Race and Ethnicity, Hypertension, and Neuroimaging Markers of Brain Aging

Abstract

Background and Objectives

The mechanisms underlying racial/ethnic differences in dementia incidence and pathology are multifactorial, and hypertension represents an actionable target for reducing these differences. We aimed to estimate the extent to which controlling for hypertension mediates racial/ethnic inequities in neuroimaging markers of brain aging.

Methods

The Health and Aging Brain Study–Health Disparities cohort is a highly phenotyped, racially and ethnically diverse cohort of cognitive aging. We used marginal structural models with inverse probability weights to estimate total and controlled direct effects of race/ethnicity, hypertension, and systolic blood pressure (SBP) at baseline, with neuroimaging markers measured on average 2 years later. Neuroimaging markers of brain aging were measured at the 2-year follow-up.

Results

Among Black and Hispanic participants with any neuroimaging data at the second visit (overall N = 1,347), 68% and 71% were women, 75% and 67% had hypertension, and the mean age was 61 and 63 years, respectively. Black and Hispanic participants had greater white matter hyperintensity volume (WMHV) compared with non-Hispanic White (NHW) participants (n = 1,333, β [95% CI]: Black 2.08 [1.68–2.59], Hispanic 0.99 [0.91–1.08]). After analytically setting hypertension status to absent, Black-NHW inequities in WMHV were attenuated (β [95% CI]: 1.3 [1.01–1.65]). Black participants had lower amyloid deposition compared with NHW participants (n = 679, β [95% CI]: −0.29 [−0.46 to −0.12]), but analytically controlling for hypertension did not appreciably change estimates. Compared with NHW participants, Hispanic participants had lower Alzheimer disease meta–region of interest cortical thickness (n = 1,005, β [95% CI]: −0.20 [−0.34 to −0.07]), but neither hypertension nor SBP significantly mediated this difference. Medial temporal lobe tau-PET standardized uptake value ratio did not significantly differ in Black or Hispanic participants compared with NHW participants (n = 408).

Discussion

Black-NHW inequities in subclinical cerebral small vessel disease may be mitigated by population-level efforts to reduce hypertension prevalence. Future studies should extend this work to examine clinical outcomes.

Introduction

Racial and ethnic inequities in Alzheimer disease and Alzheimer disease–related dementias (AD/ADRD) remain a major public health priority^1,2 and may be due to differential distribution of age-related brain pathologies across racial and ethnic groups. In neuroimaging and autopsy studies of racially and ethnically diverse cohorts, cerebral small vessel disease (cSVD) is prevalent and often co-occurs with AD pathology.^3,4 By contrast, amyloid positivity appears less prevalent among Black and Hispanic Medicare beneficiaries compared with similar non-Hispanic White (NHW) counterparts,⁵ although other studies do not observe differences across groups after adjustment for demographic covariates.^6,7 Patterns of gray matter atrophy have also differed across racial/ethnic groups in several large neuroimaging studies,^8,9 although others have not observed such differences.¹⁰ As such, public health prevention strategies should be tailored to target the most common pathology among these minoritized groups.

Racial and ethnic inequities in AD/ADRD are mirrored by inequities in hypertension,¹¹ a major modifiable risk factor of dementia.¹² The relationship between hypertension and cerebrovascular disease is well established, but its association with neurodegenerative pathology is less clear.¹³ Greater midlife systolic blood pressure (SBP) and diastolic blood pressure (DBP) have been associated with MRI measures of brain atrophy and advanced brain age.^14,15 However, there is mixed evidence on how hypertension affects amyloid deposition. Some studies do not show a significant association between hypertension and amyloid positivity or burden,^14,16 while in other community-based studies, hypertension was significantly associated with greater amyloid deposition in APOE ε4 carriers.^17,18 It remains unclear whether the reduction in hypertension prevalence at a population level would reduce racial/ethnic differences in cerebrovascular disease, neurodegenerative pathology, or both, and examining this question may inform primary prevention strategies for racial/ethnic inequities in dementia burden.

In this study, we used a causal mediation approach¹⁹ to estimate the extent to which hypertension mediated racial and ethnic differences in neuroimaging markers of brain aging. We hypothesized that hypertension would most strongly mediate racial/ethnic differences in MRI-derived measures of cSVD compared with neurodegeneration (namely amyloid or tau burden as measured by PET imaging or gray matter atrophy as measured by brain MRI).

Methods

Cohort Description and Sample Selection

As previously described,²⁰ the Health and Aging Brain Study–Health Disparities (HABS-HD) is an ongoing, community-based cohort study based at the University of North Texas Health Science Center in Fort Worth, Texas, that recruited Hispanic and NHW community members in the Dallas-Fort Worth metropolitan area starting in 2017 as part of the Health and Aging Brain among Latino Elders Study.²¹ Recruitment of Black participants began in 2021 with broader enrollment of NHW and Hispanic participants aged 30 years or older.²⁰ With a community-based participatory research approach, participants were invited to complete a standardized assessment, including a clinical interview, neuropsychological assessment, blood draw, and neuroimaging.^20,21 Willingness to undergo neuroimaging was part of the inclusion criteria for HABS-HD.²⁰

We selected participants who were dementia-free (determined by consensus diagnosis using a combination of neuropsychological testing and the Clinical Dementia Rating Scale–Sum of Boxes scores, as previously described²⁰) at baseline with either MRI or PET data available at their second visit (eFigure 1). Using the second-visit neuroimaging measure enabled us to optimize causal inference by maintaining temporal ordering of the exposure and mediators being measured before the outcome.^22,23

Standard Protocol Approvals, Registrations, and Patient Consents

This study was approved by the North Texas Regional Institutional Review Board, and all participants provided written informed consent.

Exposure: Race/Ethnicity

Race/ethnicity was self-reported at baseline and modeled as a categorical variable with NHW participants as the reference group because we conceptualize NHW participants as the most socially advantaged group. We conceptualize race/ethnicity as social constructs representing the lived experience of identifying or being identified as Black or White race or Hispanic ethnicity in the United States. We assert that Black race is an incomplete proxy for experiences of structural anti-Black racism, consistent with previous scholarship,²⁴ and that Hispanic ethnicity represents a construct that captures group cultural identity in addition to a relational dimension that contextualizes the group relative to the majority.²⁵

Mediators: Hypertension and SBP

Hypertension diagnosis was determined at baseline by consensus definition and reviewed by a medical professional associated with HABS-HD, as previously described.²⁰ Hypertension was defined as self-reported medical history of hypertension, or at least 2 BP readings in the same visit showing SBP ≥140 or DBP ≥90 (e.g., based on consensus definition); in addition, those who reported use of antihypertensive medications were defined as having hypertension in this analysis. Hypertension was modeled as a binary variable, with no hypertension as the reference group. We also modeled SBP because of its strong association with cSVD.²⁶ BP was measured using standardized protocols and measured twice per visit; the value used was the person-level average of 2 measures. SBP was modeled continuously in 2 ways: centering on the mean of SBP in NHW participants and centering on 120 mm Hg (i.e., the upper limit of intensive BP control as defined by the National Heart, Lung, and Blood Institute expert panel in 2007 and tested in the Systolic Blood Pressure Intervention Trial [SPRINT]²⁷), effectively setting these values as the reference value of 0.

Outcomes: Neuroimaging Markers of Brain Aging

Neuroimaging was obtained at the second visit (on average approximately 2 years after baseline, when hypertension was measured), which enabled us to optimize causal inference by maintaining temporal ordering of the exposure and mediators being measured before the outcome.^22,23

Global Amyloid-PET Standardized Uptake Value Ratio

Amyloid-PET scans were available for 679 participants at the second visit. Scans were collected on a Siemens Vision 40 whole-body PET/CT or a Siemens MCT 20 scanner. For the Siemens Vision 450 whole-body PET/CT, a 4-frame by 5-minute dynamic emission acquisition was started 90 minutes after injection of 8.1 mCi ± 10% of florbetaben after a low-dose CT scan used for attenuation correction.²¹ Images were processed by iterative reconstruction (8 iterations and 5 subsets). On the Siemens MCT 20 scanner, the acquisition was identical except that there was a 20-minute continuous emission. Scans on the MCT20 were processed by iterative reconstruction with 4 iterations and 24 subsets. A global amyloid deposition measure was calculated across frontal, anterior/posterior cingulate, lateral parietal, and lateral temporal cortices, based on FreeSurfer-derived regions of interest (ROIs). Using the whole cerebellum as a reference, a standardized uptake value ratio (SUVR) was calculated for global amyloid deposition. Amyloid-PET SUVR was z-scored to facilitate comparisons across outcomes.

Medial Temporal Lobe Tau-PET SUVR

Tau-PET scans were available for 408 participants at the second visit. Tau deposition was measured using tau-PET imaging with ¹⁸F-PI-2620 (PI-2620) using the Siemens Vision 450 whole-body PET/CT scanner. As previously described,²¹ 6-frame by 5-minute dynamic emission acquisition was performed, starting 45–75 minutes after injection of 5 mCi ± 10% ¹⁸F-PI-2620 and immediately after a CT attenuation scan. Images were reconstructed immediately after the 30-minute emission scan using identical reconstruction parameters to the FBB scans collected on Vision 450. As for amyloid-PET, FreeSurfer ROIs were used to identify the medial temporal lobe (MTL), and using the inferior cerebellar gray matter as a reference, SUVR was calculated. We chose MTL tau as a focus, given its clinical importance for AD pathophysiology.²⁸ Tau-PET SUVR was z-scored to facilitate comparisons across outcomes.

AD Meta-ROI Cortical Thickness

Cortical thickness (in mm) was available for 1,005 participants at the second visit. MRIs were obtained using a Siemens Magnetom Skyr whole-body scanner or Siemens Magnetom Vida 3T scanner.²⁹ Regional cortical thickness was measured using FreeSurfer software 5.3.0. Based on prior work,³⁰ an AD meta-ROI was calculated as the surface area-weighted average of the mean bilateral cortical thickness in the entorhinal cortex, fusiform, inferior temporal gyri, and middle temporal gyri. AD meta-ROI cortical thickness was z-scored to facilitate comparisons across outcomes.

White Matter Hyperintensity Volume

White matter hyperintensity volume (WMHV, in cm³) data were available for 1,333 participants at the second visit and were obtained from FLAIR and T1-weighted MR images using the Statistical Parametric Mapping Lesion Segmentation Toolbox as previously described.^21,31 WMHV was log-transformed because of its right-skewed distribution, and estimates were back-transformed to represent percent change in WMHV.

Covariates of Interest

Age at baseline, sex/gender (men vs women), years of education, income, insurance status (has insurance vs uninsured), and smoking status were self-reported and obtained at the baseline visit with standardized protocols as previously described.²¹ Antihypertensive medication use was defined as self-reported use of angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, beta-blockers, and diuretics. Body mass index (BMI) was calculated as kg/m². Physical activity was measured using the Rapid Assessment of Physical Activity, with scores for aerobic activity and strength.³² Social support was measured using the Interpersonal Support and Evaluation List.³³ Depressive symptoms were measured using the Geriatric Depression Scale.³⁴ Total cholesterol, low-density lipoprotein cholesterol (LDL-C), hemoglobin A1c (HbA1c), estimated glomerular filtration rate (eGFR), and APOE ε4 carrier status were obtained from laboratory analysis of blood samples collected at baseline. Depression was defined as a medical history of depression or Geriatric Depression Scale score ≥10. Dyslipidemia was defined as a medical history of high cholesterol or LDL-C ≥120, total cholesterol ≥240, or triglycerides ≥200. Cardiovascular disease (CVD) was defined as a medical history of heart attack, heart failure, cardiomyopathy, atrial fibrillation, or heart valve replacement. Diabetes was defined as a medical history of diabetes or HbA1c ≥ 6.5. Total intracranial volume was obtained using MRI processing as described above. MRI and PET scanner types are as described above.

Statistical Analysis

We generated descriptive statistics stratified by race/ethnicity using means and standard deviations for continuous variables and frequencies and percentages for categorical variables. We also generated descriptive statistics stratified by availability neuroimaging markers at visit 2. Correlation between covariates was also examined (eFigure 2).

For our causal mediation analysis, we chose a modeling strategy that would quantify how a theoretical intervention on hypertension or SBP would mitigate racial and ethnic differences in neuroimaging markers of brain aging. Therefore, we aimed to estimate controlled direct effects (CDEs) corresponding to racial/ethnic differences in the distribution of neuroimaging markers of brain aging, after controlling for hypertension or SBP. Separate models were fit with either hypertension or SBP as the mediator of interest. Figure 1 illustrates our conceptual framework. Because confounders of the mediator-outcome relationship were also affected by the exposure, we used marginal structural models with stabilized inverse probability weights (IPWs) to estimate total effects (TEs) and CDEs.³⁵ We interpret the TE as the difference in the expected mean of each neuroimaging outcome for Black or Hispanic participants compared with NHW participants. We interpret the CDE as the difference in the expected mean of each neuroimaging outcome for Black or Hispanic participants compared with NHW participants, while holding hypertension (present or absent) or mean SBP (mean of NHW participants or 120 mm Hg) constant across the entire sample.^35,36 In the case of null TEs, we moved forward with estimating CDEs, given previous work showing that mediating effects in opposite directions may offset TEs.³⁷

Figure 1. Conceptual Framework.

Conceptual framework informing analysis. AD = Alzheimer disease; MTL = medial temporal lobe; ROI = region of interest; SES = socioeconomic status; SUVR = standardized uptake value ratio.

Owing to concerns for missing covariate data at random for IPW models,³⁸ the mice package³⁹ was used to generate multiply imputed data sets (m = 20) for the above-described covariates, using predictive mean matching for continuous variables, logistic regression for binary variables, and polytomous logistic regression for categorical variables. eFigure 3 illustrates the missingness of covariates in our analytic sample. Analyses were performed in each imputed data set, and estimates were pooled using Rubin's rules.⁴⁰

Calculating IPWs

For all weights, we computed stabilized IPWs as the ratio of the marginal probabilities (for hypertension) or probability density functions (for SBP) as the numerator and the conditional probabilities or probability density functions as the denominator.^41,42 Covariate balance was evaluated using the cobalt package⁴³ to examine mean differences between unweighted and weighted data sets using a threshold of 0.2. All weights were truncated at the 1st and 99th percentiles and had a mean of approximately 1, with no weights exceeding a max of 10 (eFigure 4 shows Love plots and eTable 1 presents summary statistics for IPWs).

Generating Exposure and Mediator IPWs

Marginal and conditional probabilities of racial/ethnic group identity (Black, Hispanic, or NHW) were estimated using multinomial regression.³⁵ Models for conditional probabilities were adjusted for age and sex/gender to ensure balance across race/ethnicity, sex/gender, and age.

Marginal and conditional probabilities of having hypertension were estimated using logistic regression. To estimate the conditional probability, we regressed hypertension diagnosis on race/ethnicity, confounders not affected by the exposure (age, gender), and confounders potentially affected by the exposure (years of education, income, insurance status, smoking status, physical activity, LDL-C, total cholesterol, eGFR, blood glucose, HbA1c, BMI, clinical diagnoses of dyslipidemia and diabetes, and APOE ε4 carrier status [for outcomes of amyloid-PET and tau-PET SUVR]). The marginal and conditional probability density functions of SBP were estimated from linear regression models.⁴² The conditional probability density function was estimated using similar covariates as for hypertension above, with the addition of antihypertensive medication use.

Generating IPWs for Selection Bias

We further computed IPWs to account for potential selection bias due to differential availability of neuroimaging data, with the goal of re-weighting analyses to represent the dementia-free sample at baseline.⁴⁴ Marginal and conditional probabilities of having neuroimaging data at visit 2 were estimated using logistic regression. To estimate the conditional probability, we regressed a binary measure of whether the participant had a given neuroimaging marker on race/ethnicity, age, gender, years of education, income, insurance status, smoking status, social support, BMI, physical activity, depressive symptoms, LDL-C, total cholesterol, eGFR, blood glucose, HbA1c, and clinical diagnoses of depression, dyslipidemia, CVD, and diabetes.

Marginal Structural Models

We used the survey package⁴⁵ to fit weighted generalized linear models with Gaussian link to estimate the TEs and CDEs with robust estimation of standard errors. For estimation of the TE, the weight used was a product of the exposure and selection weights. For estimation of the CDEs, the weight used was a product of the exposure, mediator, and selection weights. Outcome models were adjusted for MRI or PET scanner and intracranial volume for WMHV. Alpha was set at 0.05.

To further characterize the CDEs examined in the causal mediation analyses, we estimated exposure-mediator and mediator-outcome associations, based on previous work.¹⁹ In the baseline sample, we fit logistic and linear models to test the associations between the exposure (race/ethnicity) and hypertension and SBP, respectively, adjusted for age, sex, years of education, income, insurance status, smoking status, social support, BMI, physical activity, depressive symptoms, LDL-C, total cholesterol, eGFR, blood glucose, HbA1c, and clinical diagnoses of depression, dyslipidemia, CVD, and diabetes.

Data Availability

We used publicly available, deidentified data from HABS-HD data release 6 (apps.unthsc.edu/itr/our). The analysis was conducted using R version 4.4.2, RStudio version 2024.12.0+467, and the sample code for this analysis is available at github.com/michelle-caunca/HABS_CM_RaceEthHTNImg. This article was drafted in line with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines for observational epidemiology research.⁴⁶

Results

Table 1 lists descriptive statistics stratified by race/ethnicity (Table 1). In particular, Black and Hispanic participants were younger and had fewer years of education, greater BMI, and a greater proportion of participants with no insurance, hypertension, and mild cognitive impairment, compared with NHW participants. Patterns across racial/ethnic groups were similar across neuroimaging subsamples (eTable 2). Compared with NHW participants, Black and Hispanic participants had significantly greater odds of hypertension and had a greater average SBP (eTable 3).

Table 1.

Sample Characteristics, Stratified by Race/Ethnicity

	White (N = 634)a	Black (N = 141)a	Hispanic (N = 572)a
Age at baseline (y)	68 (8)	61 (7)	63 (8)
Sex/gender
Men	266 (42)	45 (32)	167 (29)
Women	368 (58)	96 (68)	405 (71)
Years between visits 1 and 2	2 (1)	2 (1)	2 (1)
Education (y)	15.8 (2.4)	14.9 (2.6)	10.3 (4.3)
Has no insurance	16 (2.5)	7 (5.0)	127 (22)
Income (dollars)	90,628 (77,831)	67,835 (63,509)	44,209 (69,880)
Social support score	42 (6)	42 (6)	40 (6)
Ever smoker	247 (40)	40 (33)	182 (35)
BMI (kg/m2)	29 (6)	34 (9)	31 (6)
SBP (mm Hg)	132 (18)	135 (18)	138 (19)
DBP (mm Hg)	81 (10)	87 (11)	82 (10)
Rapid assessment of physical activity score, aerobic	5 (2)	4 (2)	4 (1)
Rapid assessment of physical activity score, strength, and flexibility	2 (1)	1 (1)	1 (1)
Geriatric depression score	5 (5)	5 (5)	6 (6)
Taking BP medications at baseline	275 (43)	72 (51)	248 (43)
Hypertension	404 (64)	106 (75)	381 (67)
Depression	200 (32)	46 (33)	196 (34)
Mild cognitive impairment	71 (11)	57 (40)	99 (17)
Dyslipidemia	447 (71)	91 (65)	419 (73)
Cardiovascular disease	59 (9.3)	8 (5.7)	29 (5.1)
Diabetes	84 (13)	44 (31)	204 (36)
LDL (mg/dL)	103 (33)	97 (32)	106 (34)
Glucose (mg/dL)	102 (26)	104 (30)	116 (45)
HbA1c (%)	5.57 (0.77)	6.09 (1.10)	6.40 (1.57)
eGFR (mL/min/1.73 m2)	76 (15)	75 (18)	85 (17)
Total cholesterol (mg/dL)	183 (40)	169 (39)	184 (39)
APOE4 allele positive	177 (28)	48 (36)	102 (18)
WMHV (cm3)	0.00 (0.00, 0.00)	1.29 (0.39, 4.94)	0.00 (0.00, 0.00)
Global amyloid-PET SUVR	1.01 (0.97, 1.14)	1.00 (0.97, 1.04)	1.00 (0.96, 1.05)
MTL tau-PET SUVR	1.12 (1.05, 1.19)	1.14 (1.06, 1.23)	1.14 (1.07, 1.21)
AD meta-ROI cortical thickness (mm)	2.74 (0.13)	2.78 (0.11)	2.75 (0.13)
Intracranial volume (cm3)	1,455 (144)	1,411 (124)	1,370 (126)

Abbreviations: AD = Alzheimer disease; BMI = body mass index; BP = blood pressure; DBP = diastolic BP; eGFR = estimated glomerular filtration rate; HbA1c = hemoglobin A1c; LDL = low-density lipoprotein; MTL = medial temporal lobe; ROI = region of interest; SBP = systolic BP; SUVR = standardized uptake value ratio; WMHV = white matter hyperintensity volume.

^aMean (SD); n (%); median (Q1, Q3).

Figure 2 presents the TEs and CDEs for PET imaging markers of brain aging. Compared with NHW participants, Black participants had lower global amyloid-PET SUVR (TE β [95% CI]: −0.29 [−0.46 to −0.12]). Global amyloid-PET SUVR did not significantly differ in Hispanic participants compared with NHW participants (TE β [95% CI]: −0.05 [−0.24 to 0.14]). Differences in global amyloid-PET SUVR by race/ethnicity did not significantly change after setting hypertension to present or absent, or mean SBP to mean for NHW participants or 120 mm Hg (CDEs, Figure 2A). MTL tau-PET SUVR did not differ in Black or Hispanic participants compared with NHW participants (TE β [95% CI]: Black 0.10 [−0.29 to 0.49], Hispanic 0.17 [−0.16 to 0.5]). MTL tau-PET SUVR did not significantly differ by race/ethnicity after setting hypertension to present or absent, or mean SBP to mean for NHW participants or 120 mm Hg (CDEs, Figure 2B). Neither hypertension nor SBP was significantly associated with global amyloid-PET or MTL tau-PET SUVR (eTable 3).

Figure 2. Total and Controlled Direct Effects From Marginal Structural Models for PET Imaging Markers of Brain Aging.

(A) Global amyloid-PET SUVR. (B) MTL tau-PET SUVR. Outcomes z-scored to facilitate comparisons. Estimates from marginal structural models calculated after pooled analysis across 20 multiply imputed data sets. CDE = controlled direct effect; HTN = hypertension; MTL = medial temporal lobe; NHW = non-Hispanic White; SBP = systolic blood pressure; SUVR = standardized uptake value ratio.

Figure 3 presents the TEs and CDEs for structural MRI markers of brain aging. Compared with NHW participants, Hispanic participants had lower AD meta-ROI cortical thickness (TE β [95% CI]: −0.20 [−0.34 to −0.07]). AD meta-ROI cortical thickness did not differ in Black participants compared with NHW participants (TE β [95% CI]: 0.11 [−0.15 to 0.37]). Differences in AD meta-ROI cortical thickness by race/ethnicity did not significantly change after setting hypertension to present or absent, or mean SBP to mean for NHW participants or 120 mm Hg (CDEs, Figure 3). Neither hypertension nor SBP was significantly associated with AD meta-ROI cortical thickness (eTable 3).

Figure 3. Total and Controlled Direct Effects From Marginal Structural Models for Structural MRI Markers of Brain Aging.

(A) AD meta-ROI cortical thickness. (B) White matter hyperintensity volume. Outcomes z-scored to facilitate comparisons. Estimates from marginal structural models calculated after pooled analysis across 20 multiply imputed data sets. AD = Alzheimer disease; CDE = controlled direct effect; HTN = hypertension; NHW = non-Hispanic White; ROI = region of interest; SBP = systolic blood pressure.

Compared with NHW participants, Black and Hispanic participants had higher WMHV (TEs β [95% CI]: Black 2.08 [1.68–2.59], Hispanic 0.99 [0.91–1.08]). After setting the mediator level to hypertension absent for all participants, Black participants continued to have greater WMHV compared with NHW participants, but the CDE was lower than the TE (CDE β [95% CI]: 1.3 [1.01–1.65]). After setting the mediator level to hypertension present for all participants, Black participants continued to have greater WMHV compared with NHW participants, and the CDE was higher than the TE (CDE β [95% CI]: 4.02 [3.18–5.08]). After setting the mediator levels to mean SBP for NHW participants and 120 mm Hg, Black-NHW differences were reduced more when setting mean SBP to 120 mm Hg vs mean for NHW participants (CDE β [95% CI]: mean SBP for NHW participants, 2.04 [1.65–2.52], mean SBP = 120 mm Hg, 1.87 [1.46–2.39]). After setting the mediator level to hypertension absent for all participants, Hispanic participants continued to have greater WMHV compared with NHW participants, but the CDE was stable compared with the TE (CDE β [95% CI]: 0.96 [0.86–1.08]). After setting the mediator level to hypertension present for all participants, Hispanic participants continued to have greater WMHV compared with NHW participants and the CDE was higher than the TE (CDE β [95% CI]: 1.66 [1.5–1.85]). After setting the mediator levels to mean SBP for NHW participants and 120 mm Hg, the differences in WMHV between Hispanic and NHW participants remained stable compared with TEs (Figure 3B). Hypertension, but not SBP, was significantly associated with greater WMHV (eTable 3).

Discussion

In this causal mediation analysis, we found that hypertension partially mediated the Black-NHW inequity in WMHV, an established marker of cSVD, consistent with our original hypothesis. Hypertension did not significantly mediate racial or ethnic inequities in amyloid or tau deposition. Although Hispanic participants had a lower AD meta-ROI cortical thickness compared with NHW participants, lowering BP did not significantly mediate this difference.

Secondary analyses of the SPRINT trial have suggested that intensive BP control would reduce WM lesion load over time, but it remains unclear how this would affect racial and ethnic inequities.²⁶ Our study uniquely examined the extent to which hypertension may mediate associations between racial/ethnic differences in neuroimaging markers of brain aging using causal mediation. Our study suggests that hypertension control could reduce racial/ethnic inequities in WMHV, which is one of the most robust predictors of all-cause dementia.⁴⁷ Notably, the magnitude of differences in WMHV between Black and NHW participants increased when we assumed that all participants had hypertension compared with our unmediated estimates. These findings suggest that controlling hypertension for dementia prevention is especially important in Black communities, which is consistent with previous epidemiologic data showing the greater burden of hypertension in Black communities compared with others in the United States.⁴⁸

We found evidence for Hispanic-NHW inequities in AD meta-ROI cortical thickness, but hypertension did not seem to mediate this association. Although CI included the null, the CDE after setting hypertension to absent suggests that controlling for hypertension would lead to thinner cortices contradictory to previous evidence that hypertension is related to gray matter atrophy.^9,26,49 Recent work from Fang et al.⁴⁹ suggests a possible bidirectional relationship between cortical thinning and hypertension; moreover, reverse causation could be affecting our results. Replication in larger cohorts may help clarify this paradoxical finding that did not reach statistical significance in our analysis.

In addition, our study did not provide evidence that hypertension mediates racial/ethnic differences in amyloid or tau deposition. In this and previous analyses in HABS-HD,⁵⁰ amyloid deposition was lower in Black participants compared with NHW participants, and evidence from this causal mediation analysis suggests that mitigating hypertension would not reduce these differences. This is consistent with previous studies that do not show a significant association between hypertension and amyloid deposition,^14,16 implying that hypertension does not contribute causally to amyloid deposition. In previous HABS-HD analyses, greater tau burden was shown in women of color, suggesting that there is effect modification by sex/gender not fully elucidated in this work.⁵⁰ Similarly, racial/ethnic differences in tau deposition do not seem to be strongly mediated by hypertension. Consistent with previous work, hypertension seems to be most causally associated with subclinical cerebrovascular disease, and not neurodegenerative pathology, in this diverse sample. Previous work in animal models suggests that hypertension contributes to blood-brain barrier dysfunction and beta-amyloid clearance.¹³ Therefore, future work may consider how hypertension modifies racial/ethnic differences in neurodegeneration, as vascular risk factors may contribute more to cognitive or brain reserve than as a causal contributor to neurodegenerative pathology.

There are limitations to this study. First, we categorize participants as identifying either as Black or Hispanic, but Hispanic participants represent a racially diverse population,²⁵ and future studies should consider the intersection of race/ethnicity. Furthermore, most Hispanic participants in this study identified as Mexican, representing only 1 subgroup of a diverse Hispanic diaspora. In addition, future work should also use more complete structural measures of anti-Black racism and Hispanic ethnicity. Second, we examined 1 time point of neuroimaging measurement, but examining longitudinal trajectories of brain aging pathology can inform how interventions on hypertension may affect brain aging pathology in the longer term. Third, assumptions for causal mediation include that there are no unmeasured confounders of the relationships between exposure and outcomes or the mediator and outcomes. While we are able to account for many confounders, given the in-depth measurement of medical, social, and psychological factors in HABS-AD, we cannot rule out the possibility of residual confounding. As outlined in the National Institute of Neurological Disorders and Stroke Social Determinants of Health framework,⁵¹ early-life social environment, parental education, and health in young adulthood are commonly unmeasured confounders that may differ substantially by race/ethnicity and thus strengthen the associations seen here. Fourth, the lack of detailed medication data limits our ability to examine how pharmacologic control of BP may mediate these associations. Fifth, results may not be generalizable to other countries, other racial/ethnic groups not represented here, or nonbinary genders. Finally, although we attempt to account for selection into the neuroimaging subsamples, selection bias due to the nonprobability nature of the study design limits our ability to generalize results to the target population (i.e., the Dallas-Fort Worth population, which served as the catchment area for HABS-HD).

Strengths include using data from a highly phenotyped, racially and ethnically heterogeneous cohort and a causal mediation analysis approach that rigorously addresses confounding and selection bias due to differential participation in imaging.

Overall, our findings suggest that treating hypertension in Black older adults may decrease differential burden of cSVD compared with NHW participants. Future work should explore how control of hypertension in racially/ethnically diverse samples may reduce cognitive decline and risk of dementia.

Glossary

AD

Alzheimer disease

ADRD

Alzheimer disease–related dementia

BMI

body mass index

BP

blood pressure

CDE

controlled direct effect

cSVD

cerebral small vessel disease

CVD

cardiovascular disease

DBP

diastolic BP

eGFR

estimated glomerular filtration rate

HABS-HD

Health and Aging Brain Study—Health Disparities

HbA1c

hemoglobin A1c

IPW

inverse probability weight

LDL-C

low-density lipoprotein cholesterol

MTL

medial temporal lobe

ROI

region of interest

SBP

systolic BP

SPRINT

Systolic Blood Pressure Intervention Trial

SUVR

standardized uptake value ratio

TE

total effect

WMHV

white matter hyperintensity volume

Author Contributions

M. Caunca: drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data. S. Gutierrez: drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data. K. Wheeler: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. M.N. Braskie: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. J. Torres: drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data. K. Yaffe: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data.

Study Funding

This publication was supported by the National Institute of Neurological Disorders and Stroke, NIH, through UCSF grant number 5UE5NS070680-15, and the National Institute on Aging of the NIH under Award Numbers R01AG054073, R01AG058533, P41EB015922, U19AG078109, and R35AG071916. Its contents are solely the responsibility of the authors and do not necessarily represent the views of the NIH.

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.