance for environmental-neuroepidemiologic studies of ADRD to include high-risk minority populations. After we adjusted for PM_2.5, the increased AD risk among Black women (85%–141% higher than White women) was reduced to 81%–126% (Table 2). This finding suggests that the environmental risk imparted by PM_2.5 may partially contribute to the racial disparities in AD risk. Several studies on Western countries have reported an association between long-term exposure to late-life PM_2.5 and ADRD (16) in White populations, while the present study is the first to depict this relationship in a biracial population. The REasons for Geographic and Racial Differences in Stroke cohort was the only other geographically diverse, biracial U.S. population used to study PM_2.5 neurotoxicity, but investigators did not find a significant association between PM_2.5 and incident cognitive impairment (47). The advances of environmental neurosciences of air pollution and brain aging need to be built on studying racially diverse populations, as the clinical manifestation of AD may differ across racial groups (1).

Our study findings also raised several questions concerning the environmental health disparities in AD as related to air pollution neurotoxicity on brain aging. The increased risk of AD associated with PM_2.5 was ~2 times greater in Black (HR_PM2.5: 2.10–2.60) than in White (HR_PM2.5: 1.07–1.15) women (Figure 1). Studies have demonstrated that Black individuals may be more susceptible to the adverse effect of PM_2.5 on several health outcomes (48–50), but our study was the first to examine the race-related environmental health disparities in brain aging. Although the reasons for this increased susceptibility among Black populations are unclear, our ad hoc analyses suggest that cardiovascular risk factors and stressful life events typically found to be more prevalent in Black individuals could not explain this finding (Supplementary Tables 8 and 9). Understanding the reasons for these greater health burdens from air pollution among Black populations is important, as the EPA’s Office of Environmental Justice mandates that no race should experience disproportionate negative consequences from environmental stressors (51). Black populations are likely much more vulnerable to the adverse health effects of ambient air pollution due to the lifetime adversities and injustices they have experienced as a result of structural racism. For example, the race-based residential segregation that has persisted in the United States for decades has shaped the racial inequities in exposure to air pollution (52), quality of education (53), socioeconomic opportunities (54), psychosocial stress (55), and access to high-quality food (56), all contributing to brain aging (31). These lifetime experiences shape cognitive reserve, which has important implications for cognitive aging and AD-related brain pathologies (57). Residing in highly segregated neighborhoods in young adulthood has even been directly linked to worse cognitive performance in midlife among Black individuals (58). Health disparities related to aging emerge over the life course as a product of several factors that likely interact across many levels, such as environmental (eg, toxins, exposures) and sociocultural (eg, institutional racism), to compromise health (59). Therefore, future studies should investigate whether the neurotoxic effects of PM_2.5 may interact with these contextual factors resulting from structural racism to impact brain aging. Additionally, the collective evidence demonstrating the increased susceptibility to the health effects of PM_2.5 among Black populations highlights the need for stronger enforcement of the Clean Air Act’s mandate to provide safe margins of air quality for susceptible populations (60).

We recognize several limitations of our study. First, limited access to quality education and test anxiety from perceived discrimination (61) may lower cognitive test performance in Black individuals, leading to an overdiagnosis of AD in Black populations and an overestimation of the racial/ethnic differences (3). However, the longitudinal nature of WHIMS protocols alleviates concerns of overdiagnosis (1) and if present, such bias would likely be nondifferential with respect to PM_2.5 exposure. Second, although PM_2.5 was estimated with a statistically cross-validated (average Pearson’s R² = .70) model, the possible nondifferential measurement errors would attenuate the observed associations. Third, we did not investigate other environmental exposures that may be more prevalent among Black individuals (eg, occupational dust) or other exposure sources of PM_2.5 (eg, traffic emissions). Fourth, our covariates measured at baseline do not capture the entirety of an individual’s lifetime experience (eg, adverse life experiences including racial discrimination in early life), which may contribute to the racial differences in AD risk (45). Fifth, the number of Black participants is relatively small compared to the number of White participants; however, the relative proportion of Black and White women in WHIMS was comparable to U.S. women aged 65 years and older in the 1990s (62). Sixth, our findings may not be generalizable to men or younger women.

This study also had several strengths. Our study utilized a large, biracial sample of Black and White individuals prospectively followed annually up to 15 years. The population was a nationwide, geographically diverse cohort, capturing individuals with varying backgrounds and allowing sufficient PM_2.5 gradients. Our PM_2.5 estimates allowed us to explore environmental risk factors potentially contributing to the racial disparities in AD risk. We accounted for competing risk of death, reducing the risk of biased effect estimates (63). Lastly, the rich data allowed us to adjust for potential confounding and reduce other sources of biases.

In summary, this study on a geographically diverse, biracial cohort showed that Black older women had a greater risk for clinical AD than White women. Such racial/ethnic differences persisted after accounting for conventional risk factors, with the remaining disparities partly explained by late-life exposure to PM_2.5. Additionally, the association between PM_2.5 and increased AD risk was much stronger in Black, compared to White, older women. These results provide new evidence for the potential role of environmental neurotoxins in the racial/ethnic disparities in AD risk and may help inform strategies aimed towards AD prevention in at-risk minority populations.

Acknowledgment

We would like to thank the women who participated in the Women’s Health Initiative suite of studies.

Funding

This work was supported by a grant from the Alzheimer’s Association (AARF-19-591356). The air pollution exposure data were derived from the grant R01AG033078 (PI: J.-C.C.) supported by the National Institute on Aging. D.Y., X.W., M.A.E., and J.-C.C. are supported by RF1AG054068 (PI: J.-C.C.). D.Y. and J.-C.C. are also supported by P01AG055367 and R01ES025888. Research work of D.Y. and J.-C.C. is also supported by the National Institute of Environmental Health Sciences (5P30ES007048). H.C.C. is supported by P50AG05142. M.A.E. receives funding from the Wake Forest Alzheimer’s Disease Core Center (P30AG049638-01A1). The Women’s Health Initiative (WHI) program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through contracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C, HHSN268201100003C, HHSN268201100004C, and HHSN271201100004C. For a list of all the investigators who have contributed to WHI science, please visit: https://www.whi.org/researchers/Documents%20%20Write%20a%20Paper/WHI%20Investigator%20Long%20List.pdf. The WHI Memory Study was funded as an ancillary study to the WHI by Wyeth Pharmaceuticals, Inc.; Wake Forest University; the National Heart, Lung, and Blood Institute, National Institutes of Health; and the National Institute of Aging, National Institutes of Health (contract number HHSN271201100004C). The Women’s Health Initiative Magnetic Resonance Imaging Study was funded by contract N01-WH-44221, from the National Heart, Lung, and Blood Institute.

Conflict of Interest

None declared.

Author Contributions

Concept and design: D.Y. and J.-C.C. Acquisition, analysis, or interpretation of data: D.Y., X.W., T.G., M.G., M.L.S., W.V., F.L., H.C.C., J.E.M., R.B.W., M.A.E., and J.-C.C. Revision of the manuscript for intellectual content: D.Y., X.W., T.G., M.G., M.L.S., M.N.B., N.F.W., K.K., L.G., J.E.M., M.A.E., and J.-C.C.