Neuroimaging markers of dementia across race/ethnicity and sex/gender using an intersectional approach within the HABS‐HD cohort

Abstract

INTRODUCTION

Racial/ethnic and sex/gender differences in neuroimaging markers of dementia have been previously explored, but rarely with an intersectional approach.

METHODS

Using data from the Health and Aging Brain Study–Health Disparities cohort, we examined neuroimaging markers of dementia using both interaction between race/ethnicity and sex/gender and effect modification of race/ethnicity by sex/gender.

RESULTS

We analyzed data from 3433 dementia‐free participants with either magnetic resonance imaging or positron emission tomography (PET) data at baseline (mean [standard deviation] age: 65 [9] years, 36% non‐Hispanic White [NHW], 27% Black, 37% Hispanic, and 63% women). Compared to NHW, Black men had lower global amyloid PET standardized uptake value ratio (SUVR; β [95% confidence interval]: –0.32 [–0.53, –0.11]), and Hispanic (0.65 [0.39, 0.91]) and Black women had greater medial temporal lobe tau SUVR (0.49 [0.30, 0.69]).

DISCUSSION

We observed that the distribution of neuroimaging markers of dementia differed across racial/ethnicity groups by sex/gender. An intersectional approach can aid in tailoring research and clinical efforts in preventing and treating dementia.

Highlights

• Hispanic and Black women had greater medial temporal lobe tau deposition, compared to their non‐Hispanic White counterparts.

• Black men had lower global amyloid deposition compared to non‐Hispanic White men.

• Black men and women had higher burden of cerebral small vessel disease compared to their non‐Hispanic White counterparts, with stronger associations in Black men.

1. BACKGROUND

Differences in the risk of Alzheimer's disease (AD) and Alzheimer's disease‐related dementias (ADRD) have been observed across both racial/ethnic ¹ , ² , ³ and sex/gender groups, ⁴ representing an important public health problem as the burden of AD/ADRD continues to rise. ⁵ Prior studies have demonstrated that non‐Hispanic Black and Hispanic older adults are at higher risk for incident AD/ADRD compared to their non‐Hispanic White (NHW) counterparts. ¹ , ² , ³ Studies have also reported sex/gender differences in AD/ADRD risk, and overall, there are mixed results, especially when comparing cross‐nationally. ⁶ Mechanisms underlying these differences in AD/ADRD risk by race/ethnicity and sex/gender are unclear and warrant further investigation.

To better characterize the underlying causes for these racial/ethnic and sex/gender differences in AD/ADRD risk, ongoing research has explored how the pathophysiology of dementia may differ across these identities, including both neurodegenerative pathology and subclinical cerebrovascular disease. ⁷ For example, recent data from the Imaging Dementia–Evidence for Amyloid Scanning (IDEAS) Cohort Study showed that Medicare beneficiaries of color had a lower burden of amyloid positivity compared to their NHW counterparts. ⁸ Further, in large cohort studies, female sex has been related to a greater odds of amyloid positivity in those with mild cognitive impairment (MCI) or dementia ⁸ and among cognitively normal people. ⁹ Similarly, women have been shown to have greater tau deposition compared to men, ¹⁰ but there is a paucity of research on differences in tau deposition across racial/ethnic groups. Finally, prior studies have shown that Black persons in particular have a greater burden of cerebral small vessel disease compared to their NHW counterparts. ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ Understanding how dementia pathology differs across groups may help to uncover mechanisms that explain inequities and drive research into interventions that may consequently reduce these differences.

Efforts to characterize differences across racial/ethnic and sex/gender groups are crucial for shaping tailored public health interventions and guiding precision medicine, but rarely is the intersection of both race/ethnicity and sex/gender considered. Intersectional theory, put forth by Kimberlé Crenshaw as a framework for describing the specific oppression experienced by Black women, reflects how multiple social systems of power may interact synergistically to increase or decrease one's exposure to specific oppressive experiences. ¹⁶ , ¹⁷ In the context of health research, an intersectional approach allows us to examine heterogeneity in disease distribution within our complex social environment and can help guide more tailored approaches from the level of public health policy to precision medicine. ¹⁸ , ¹⁹ Most commonly used in quantitative research are subgroup analyses are performed using interaction and effect modification, ²⁰ and though often used interchangeably, these methods answer slightly different questions. ²¹ In the context of racial/ethnic and sex/gender inequities in ADRD research, we posit that explicitly testing for interaction will answer whether these identities have synergistic impacts on dementia pathology, while effect modification will test whether the association between race/ethnicity and burden of dementia pathology differs across sex/gender.

In this study, we aimed to describe the distribution of neuroimaging markers of dementia across both race/ethnicity and sex/gender, using an approach grounded in intersectional theory. We aimed to quantify how the distribution of neuroimaging markers differed across groups by testing both interaction between race/ethnicity and sex/gender and effect modification of race/ethnicity by sex/gender. We hypothesized that we would detect significant interactions between race/ethnicity and sex/gender, signaling a synergistic effect. Given the compounding health risks of anti‐Black racism and sexism, we expected that Black women would have a greater burden of dementia‐related pathology compared to NHW men.

2. METHODS

2.1. Cohort recruitment and sample selection

This study's analytic sample included 3433 dementia‐free participants from the cohort with either magnetic resonance imaging (MRI) or positron emission tomography (PET) data available at their first visit (baseline). Briefly, 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, USA. Initially, HABS‐HD recruited Hispanic and NHW community participants starting in 2017, and in 2021, recruitment of Black participants began. Using a community‐based participatory research approach, participants were invited to complete a standardized assessment including a clinical interview, neuropsychological assessment, blood draw, and neuroimaging. ²² This study was approved by the North Texas Regional Institutional Review Board, and all participants provided written informed consent. We used data from HABS‐HD release 6.

2.2. Exposures of interest: race/ethnicity and sex/gender

Race/ethnicity and sex/gender were self‐reported at the baseline visit based on standardized interview protocols. Participants reported their race as either NHW or Black and their ethnicity as either non‐Hispanic or Hispanic. Race/ethnicity was constructed as participants who identified as either NHW, non‐Hispanic Black, or Hispanic. Sex/gender was self‐reported as either male or female. Data regarding the full spectrum of non‐binary sex/gender was not collected in the initial data collection for this study.

2.3. Outcomes of interest: neuroimaging markers of dementia

Neuroimaging was obtained at the baseline visit. The markers below were chosen based on a priori knowledge of their relationship to risk of dementia, including AD. ⁷

2.3.1. PET imaging markers of neurodegeneration

2.3.1.1. Global amyloid deposition

Amyloid PET were available for 1584 participants at baseline. For the Siemens Vision 450 whole‐body PET/computed tomography (CT), a four frame by 5 minute dynamic emission acquisition was started 90 minutes post‐injection 8.1 mCi ± 10% of florbetaben (FBB) after a low‐dose CT scan used for attenuation correction. ²² Images are processed by iterative reconstruction (eight iterations and five subsets). On the Siemens MCT 20 scanner, the acquisition was identical except that there was a 20 minute continuous emission. Scans on the MCT 20 were processed by iterative reconstruction with four iterations and 24 subsets. Scans were processed at the University of Southern California Stevens Neuroimaging and Informatics Institute. Using FreeSurfer‐derived regions of interest (ROIs), a global amyloid measure was calculated across frontal, anterior/posterior cingulate, lateral parietal, and lateral temporal cortices. Using the whole cerebellum as a reference, a standardized uptake value ratio (SUVR) was calculated.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature using PubMed, searching all published health manuscripts, presentations, and abstracts on racial/ethnic and gender differences in neuroimaging markers of dementia. Relevant works are cited.

2. Interpretation: Our findings suggest that gender modifies the effect of race/ethnicity on neuroimaging markers of dementia. In particular, Black men had lower global amyloid deposition and Black and Hispanic women exhibited higher medial temporal lobe tau deposition, compared to their non‐Hispanic White counterparts. These data enhance the literature by providing an analysis of neuroimaging markers grounded in intersectional theory.

3. Future directions: Future studies will focus on mediators of the relationships among race/ethnicity, gender, and neuroimaging markers of dementia, as well as cognitive performance and dementia burden. Replication of these findings in other studies is also warranted.

2.3.1.2. Medial temporal lobe tau deposition

Tau PET were available for 914 participants at baseline. Tau deposition was measured using tau PET imaging with¹⁸F‐PI‐2620 (PI‐2620) with the Siemens Vision 450 whole‐body PET/CT scanner. As previously described, ²² a six frame by 5 minute dynamic emission acquisition was performed, starting 45 to 75 minutes post injection of 5 mCi ± 10% ¹⁸F‐PI‐2620 and immediately after a CT attenuation scan, similar to above. Images were reconstructed immediately after the 30 minute emission scan using identical reconstruction parameters to the FBB scans collected on the Vision 450. We used FreeSurfer to derive ROIs in the medial temporal lobe (MTL), posterior cingulate, and lateral parietal cortex. Using the inferior cerebellar gray matter as a reference, SUVR was calculated in those regions. We chose MTL tau as a focus given its clinical importance for AD pathophysiology. ⁷

2.3.2. MRI markers of neurodegeneration

Cortical thickness and hippocampal volume data were available for 2564 and 3108 participants at baseline, respectively. MRIs were obtained using a Siemens Magnetom Skyr whole body scanner or Siemens Magnetom Vida 3T scanners. ²³ The following sequences were obtained: T1‐weighted whole brain volumetric spoiled magnetization‐prepared rapid gradient (MPRAGE), whole brain volumetric fluid attenuated inversion recovery (FLAIR), and susceptibility‐weighted imaging (SWI). Regional cortical thickness was measured using FreeSurfer software 5.3.0. Based on Jack et al., ²⁴ 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. Hippocampal volumes and intracranial volumes were obtained using HippoDeep software. We calculated the total hippocampal volume as the sum of the left and right hippocampal volumes for each participant.

2.3.3. MRI markers of small vessel disease

White matter hyperintensity volume (WMHV) data were available for 3287 participants at baseline, and lacunar infarct and cerebral microbleed data were available in 3378 participants at baseline. WMHV was obtained from FLAIR and T1‐weighted MRIs using the Statistical Parametric Mapping Lesion Segmentation Toolbox as previously described. ²² , ²⁵ Participants with confounding pathology (e.g., possible multiple sclerosis lesions, white matter disease due to traumatic brain injury) were excluded from this analysis. Lacunar infarcts and cerebral microbleeds were detected and counted by a neuroradiologist as previously described. ²⁶

2.4. Covariates of interest

Age, years of education, income, marital status, insurance status, and smoking status were self‐reported and obtained at the baseline visit with standardized protocols as previously described. ²² The type of PET and MRI scanner was also included in models and was obtained at the time of imaging. Systolic and diastolic blood pressure were measured using standardized methods, and an average of two readings was used for analysis. Total cholesterol, hemoglobin A1c (HgbA1c), and apolipoprotein E (APOE) ε4 carrier status was obtained from lab analysis of blood samples collected at baseline. Total intracranial volume was obtained using MRI processing as described above.

2.5. Conceptual framework

To guide our covariate selection, we constructed a directed acyclic graph (DAG) and used Daggity online ²⁷ (code available in Supplemental Methods in supporting information) to determine the minimal set of covariates needed to estimate the total direct effect of both race/ethnicity and sex/gender on neuroimaging markers of dementia. Figure 1 illustrates our conceptual framework (rendered in draw.io for clarity). We conceptualize race/ethnicity as a social construct with the understanding that it does not capture the totality of the multifaceted construct of structural racism. ²⁸ To this end, we also adjusted for other aspects of structural racism in models described below. Though studies use the terms sex and gender interchangeably, we chose the term sex/gender to wholly encompass biological, social, and cultural context. ⁶ We chose covariates based on prior literature on the most important confounders and downstream mediators of the association of interest.

FIGURE 1.

A representation of our conceptual framework. To estimate the total effect of race/ethnicity and sex/gender on neuroimaging markers of dementia, adjustment by covariates other than age are actually not needed. However, estimating the direct effect of race/ethnicity and sex/gender on neuroimaging markers requires adjustment of covariates above. APOE, apolipoprotein E; BP, blood pressure; DM, diabetes mellitus; HLD, hyperlipidemia

2.6. Statistical analysis

For our descriptive analysis, distributions of covariates were examined in the sample stratified by race/ethnicity and sex/gender. We describe these differences and did not calculate inferential statistics, consistent with Strengthening the Reporting of Observational Studies in Epidemiology guidelines. ²⁹ We also plotted age‐adjusted means of our outcomes of interest and computed P values for differences across race/ethnicity using pairwise t tests for continuous variables and chi‐squared tests for categorical variables.

For our main analysis, we estimated associations among race/ethnicity, sex/gender, and neuroimaging biomarkers. For continuous outcomes, we fit multivariable linear regression models to calculate beta estimates and 95% confidence intervals (CIs). For binary outcomes, we fit multivariable logistic regression to calculate odds ratios (ORs) and 95% CIs. For global amyloid PET SUVR, medial temporal lobe tau PET SUVR, AD meta‐ROI cortical thickness, and hippocampal volume, each was z scored to facilitate comparability across estimates; thus, beta estimates represent the standard deviation (SD) change in the outcome. WMHV was log‐transformed due to a highly right‐skewed distribution; beta estimates were then back‐transformed and thus represent percent change in WMHV. We plotted residuals on Q–Q plots to check the normality assumption for linear models, which was met for all continuous variables. For all analyses, a two‐sided P value < 0.05 was considered statistically significant.

We performed interaction analyses by adding race/ethnicity x sex/gender interaction terms in the models described above. We performed effect modification analyses by stratifying the sample by sex/gender and re‐fitting models for men and women separately. Because we explicitly tested for both interaction and effect modification a priori, we did not use P values from interaction terms between race/ethnicity and sex/gender to aid in our decision to perform stratified analyses. We highlight estimates with P value < 0.05 and present main effects, interaction terms, and estimates from stratified analyses. ³⁰ , ³¹ For continuous outcomes, we chose to focus on the additive interaction (represented by the race/ethnicity x sex/gender term in the model) as the distribution of the data fits a linear model best (vs. a log‐linear model in which multiplicative interactions could theoretically be tested), and the utility of additive versus multiplicative interactions is not as clear for continuous outcomes as for binary. ¹⁸ , ³¹ For the binary outcomes, we report the relative excess risk due to interaction calculated via odds ratios (RERI_OR) to assess possible additive interactions between race/ethnicity and sex/gender, and 95% CIs for the RERI_OR were estimated via the delta method. ³¹ , ³² For binary outcomes, estimates for multiplicative interaction terms (represented by the race/ethnicity x sex/gender term in the model) are also presented with 95% CIs.

We fit two models to estimate associations between race/ethnicity and outcomes of interest, informed by our DAG (Figure 1). Model 1 includes age, years of education, marital status, income, insurance status, and PET/MRI scanner (except for tau deposition, for which all participants were scanned on the same machine). Model 2 includes covariates from Model 1 and additionally adjusts for systolic blood pressure, diastolic blood pressure, total cholesterol, HgbA1c, smoking status, and APOE ε4 allele status. Models for hippocampal volume and WMHV additionally adjust for total intracranial volume to account for differences in head size that may impact measurement of these outcomes and capture early‐life exposures that may impact brain development. ³³

Sensitivity analyses were performed to test the robustness of our findings. To address possible selection bias into the imaging subsamples, ³⁴ we first compared covariate distributions across subgroups of participants able to obtain PET versus MR imaging. Then, given that a significantly smaller group of participants obtained PET imaging, we re‐analyzed the analyses of amyloid and tau PET using stabilized inverse probability of selection weights to better understand the impact of selection bias on our estimates ³⁵ (methods in Supplemental Methods). To understand how those with prevalent MCI may have impacted our estimates, we repeated our fully adjusted analyses excluding those with MCI. In a post hoc analysis, we also conceptualized intersectional groups as contextual variables using intersectional multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) to examine whether multilevel models would fit the data better than linear regression analyses as described above (Supplemental Methods). ³⁶

Analyses were performed in R version 4.4.2. HABS‐HD data are publicly available at https://apps.unthsc.edu/itr/habs‐hd. Code for this analysis is available on Github (https://github.com/michelle‐caunca/HABSRaceEthGenImg.git).

3. RESULTS

3.1. Descriptive statistics

Descriptive statistics stratified by race/ethnicity and sex/gender for the analytic sample are outlined in Table 1 (N = 3433). In the overall analytic sample, participants identified as 36% NHW, 27% Black, 37% Hispanic, and 63% women with a mean (SD) age of 65 (9) years. Across race/ethnicity, patterns were similar between men and women for age, marital status, years of education, income, HgbA1c, smoking status, diastolic blood pressure, and raw values of neuroimaging markers of dementia (Table 1). Hispanic participants had the highest average systolic blood pressure compared to their NHW and Black counterparts, with a greater difference observed for men than for women. Among men, Hispanic men had the highest average total cholesterol compared to NHW and Black men. Among women, however, NHW women had the highest average total cholesterol compared to Black and Hispanic women (Table 1). There were significant differences in the distribution of neuroimaging markers of dementia between the racial/ethnic groups, with similar distributions in men and women (Figure 2). Distributions of exposures, covariates, and outcomes were largely similar across subsamples stratified by availability of specific neuroimaging variables (Table S1 in supporting information).

TABLE 1.

Descriptive statistics by race/ethnicity and sex/gender for analytic sample (N = 3433).

	Non‐Hispanic White (n = 1228)	Black (n = 942)	Hispanic (n = 1263)
Men (n = 1261)	n = 512	n = 339	n = 410
Age (years, mean, SD)	69 (9)	63 (7)	63 (8)
Married (n, %)	370 (73)	188 (56)	299 (73)
Years of education (mean, SD)	16 (3)	14 (3)	11 (5)
Income (dollars, mean, SD)	99,642 (93,772)	87,830 (112,780)	52,754 (58,785)
Has no insurance (n, %)	18 (4)	34 (10)	91 (22)
Systolic blood pressure (mmHg, mean, SD)	135 (16)	139 (18)	143 (19)
Diastolic blood pressure (mmHg, mean, SD)	82 (10)	88 (11)	87 (11)
Total cholesterol (mg/dL, mean, SD)	167 (40)	163 (37)	175 (40)
HgbA1c (%, mean, SD)	5.7 (0.9)	6.1 (1.3)	6.4 (1.7)
Current smoker (n, %)	26 (5)	75 (22)	42 (10)
APOE ε4 allele positive (n, %)	139 (29)	106 (40)	63 (15)
Missing APOE data	37 (7)	30 (7)	30 (7)
Global amyloid PET SUVR (median, Q1, Q3)	1.0 (0.9, 1.1)	1.0 (0.9, 1.0)	1.0 (0.9, 1.0)
Medial temporal lobe tau PET SUVR (median, Q1, Q3)	1.1 (1.0, 1.2)	1.1 (1.0, 1.2)	1.2 (1.1, 1.2)
AD meta‐ROI cortical thickness (mm, mean, SD)	2.7 (0.1)	2.8 (0.1)	2.7 (0.1)
Hippocampal volume (mm3, mean, SD)	6593 (877)	6523 (745)	6684 (777)
White matter hyperintensity volume (cm3, median, Q1, Q3)	1.7 (0.5, 4.6)	1.6 (0.5, 4.6)	0.9 (0.2, 2.5)
Intracranial volume (mm3, mean, SD)	1,583,629 (109,524)	1,528,918 (110,653)	1,503,280 (106,853)
Lacunar infarcts present (n, %)	3 (1)	18 (5)	4 (1)
Cerebral microbleeds present (n, %)	23 (5)	49 (15)	7 (2)

Women (n = 2172)	n = 716	n = 603	n = 853
Age (years, mean, SD)	68 (9)	63 (8)	62 (8)
Married (n, %)	384 (54)	231 (38)	474 (56)
Years of education (mean, SD)	15 (2)	15 (3)	10 (4)
Income (dollars, mean, SD)	87,269 (89,856)	78,397 (87,456)	41,789 (58,632)
Has no insurance (n, %)	28 (4)	27 (5)	217 (26)
Systolic blood pressure (mmHg, mean, SD)	130 (18)	134 (17)	135 (20)
Diastolic blood pressure (mmHg, mean, SD)	80 (10)	86 (11)	82 (10)
Total cholesterol (mg/dL, mean, SD)	195 (38)	184 (39)	189 (41)
HgbA1c (%, mean, SD)	5.5 (0.6)	6.0 (1.0)	6.4 (1.6)
Current smoker (n, %)	17 (2)	49 (8)	50 (6)
APOE ε4 allele positive (n, %)	184 (26)	197 (33)	150 (18)
Missing APOE data	50 (7)	131 (22)	83 (10)
Global amyloid PET SUVR (median, Q1, Q3)	1.0 (1.0, 1.0)	1.0 (1.0, 1.0)	1.0 (1.0, 1.0)
Medial temporal lobe tau PET SUVR (median, Q1, Q3)	1.1 (1.0, 1.2)	1.1 (1.1, 1.2)	1.1 (1.1, 1.2)
AD meta‐ROI cortical thickness (mm, mean, SD)	2.8 (0.1)	2.8 (0.1)	2.8 (0.1)
Hippocampal volume (mm3, mean, SD)	6332 (736)	6232 (664)	6386 (689)
White matter hyperintensity volume (cm3, median, Q1, Q3)	1.0 (0.3, 3.5)	0.8 (0.2, 2.3)	0.5 (0.1, 1.9)
Intracranial volume (mm3, mean, SD)	1,369,615 (96,175)	1,359,129 (95,125)	1,319,143 (94,162)
Lacunar infarcts present (n, %)	7 (1)	19 (3)	4 (1)
Cerebral microbleeds present (n, %)	19 (3)	61 (10)	19 (2)

Abbreviations: AD, Alzheimer's disease; APOE, apolipoprotein E; HgbA1c, hemoglobin A1c; MRI, magnetic resonance imaging; ns, not significant; PET, positron emission tomography; ROI, region of interest; SD, standard deviation; SUVR, standardized uptake value ratio; WMHV, white matter hyperintensity volume.

FIGURE 2.

Distribution of neuroimaging markers of dementia, stratified by race/ethnicity and sex/gender. Age‐adjusted means presented for continuous variables, error bars represent ± standard deviation. P values from pairwise t tests for continuous variables and chi‐squared tests for categorical variables (*** indicates P < 0.05). A, Global amyloid PET SUVR. B, Medial temporal lobe tau PET SUVR. C, AD meta‐ROI cortical thickness. D, Hippocampal volume. E, Log‐transformed White matter hyperintensity volume. F, Lacunar infarcts. G, Microbleeds. AD, Alzheimer's disease; MRI, magnetic resonance imaging; ns, not significant; PET, positron emission tomography; ROI, region of interest; SUVR, standardized uptake value ratio; WMHV, white matter hyperintensity volume

3.2. PET imaging markers of neurodegeneration

3.2.1. Global amyloid deposition

Black participants had lower global amyloid PET SUVR compared to NHW participants (β [95% CI]: –0.32 [–0.51, –0.12], P < 0.05). Global amyloid PET SUVR for Hispanics, women, and the interaction between race/ethnicity x sex/gender was not significantly different from NHW participants, men, and NHW men, respectively (Table 2). In sex/gender‐stratified analyses, Black men had lower global amyloid PET SUVR (β [95% CI]: –0.32 [–0.53, –0.11], P < 0.05) compared to NHW men, which was similar after further adjustment for cardiometabolic measures and APOE ε4 allele status (Figure 3A). Global amyloid PET SUVR was significantly lower in Black women compared to their NHW counterparts after adjustment for cardiometabolic measures and APOE ε4 allele status (β [95% CI]: –0.21 [–0.39, –0.03], P < 0.05). Global amyloid PET SUVR for Hispanic men and women were not significantly different from their NHW counterparts (Figure 3A).

TABLE 2.

Associations Between Race/Ethnicity, Sex/Gender, and Neuroimaging Markers of Neurodegeneration.

	Global amyloid‐PET SUVR (N = 1584)			MTL tau‐PET SUVR (N = 914)			AD meta‐ROI cortical thickness (N = 2564)			Hippocampal volume (N = 3108)
	Beta	LCL	UCL	Beta	LCL	UCL	Beta	LCL	UCL	Beta	LCL	UCL
Model 1
Black	−0.32	−0.51	−0.12	0.24	0.01	0.47	−0.30	−0.60	−0.01	−0.26	−0.40	−0.12
Hispanic	−0.23	−0.47	0.01	0.63	0.33	0.94	−0.04	−0.43	0.36	0.08	−0.05	0.21
Women	0.04	−0.16	0.24	0.02	−0.22	0.26	0.20	−0.09	0.50	0.28	0.16	0.39
Black*Women	0.20	−0.05	0.44	0.24	−0.05	0.54	0.01	−0.36	0.38	−0.13	−0.29	0.03
Hispanic*Women	0.10	−0.19	0.38	−0.03	−0.40	0.34	0.01	−0.45	0.47	−0.11	−0.25	0.03
Model 2
Black	−0.31	−0.53	−0.09	0.36	0.10	0.62	−0.27	−0.60	0.06	−0.24	−0.39	−0.08
Hispanic	−0.10	−0.37	0.16	0.55	0.22	0.88	−0.06	−0.48	0.37	0.08	−0.06	0.21
Women	0.13	−0.10	0.35	0.04	−0.23	0.31	−0.05	−0.38	0.29	0.23	0.11	0.36
Black*Women	0.10	−0.17	0.36	0.08	−0.25	0.40	0.09	−0.33	0.50	−0.17	−0.34	0.00
Hispanic*Women	−0.06	−0.37	0.26	0.04	−0.36	0.45	0.11	−0.40	0.62	−0.11	−0.26	0.03

Notes: Beta estimates with 95% confidence intervals (LCL = lower confidence limit, UCL = upper confidence limit) from multivariable linear regression models are presented. Reference group for race/ethnicity is NHW and for sex/gender is men. Outcomes modeled as z‐scores to facilitate comparability. Highlighted estimates indicate P < 0.05. Model 1: age, years of education, marital status, income, insurance status, PET/MRI scanner (and intracranial volume for hippocampal volume). Model 2: Model 1 + systolic blood pressure, diastolic blood pressure, total cholesterol, HgbA1c, smoking status, and APOE ε4 allele status.

Abbreviations: APOE, apolipoprotein E;

FIGURE 3.

Associations between race/ethnicity and neuroimaging markers of neurodegeneration, stratified by sex/gender. Beta estimates with 95% confidence intervals from multivariable linear regression models are presented. Outcomes modeled as z scores to facilitate comparability. Model 1: age, years of education, marital status, income, insurance status, PET/MRI scanner (and intracranial volume for hippocampal volume). Model 2: Model 1 + systolic blood pressure, diastolic blood pressure, total cholesterol, HgbA1c, smoking status, and APOE ε4 allele status. A, Global amyloid PET SUVR. B, Medial temporal lobe tau PET SUVR. C, AD meta‐ROI cortical thickness. D, Hippocampal volume. AD, Alzheimer's disease; APOE, apolipoprotein E; HgbA1c, hemoglobin A1c; LCL, lower confidence limit; MRI, magnetic resonance imaging; PET, positron emission tomography; ROI, region of interest; SUVR, standardized uptake value ratio; UCL, upper confidence limit

3.2.2. MTL tau deposition

Compared to NHW participants, both Hispanic and Black participants had higher MTL tau PET SUVR, with stronger associations in Hispanics (β [95% CI]: Hispanic 0.63 [0.33, 0.94], P < 0.05; Black 0.24 [0.01, 0.47], P < 0.05). MTL tau PET SUVR for women and race/ethnicity x sex/gender interactions terms were not significantly different from men and NHW men, respectively (Table 2). In sex/gender‐stratified analyses, compared to NHW men, Hispanic men had higher MTL tau PET SUVR (β [95% CI]: 0.55 [0.21, 0.89], P < 0.05). Black men also had higher MTL tau PET SUVR compared to NHW men after adjustment for cardiometabolic measures and APOE ε4 allele status (β [95% CI]: 0.36 [0.07, 0.64], P < 0.05). Compared to NHW women, Hispanic and Black women had higher MTL tau PET SUVR (β [95% CI]: Hispanic women 0.65 [0.39, 0.91], P < 0.05; Black women 0.49 [0.30, 0.69], P < 0.05), which remained stable after adjustment for cardiometabolic measures and APOE ε4 allele status (Figure 3B).

3.3. MRI markers of neurodegeneration

3.3.1. AD meta‐ROI cortical thickness

Compared to NHW participants, Black participants had lower AD meta‐ROI cortical thickness (β [95% CI]: –0.30 [–0.60, –0.01], P < 0.05), though the association was slightly attenuated and did not reach statistical significance after adjustment for cardiometabolic measures and APOE ε4 (β [95% CI]: –0.27 [–0.60, 0.06], P > 0.05). AD meta‐ROI cortical thickness for Hispanic participants, women, and interaction terms between race/ethnicity and sex/gender were not significantly different from NHW participants, men, and NHW men, respectively (Table 2). In sex/gender‐stratified analyses, Black women had lower AD meta‐ROI cortical thickness compared to NHW women (β [95% CI]: –0.28 [–0.43, –0.14], P < 0.05), which remained stable after adjustment for cardiometabolic measures and APOE ε4 (Figure 3C). AD meta‐ROI cortical thickness for Black and Hispanic men and Hispanic women were not significantly different from their NHW counterparts (Figure 3C).

3.3.2. Hippocampal volume

Compared to NHW participants, Black participants had lower hippocampal volume (β [95% CI]: –0.26 [–0.40, –0.12], P < 0.05). Compared to men, women had greater hippocampal volume (β [95% CI]: 0.28 [0.16, 0.39], P < 0.05). These associations remained stable after adjustment for cardiometabolic measures and APOE ε4 allele status (Table 2). Hippocampal volume for Hispanic participants and interaction terms between race/ethnicity and sex/gender were not significantly different from NHW participants or NHW men, respectively (Table 2). In sex/gender‐stratified analyses, Black men and women had lower hippocampal volumes compared to their NHW counterparts (β [95% CI]: Black men –0.34 [–0.50, –0.18], P < 0.05; Black women –0.38 [–0.50, –0.26], P < 0.05), which remained stable after adjustment for cardiometabolic measures and APOE ε4 allele status (Figure 3D). Hippocampal volumes for Hispanic men and women were not significantly different from their NHW counterparts (Figure 3D).

3.4. MRI markers of small vessel disease

3.4.1. WMHV

Compared to NHW participants, Black participants had higher WMHV (β [95% CI]: 1.47 [1.29, 1.68], P < 0.05), which was slightly attenuated after adjustment for cardiometabolic measures and APOE ε4 allele status (β [95% CI]: 1.33 [1.15, 1.54], P < 0.05). WMHV for Hispanic participants, women, and interaction terms between race/ethnicity and sex/gender were not statistically significant from NHW participants, men, or NHW men, respectively (Table 3). In sex/gender‐stratified analyses, Black men had higher WMHV compared to NHW men (β [95% CI]: 1.49 [1.26, 1.77], P < 0.05), which was slightly attenuated after adjustment for cardiometabolic measures and APOE ε4 allele status (β [95% CI]: 1.35 [1.13, 1.62], P < 0.05). Black women had higher WMHV compared to NHW women (β [95% CI]: 1.18 [1.06, 1.31], P < 0.05), which remained stable after adjustment for cardiometabolic measures and APOE ε4 allele status (β [95% CI]: 1.14 [1.01, 1.29], P < 0.05). Similar patterns were observed for Hispanic men and women compared to their NHW counterparts, but did not reach statistical significance (Figure 4A).

TABLE 3.

Associations Between Race/Ethnicity, Sex/Gender, and MRI Measures of Cerebral Small Vessel Disease.

	White matter hyperintensity volume (N = 3287)			Lacunar infarcts (N = 3379)			Cerebral microbleeds (N = 3379)
	Beta	LCL	UCL	OR	LCL	UCL	OR	LCL	UCL
Model 1
Black	1.47	1.29	1.68	10.15	3.26	44.78	5.36	3.06	9.69
Hispanic	0.91	0.81	1.03	0.68	0.10	4.27	0.60	0.22	1.45
Women	0.99	0.88	1.11	1.56	0.43	7.31	0.65	0.33	1.26
Black*Women	0.81	0.69	0.94	0.37	0.07	1.58	0.97	0.44	2.14
Hispanic*Women	1.02	0.89	1.17	0.30	0.03	2.51	1.98	0.68	6.24
Model 2
Black	1.33	1.15	1.54	11.96	3.10	79.09	7.69	3.77	16.80
Hispanic	0.88	0.78	1.00	0.93	0.12	8.49	1.22	0.42	3.34
Women	1.05	0.93	1.19	2.18	0.48	15.33	0.77	0.32	1.86
Black*Women	0.85	0.72	1.01	0.22	0.03	1.22	0.79	0.29	2.15
Hispanic*Women	1.00	0.87	1.16	0.24	0.02	2.27	1.61	0.47	5.82

Note: Beta estimates or odds ratios (OR) with 95% confidence intervals (LCL = lower confidence limit, UCL = upper confidence limit) are presented. WMHV estimates transformed to represent percent change in WMHV. Highlighted estimates indicate P < 0.05. Model 1: age, years of education, marital status, income, and insurance status, (and MRI scanner and intracranial volume for WMHV). Model 2: Model 1 + systolic blood pressure, diastolic blood pressure, total cholesterol, HgbA1c, smoking status, and APOE ε4 allele status.

Abbreviations: APOE, apolipoprotein E.

FIGURE 4.

Associations between race/ethnicity and MRI measures of small vessel disease, stratified by sex/gender. Beta estimates or odds ratios with 95% confidence intervals (LCL = lower confidence limit, UCL = upper confidence limit) are presented. WMHV estimates transformed to represent percent change in WMHV. Highlighted estimates indicate P < 0.05. Model 1: age, years of education, marital status, income, and insurance status (and MRI scanner and intracranial volume for WMHV). Model 2: Model 1 + systolic blood pressure, diastolic blood pressure, total cholesterol, HgbA1c, smoking status, and APOE ε4 allele status. A, White matter hyperintensity volume. B, Lacunar infarcts. C, Cerebral microbleeds. APOE, apolipoprotein E; HgbA1c, hemoglobin A1c MRI, magnetic resonance imaging; WMHV, white matter hyperintensity volume

3.4.2. Lacunar infarcts

Black participants had higher odds of lacunar infarcts compared to NHW participants (OR [95% CI]: 10.15 [3.26, 44.78], P < 0.05), which was stable after adjustment for cardiometabolic measures and APOE ε4 allele status (Table 3). Odds of lacunar infarcts for Hispanic participants, women, and multiplicative interaction terms between race/ethnicity and sex/gender were not significantly different from NHW participants, men, or NHW men, respectively (Table 3). RERI_OR suggested a possible negative additive interaction for both Black and Hispanic participants with sex/gender, but was not statistically significant (RERI_OR [95% CI], Black x Women: –7.35 [–21.52, 6.82], P = 0.85, Hispanic x Women: –1.62 [–5.93, 2.68], P = 0.77). In sex/gender‐stratified analyses, Black men had higher odds of lacunar infarcts compared to NHW men (OR [95% CI]: 16.30 [4.79, 76.58], P < 0.05). Of note, estimates were large and unstable due to small cell sizes, and stable after adjustment for cardiometabolic measures and APOE ε4 allele status (Figure 4B). Estimates for Hispanic men and women were not statistically significant (Figure 4B). Black women had higher odds of lacunar infarcts compared to NHW women (OR [95% CI]: 3.00 [1.23, 8.10], P < 0.05), which was attenuated and no longer statistically significant after adjustment for cardiometabolic measures and APOE ε4 allele status (OR [95% CI]: 2.14 [0.71, 6.92], P > 0.05). Hispanic women had lower odds of lacunar infarcts compared to NHW women (OR [95% CI]: 0.13 [0.02, 0.65], P < 0.05), and this association was no longer statistically significant after adjustment for cardiometabolic measures and APOE ε4 allele status (OR [95% CI]: 0.02 [0.03, 1.15]).

3.4.3. Cerebral microbleeds

Compared to NHW participants, Black participants had higher odds of having cerebral microbleeds (OR [95% CI]: 5.36 [3.06, 9.69], P < 0.05), which was strengthened after adjustment for cardiometabolic measures and APOE ε4 allele status (OR [95% CI]: 7.69 [3.77, 16.80], P < 0.05). Odds for Hispanic participants, women, and race/ethnicity x sex/gender interaction terms were not significantly different from NHW participants, men, and NHW men, respectively (Table 3). RERI_OR suggested a possible negative additive interaction for Black identity and sex/gender, and a possible positive additive interaction for Hispanic identity with sex/gender, but these estimates were not statistically significant (RERI_OR [95% CI], Black x Women: –2.77 [–6.90, 1.36], P = 0.91, Hispanic x Women: 0.52 [–0.81, 1.85], P = 0.22). In sex/gender‐stratified analyses, Black participants had higher odds of microbleeds compared to NHW participants (OR [95% CI]: Black men 5.45 [3.01, 10.22], P < 0.05; Black women 5.38 [3.07, 9.88], P < 0.05). Associations for Black men were strengthened after adjustment for cardiometabolic measures and APOE ε4 allele status (OR [95% CI]: 8.98 [4.19, 20.60], P < 0.05), but for Black women, estimates were stable after adjustment (Figure 4C). Associations for Hispanic men and women were not statistically different from their NHW counterparts (Figure 4C).

3.5. Supplemental analyses

Estimates from fully adjusted models were similar when re‐weighted for selection into the amyloid PET subsample (Table S2 in supporting information). After excluding participants with MCI, associations with global amyloid PET were attenuated, while associations with MTL tau PET, AD meta‐ROI cortical thickness, hippocampal volume, WMHV, and cerebral microbleeds were strengthened, though estimates were largely no longer statistically significant for the overall sample across all neuroimaging markers (Table S3 in supporting information). These patterns were largely similar in sex/gender‐stratified analyses (Table S3). Due to small cell sizes, logistic regression models in the cognitively normal subsample did not converge. In MAIHDA models, intersectional stratum explained the variance in only WMHV and hippocampal volume (Table S4A in supporting information, stratum‐specific variance_WMHV = 0.002, stratum‐specific variance_Hipp = 0.002). Estimated means of logWMHV and hippocampal volume were similar to results from our linear regression models (Table S4B).

4. DISCUSSION

In this cross‐sectional analysis of a diverse, community‐based, dementia‐free cohort, we did not observe statistically significant interactions between race/ethnicity and sex/gender, but sex/gender‐stratified analyses suggested effect modification for race/ethnicity. In particular, Black participants had significantly lower amyloid deposition compared to NHW participants, with strongest associations among Black men. Hispanic participants had higher MTL tau deposition compared to NHW participants, with the strongest estimates found for Hispanic women. Adjustment for cardiometabolic measures and APOE ε4 allele status did not significantly affect associations with neuroimaging markers of neurodegeneration, but did impact some estimates for markers of cerebral small vessel disease. Overall, our study reflects how an intersectional approach to biomarker research can provide insight into the possible heterogeneity of different pathologies causing ADRD. Amyloid pathology may not be as prevalent in communities of color, but tau deposition and cerebral small vessel disease may be more potent targets for intervention.

We showed that Hispanic and Black women had greater MTL tau deposition compared to their NHW counterparts, which fills a gap in the literature as there is a paucity of studies that have explored racial/ethnic or joint racial/ethnic and sex/gender differences in tau PET imaging. Similar to our study, Buckley et al. found that, across two convenience‐sampled cohorts of cognitively normal participants, women had greater entorhinal cortex tau deposition compared to men, ¹⁰ and a study in the Alzheimer's Disease Neuroimaging Initiative cohort by Yan et al. showed that female APOE ε4 heterozygotes had greater tau deposition compared to male heterozygotes. ³⁷ In terms of amyloid PET imaging, our findings that Black participants have a lower amyloid burden are consistent with some studies, including the IDEAS Cohort Study in Medicare beneficiaries, which showed that people of color had a lower burden of amyloid positivity than NHW beneficiaries. ⁸ However, our results are inconsistent with other studies, ³⁸ , ³⁹ including in a narrative review positing that there is no significant racial/ethnic variability in amyloid PET imaging after accounting for cognitive status and cardiovascular risk. ⁴⁰ In large cohort studies, female sex has been related to a greater odds of amyloid positivity in those with MCI or dementia ⁸ and among cognitively normal people. ⁹ Mechanisms driving these differences have yet to be elucidated, but we posit that greater tau deposition in Hispanic and Black women may signal their dual vulnerabilities to structural racism, structural sexism, and possible biological mechanisms regarding female sex including menopause‐related hormone changes, inflammation, and stress responses. ⁴¹

Similar to PET studies, there is a paucity of data on the distribution of MRI markers for dementia at the intersection of race/ethnicity and sex/gender. Black participants in our study had lower hippocampal volumes compared to NHW participants, consistent with prior cohort studies ³⁹ and inconsistent with others. ⁴² Consistent with prior work, we found that Black participants had lower cortical thickness measures. ¹¹ In the UK Biobank, sex differences in hippocampal volume were not evident after correction for total brain volume, ⁴³ which has also been illustrated in a meta‐analysis. ⁴⁴ In contrast to our study, other community‐based cohorts have found that women have greater cortical thickness measures than men. ⁴³ , ⁴⁵ The mechanisms underlying these differences are more difficult to elucidate, because these markers of gray matter atrophy are less specific markers of AD‐related neurodegeneration compared to PET imaging markers. Though differences in hippocampal volume can be explained by head size, intracranial volume has been hypothesized as a marker of early life socioeconomic status and thus early life exposures should be considered in future studies. ⁴⁶ Also, these MRI markers can indicate the degree of brain reserve, which is an important but less considered factor in the diagnosis of ADRD, ⁴⁷ especially with regard to sex/gender differences. ⁴⁸

As observed in several prior studies, we also found that Black participants have a greater burden of WMHV ¹¹ , ¹² and lacunar infarcts, ¹³ , ⁴⁹ compared to NHW participants. ¹⁴ , ¹⁵ Some differences have also been shown in Hispanic cohorts, ⁴⁹ though there is a paucity of literature in this population. Women have greater WMHV than men, ⁵⁰ , ⁵¹ including in a large individual‐level meta‐analysis across multiple cohort studies. ⁵² In some community‐based cohorts, men had a greater propensity for lobar microbleeds compared to women, ⁵³ , ⁵⁴ though others did not find any differences by sex/gender. ⁵⁵ Evidence from our study emphasizes the differential burden that Black participants have for small vessel disease. Additionally, possible mediation by cardiometabolic factors for Black women in particular is a target of future study. These findings further support primary prevention efforts in communities of color to reduce the burden of cardiometabolic disease to prevent dementia and reduce racial/ethnic inequities.

Methodologically, we assessed both interaction and effect modification. ²⁰ , ²¹ , ³⁰ , ³¹ Overall, our study more robustly demonstrated effect modification of the association between race/ethnicity and neuroimaging markers of dementia by sex/gender, and did not show strong evidence for interactions, which was inconsistent with our original hypothesis. Effect modification can be observed without interaction in cases in which the effect modifier of interest may act as a proxy for another variable that has a causal effect on our outcomes of interest, but no causal relation to our exposure of interest. ²¹ Further, interactions often occur when both risk factors have a causal link to one another, for example, diet and exercise for cardiovascular disease risk reduction. ²¹ Therefore, our study suggests that although certainly race/ethnicity and sex/gender are related in important ways, they may not act synergistically to impact dementia pathology. Observing more robust effect modification indicates that there are distinct mechanisms through which race/ethnicity and sex/gender separately impact dementia pathology. Differences by race/ethnicity, though not a complete proxy for structural racism, certainly imply that there are contextual risk factors underlying these differences. However, the impact of structural racism does affect many downstream mediators, such as cardiometabolic risk factors and epigenetic mechanisms, that should be explored in further research. Sex/gender represents social, cultural, and biological constructs, and further research can parse out these different effects. Additionally, our post hoc analysis examining these intersectional groups as contextual risk factors was significant for hippocampal volume and WMHV, suggesting the contextual effect of race/ethnicity and sex/gender is most salient for these neuroimaging markers of dementia. Future work should examine this in more depth, especially other contextual risk factors of racism and sexism. Overall, our results support the notion that intervening on downstream targets of racism could have differential impacts on different sex/genders in the context of developing dementia‐related pathology.

This study has several limitations. First, though conducted in a racially/ethnically diverse cohort, results may not be nationally representative as this was a single‐site study. We do not have the ability to examine distributions across Asian, Pacific Islander, and American Indian or Alaskan Native participants. Additionally, the sample imbalance does not reflect a nationally representative sample and thus this may introduce bias and limit generalizability of this work. Second, definitions of sex/gender were binary and efforts in the future to include transgender and non‐binary identities are absolutely crucial to improving intersectional work in dementia research. Replication in other cohorts with similar phenotyping and greater diversity of race/ethnicity and sex/gender will be an important next step. Third, novel neuroimaging markers that take intersectionality into account would enhance precision medicine and should be explored in future studies. Last, though we fit multivariable models, we do not formally examine mediation, and therefore mechanisms on a population level are not examined closely in this study. ¹⁹ This is an area of future research for our group, especially examining causal mediation effects of cardiometabolic risk factors between race/ethnicity and neuroimaging markers of dementia. Strengths of this study include the availability of a highly phenotyped, diverse, community‐based sample and the use of an intersectional approach.

Overall, we found that the distribution of neuroimaging markers of dementia differs jointly by race/ethnicity and sex/gender, supporting the need for an intersectional approach to dementia neuroimaging research. Future studies should replicate our findings in larger cohorts and examine mediating factors to identify targets for intervention, encourage more diverse enrollment in clinical trials, and further tailor clinical practice.

CONFLICT OF INTEREST STATEMENT

The authors do not have any conflicts of interest to report. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All human subjects provided informed consent.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Caunca MR, Bahorik A, Jiang X, Braskie MN, O'Bryant S, Yaffe K. Neuroimaging markers of dementia across race/ethnicity and sex/gender using an intersectional approach within the HABS‐HD cohort. Alzheimer's Dement. 2025;21:e70733. 10.1002/alz.70733