Projected Cognitive and Brain Aging Benefits of Eliminating Cardiometabolic Risks in Non-Hispanic White and Black Males – HABS-HD

Abstract

Background:

Cardiometabolic risk factors contribute to cognitive decline and cerebrovascular pathology and are more prevalent among non-Hispanic Black (NHB) adults than non-Hispanic White (NHW) adults, with the greatest burden observed in males.

Methods:

We analyzed 974 male participants (581 non-Hispanic White [NHW], 393 NHB) from the Healthy Aging Brain Study – Health Disparities baseline visit. Multivariable linear regression models were used to examine associations between cardiometabolic risk factors, including hypertension, diabetes, dyslipidemia, obesity, and tobacco dependence, and outcomes of cognitive domain performance (memory, executive function, processing speed, and language) and white matter hyperintensity burden, adjusting for age, education, apolipoprotein ε4 status, and race. Population intervention models (PIM), a counterfactual regression-based approach, were applied to estimate projected changes in cognition and WMH burden under hypothetical scenarios in which individual cardiometabolic risk factors were absent, with analyses stratified by racial ethnicity and mutually adjusted for the other cardiometabolic risks.

Results:

Diabetes was associated with lower memory (β = −0.14, 95% CI: −0.27 to −0.01) and language (β = −0.15, 95% CI: −0.29 to −0.02). Tobacco dependence was linked to poorer performance across all domains (β range = −0.20 to −0.29). Hypertension was associated with greater WMH volume (β = 0.61, 95% CI: 0.09 to 1.12). PIM analyses projected memory gains from eliminating diabetes of 0.027 (95% CI: 0.002–0.052) in NHW and 0.039 (95% CI: 0.004–0.075) in NHB males, and gains from eliminating tobacco dependence of 0.015 (95% CI: 0.004–0.027) and 0.056 (95% CI: 0.017–0.098), respectively. Removing hypertension was projected to reduce WMH by −0.394 (95% CI: −0.769 to −0.014) in NHW and −0.481 (95% CI: −0.940 to −0.018) in NHB participants.

Introduction

Racial ethnic disparities (referred to as racial disparities moving forward) in cognitive outcomes remain a pressing public health concern, with evidence suggesting disproportionate burdens of comorbid neuropathologies among minoritized populations. White matter hyperintensities (WMH), a radiographic marker of small vessel cerebrovascular disease, are more prevalent and severe non-Hispanic Black (NHB) adults [1], [2], and are associated with poorer cognitive performance across multiple domains [3], [4]. Epidemiological studies, including the Lancet Commission on dementia prevention, have identified modifiable lifestyle and cardiometabolic risk factors such as diabetes, hypertension, dyslipidemia, and tobacco dependence as key drivers of both vascular and neurodegenerative processes [5], [6], [7], [8], [9]. Importantly, the prevalence and clustering of these risk factors differ across racial ethnic groups, contributing to observed disparities in dementia risk [10]. Our previous study confirmed that males often present with an increased prevalence of cardiometabolic risk factors that could contribute to detrimental age-related outcomes [7]. Moreover, males from underrepresented racial ethnic groups are less likely to be enrolled in Alzheimer’s disease and related dementias (ADRD) research, leading to gaps in our understanding of their unique risk profiles [11], [12].

Identifying effective intervention points within this complex web of determinants is challenging [13]. Traditional epidemiological approaches often fall short in quantifying the population-level benefits of targeted prevention strategies, particularly when multiple risk factors interact to influence disease progression [14], [15]. Cardiometabolic dysfunction has emerged as a potential “bottleneck” mechanism, serving as a convergent pathway through which diverse exposures accelerate both vascular injury and Alzheimer’s disease (AD) pathology [16], [17]. By intervening at cardiometabolic pathways, it may be possible to mitigate downstream neurodegenerative changes and cognitive decline. Prior work, including studies estimating population attributable fraction (PAF), suggests that addressing modifiable cardiometabolic risks could substantially reduce dementia incidence [5], [14], [15]. However, PAF-based analyses typically assume causal independence between risk factors and do not account for their complex interplay, variable prevalence, or distribution across racial ethnic groups [18].

Population intervention models (PIM) provide an alternative framework to PAF-based analyses for estimating the potential impact of hypothetical interventions under counterfactual scenarios while accounting for confounding and effect modification [19], [20], [21], [22]. By simulating the elimination of a specific risk factor in a given population, PIM yield estimates of the average treatment effect and the magnitude of cognitive improvement that could be achieved. The implementation of PIM is particularly valuable in racially ethnic diverse cohorts, where the distribution of risk factors and their associations with outcomes may differ substantially, making “one-size-fits-all” prevention strategies less effective.

Using data from the Health and Aging Brain Study – Health Disparities (HABS-HD), focusing exclusively on non-Hispanic Black (NHB) and non-Hispanic White (NHW) males, two objectives were identified. First, we sought to evaluate racial ethnic disparities in the associations between individual cardiometabolic risk factors (diabetes, hypertension, dyslipidemia, obesity, and tobacco dependence) and cognitive outcomes and WMH volume. Second, we aimed to quantify the potential gains in cognitive performance and reductions in WMH volume that could be achieved under counterfactual scenarios in which each cardiometabolic risk factor was eliminated, stratified by racial ethnic group.

Methods

Study Design

The Healthy Aging Brain Study – Health Disparities (HABS-HD) is an ongoing longitudinal investigation conducted by the Institute for Translational Research at the University of North Texas Health Science Center. The cohort comprises NHW, NHB, and Hispanic older adults, who are recruited through community-engaged research strategies in the Dallas–Fort Worth metropolitan area [23]. The study procedures have received approval from the North Texas Regional Institutional Review Board and adhere to the ethical principles outlined in the 1975 Declaration of Helsinki. All participants provided written informed consent before completing demographic and health history questionnaires, undergoing cognitive assessments, and participating in clinical laboratory testing, which includes blood collection. In addition, participants receive brain imaging via magnetic resonance imaging (MRI) and positron emission tomography (PET) to assess Alzheimer’s disease–related biomarkers such as amyloid and neurofibrillary tau. A detailed description of the HABS-HD methodology has been published previously [23], [24].

Inclusion Criteria

We used baseline HABS-HD data from release 6 (December 6, 2024), which included 3,833 participants identifying as NHW, NHB, or Hispanic. For this analysis, we restricted the sample to self-reported male participants, resulting in 1,457 individuals. Because previous research has documented substantial racial ethnic disparities in cognitive aging and cerebrovascular disease between NHW and NHB populations [25], [26], we focused on NHW and NHB participants. This resulted in 974 males for the final analytic sample. Participants had completed demographic and background questionnaires, reported their education level, had information on the presence or absence of cardiometabolic risk factors, and underwent neuropsychological evaluation for cognitive testing, and neuroimaging via magnetic resonance imaging (MRI). Missingness is detailed in Table 1.

Table 1.

Sample Characteristics Compared Between NHW and NHB Males

Variable	Overall (N = 974)	NHW (N = 581)	NHB (N = 393)	p
Age (years), Mean (SD)	67.1 (8.7)	69.5 (8.7)	63.5 (7.3)	<0.001
Education (years), Mean (SD)	15.3 (2.8)	15.9 (2.7)	14.3 (2.7)	<0.001
APOE ε4 Carrier, n (%)	291 (29.9)	167 (28.7)	124 (31.6)	0.010
Missing APOE ε4, n (%)	140 (14.4)	53 (9.1)	87 (22.1)	—
APOE Genotype, n (%)				0.010
E2E2	5 (0.5)	2 (0.3)	3 (0.8)
E2E3	104 (10.7)	66 (11.4)	38 (9.7)
E2E4	29 (3.0)	13 (2.2)	16 (4.1)
E3E3	434 (44.6)	293 (50.4)	141 (35.9)
E3E4	225 (23.1)	139 (23.9)	86 (21.9)
E4E4	37 (3.8)	15 (2.6)	22 (5.6)
Missing, n (%)	140 (14.4)	53 (9.1)	87 (22.1)
Diabetes, n (%)	216 (22.2)	109 (18.8)	107 (27.2)	0.002
Hypertension, n (%)	693 (71.1)	384 (66.1)	309 (78.6)	<0.001
Dyslipidemia, n (%)	638 (65.5)	410 (70.6)	228 (58.0)	<0.001
Obese, n (%)	430 (44.1)	233 (40.1)	197 (50.1)	<0.001
Missing Obese, n (%)	5 (0.5)	4 (0.7)	1 (0.3)
Tobacco Dependence, n (%)	115 (11.8)	34 (5.9)	81 (20.6)	<0.001
CDR Sum, Mean (SD)	0.7 (1.6)	0.6 (1.5)	0.9 (1.7)	0.003
Missing CDR Sum, n (%)	1 (0.1)	0 (0.0)	1 (0.3)
Memory, Mean (SD)	0.0 (0.9)	0.2 (0.9)	−0.3 (0.8)	<0.001
Executive Function, Mean (SD)	0.0 (0.9)	0.2 (0.8)	−0.3 (0.9)	<0.001
Processing Speed, Mean (SD)	0.0 (0.9)	0.2 (0.8)	−0.3 (1.1)	<0.001
Missing Proc. Speed, n (%)	4 (0.4)	2 (0.3)	2 (0.5)
Language, Mean (SD)	0.0 (0.9)	0.1 (0.9)	−0.2 (0.9)	<0.001
Magnetic Resonance Scanner, n (%)				<0.001
Skyra	371 (38.1)	371 (63.9)	0 (0.0)
Vida	565 (58.0)	186 (32.0)	379 (96.4)
Missing, n (%)	38 (3.9)	24 (4.1)	14 (3.6)
Raw WMH Volume (mL), Mean (SD)	10.1 (61.6)	4.3 (7.2)	19.4 (98.3)	<0.001
Missing, n (%)	76 (7.8)	28 (4.8)	48 (12.2)
Intracranial Volume mm3, Mean (SD)	1,560,000 (115,000)	1,580,000 (110,000)	1,530,000 (115,000)	<0.001
Missing, n (%)	75 (7.7)	31 (5.3)	44 (11.2)
Normalized WMH Volume (log-transformed), Mean (SD)	0.0 (3.2)	−0.1 (3.2)	0.2 (3.1)	0.160
Missing, n (%)	107 (11.0)	36 (6.2)	71 (18.1)
CVD, n (%)	100 (10.3)	73 (12.6)	27 (6.9)	0.006
Cognitive Diagnosis, n (%)				<0.001
Normal	632 (64.9)	420 (72.3)	212 (53.9)
MCI	261 (26.8)	115 (19.8)	146 (37.1)
Dementia	78 (8.0)	45 (7.7)	33 (8.4)
Missing	3 (0.3)	1 (0.2)	2 (0.5)

Values are presented as Mean (Standard Deviation) for continuous variables and count (percentage) for categorical variables. APOE = apolipoprotein E; NHW = Non-Hispanic White; NHB = Non-Hispanic Black; MCI = Mild Cognitive Impairment; CDR = Clinical Dementia Rating; CVD = Cardiovascular Disease.

Justification of Focusing Only on NHW and NHB Males

Similar to our prior work [9], the rationale for restricting this analysis to males in HABS-HD is to further elucidate male-specific biological mechanisms and risk factors while minimizing potential confounding from sex hormones and other sex-linked influences on brain health and cognitive impairment. In this study, we specifically compared NHW and NHB males to address a critical gap in understanding how racial ethnic identity and cardiometabolic health converge to shape cerebrovascular aging, with a future goal of informing cross-cohort validation. Prior research in this area has been constrained by three key issues: (i) pronounced midlife increases in hypertension and diabetes among males that complicate direct sex comparisons, (ii) survival differences that reduce comparability with older adult females, and (iii) the disproportionate exposure of NHW and NHB males to adverse lifestyle and social risk factors, which may amplify the cumulative impact of brain pathologies such as WMH and cognitive impairment. The present study was specifically designed to overcome these limitations. Additionally, it is important to note that race is a social construct representing differential exposure to societal, environmental and systemic environments. Thus, the observed disparities between racial groups are contextualized by societal and historical contexts and are not inherently biologically driven.

Outcome Measures

The primary outcomes were cognitive performance and WMH volume.

Neuropsychological Assessment

Neuropsychological testing procedures for HABS-HD have been described in detail elsewhere [23]. Briefly, participants completed the following measures: Mini-Mental State Examination (MMSE) [27]; Wechsler Memory Scale–Third Edition (WMS-III) Digit Span and Logical Memory [28]; Digit Symbol Substitution; Trail Making Test Parts A and B [29]; Spanish–English Verbal Learning Test (SEVLT) [30]; Animal Naming (semantic fluency) [30]; and FAS (phonemic fluency) [31].

Cognitive Outcomes

Cognitive domain scores were derived following established HABS-HD methodology [32]. Episodic memory was assessed using a composite score that averaged z-scores from four tests: SEVLT Delayed Recall, SEVLT Immediate Recall, Logical Memory I, and Logical Memory II. Executive function was represented by the mean z-score from the Digit Span and Digit Symbol Substitution tests. Processing speed was indexed by combining z-scores from the Trail Making Test Parts A and B and then inverting the scores so that higher values indicated better performance. Language ability was evaluated using the mean z-score from the Animal Naming and FAS Verbal Fluency tests. For each domain, raw scores were standardized into z-scores and averaged to create the composite measure for each domain.

Neuroimaging Data for MRI

Details of the HABS-HD neuroimaging acquisition and processing protocols have been reported previously [23]. All participants underwent MRI scanning on a 3 Tesla Siemens system (Siemens Healthineers AG, Erlangen, Germany) using a standardized protocol. High-resolution T1-weighted images were acquired with a magnetization-prepared rapid gradient-echo (MPRAGE) sequence (repetition time = 2300 ms; echo time = 2.93 ms; field of view = 270 mm; matrix size = 256; slice thickness = 1.2 mm; voxel resolution = 1.1 × 1.1 × 1.2 mm). T2-weighted fluid-attenuated inversion recovery (FLAIR) sequences were collected with a repetition time of 4800 ms, echo time of 441 ms, inversion time of 1650 ms, and voxel dimensions of 1.0 × 1.0 × 1.2 mm, maintaining a slice thickness of 1.2 mm and a field of view of 256 mm. All scans were obtained in the sagittal plane.

WMH Volume

WMH volumes were quantified from MPRAGE and FLAIR images using the Lesion Growth Algorithm within the Lesion Segmentation Toolbox for Statistical Parametric Mapping (www.statisticalmodelling.de/lst.html) [33]. Quality control was performed by trained raters who visually inspected each WMH mask to ensure anatomical accuracy and absence of substantial omissions or artifacts. Masks with extreme lesion volumes were re-evaluated to rule out scans indicative of other neurological pathologies. Fewer than 5% of segmentations were excluded following this process.

Raw WMH volume values exhibited strong right skewness. To retain zero values, a minimal constant (1 × 10⁻⁶) was added prior to applying a natural log transformation. To account for differences in head size, we used a residual-adjustment method: log-transformed WMH volume was regressed on intracranial volume (ICV), and the residuals representing ICV-adjusted WMH volume were used as the primary outcome measure in all subsequent analyses [7], [8], [9], [34], [35]. The skewedness and transformation depictions are shown in Supplemental Figure 1.

Exposure variables

Cardiometabolic Risk Factors

Hypertension, diabetes, dyslipidemia, obesity, and tobacco dependence were identified through a standardized consensus approach, and these variables have been used in other HABS-HD studies [7], [8], [9], [36]. Hypertension was defined by a self-reported history, current use of antihypertensive medications, or an average of two blood pressure measurements exceeding 140/90 mmHg. Diabetes was determined by self-report, treatment with insulin or oral hypoglycemic agents, or an HbA1c value above 6.5%. Dyslipidemia was classified based on self-reported high cholesterol or triglycerides, current use of lipid-lowering therapy, total cholesterol levels over 200 mg/dL, or triglycerides above 150 mg/dL. Obesity was defined as a body mass index (BMI) ≥ 30 kg/m². Tobacco dependence was assessed via questionnaire regarding current and past use of tobacco products.

Covariates

Demographic Questionnaire

As part of the HABS-HD protocol, participants self-identify their race and ethnicity as NHW or NHB. At the baseline visit, age and total years of education are recorded for all participants.

Apolipoprotein E (APOE)

Genotyping for the APOE ε4 allele was conducted using ThermoFisher Scientific TaqMan assays for single-nucleotide variants rs429158 and rs7412, with GTXpress Master Mix on an Applied Biosystems 7500 real-time PCR system [23], [37]. Allele-specific fluorescence signals from these loci were used to assign genotypes: ε2/ε2 (T T), ε2/ε3 (T CT), ε2/ε4 (CT CT), ε3/ε3 (T C), ε3/ε4 (CT C), and ε4/ε4 (C C). For analyses, participants were classified as ε4 carriers (≥1 ε4 allele) or non-carriers (ε2/ε2, ε2/ε3, or ε3/ε3).

Stroke and Cardiovascular Disease History

Participants self-reported whether they had experienced a stroke and/or had a history of cardiovascular disease (CVD), which included heart attack, heart failure, cardiomyopathy, atrial fibrillation, or heart valve replacement.

Statistical Analysis

Sample Comparisons

Baseline participant characteristics were compared between NHW and NHB participants. Continuous variables, including demographic, clinical, cognitive, and neuroimaging measures, were assessed for differences across racial ethnic groups using independent two-sample t-tests. For each comparison, we calculated mean values for each group, the corresponding t-statistic, and p-values. Categorical variables, including APOE ε4 carriership, genotype distributions, cardiometabolic risk factors, and clinical diagnoses, were compared between groups using Pearson’s chi-square tests. When the expected frequency in any cell was less than five, Fisher’s exact test with Monte Carlo simulation (10,000 replicates) was applied to obtain accurate p-values.

Multivariable Linear Regression Models: Main Effects

Cognitive domain scores were modeled as a function of age, education (continuous), racial ethnicity (NHW vs. NHB) APOE ε4 carriership (present/absence), and each cardiometabolic condition (coded as binary indicators for presence versus absence). For all linear regression models, we reported beta coefficients (β) and 95% confidence intervals (CIs) and the p-values for the five cardiometabolic risk factors in each respective model for the cognitive outcome. Identical models were run with WMH volume as the outcome. For models that had WMH as the outcome, a covariate for the MRI scanner was added. We also ran interaction models between racial ethnicity and the cardiometabolic risk factors and found no significant interactions (all p>0.05, results not shown).

Population Intervention Model

To estimate population-level changes in cognition and WMH volume under hypothetical modifications of cardiometabolic risk factors, we applied population intervention measures (PIM) using a regression-based counterfactual framework [38]. PIM differ from population-attributable fractions (PAF). While both metrics incorporate the strength of the exposure–outcome association and the prevalence of the exposure, PIM rely on regression-based counterfactual predictions that explicitly adjust for measured confounders and allow estimation within specific populations. In contrast, PAF typically rely on stronger assumptions and do not directly incorporate multivariable-adjusted counterfactual outcomes [39].

PIM project how the mean outcome in a population would change under a specified intervention on an exposure while holding the distribution of measured covariates constant. Although this approach is rooted in causal inference, we emphasize that the present analyses do not meet all assumptions required for formal causal interpretation, particularly the absence of unmeasured confounding; therefore, PIM estimates should be interpreted as counterfactual projections based on observed data rather than definitive causal effects [40]. PIM are conceptually related to population-attributable fractions in that both incorporate the prevalence of a risk factor and its association with the outcome, but PIM are estimated using multivariable-adjusted regression models and generate explicit counterfactual predictions within the study population [19], [21], [22]. For each cognitive domain (memory, executive function, processing speed, and language), we fit linear regression models including age, sex, education, APOE ε4 status, racial ethnicity (NHW vs. NHB), and all five cardiometabolic risk factors.

Cognitive scores were predicted under three counterfactual exposure conditions: (1) all participants exposed, (2) none exposed, and (3) observed exposure. Differences between the observed and full non-exposure scenarios reflected the potential cognitive benefit of eliminating the risk factor. These estimates were stratified by racial ethnicity to assess group-specific effects across four harmonized cognitive domains: memory, executive function, processing speed, and language. We calculated 95% CIs using 1,000 bootstrap resamples. Positive PIM estimates indicate cognitive benefit from eliminating the exposure, whereas negative estimates suggest potential harm. PIM results were reported for exposures with significant main effects in adjusted models only. Identical PIM models were run with WMH as the outcome to estimate the reductions in WMH when a significant risk factor was eliminated. The direction and magnitude of the PIM reflect the adjusted association between the risk factor and the outcome, with positive values indicated improvement following risk factor removal on the cognitive outcomes and negative for the WMH volume outcome. Appropriately, we interpret PIM estimates as counterfactual projections rather than causal effects [22], [41].

All analyses were conducted using R version 4.2.3. All regression models were estimated using R’s default complete-case approach, whereby observations with missing values in the outcome or any included covariates were automatically omitted on a model-by-model basis. Statistical significance was determined with a threshold of p < 0.05. All statistical tests were two-tailed.

Data Availability:

The HABS-HD data is available upon request at: https://ida.loni.usc.edu/login.jsp

Results

Sample Comparisons

Baseline demographic, genetic, cardiometabolic, lifestyle, cognitive, and neuroimaging characteristics of the analytic sample (974 male participants), stratified 581 NHW males and 393 NHB males is presented in Table 1. NHB participants were, on average, younger (63.5 ± 7.3 years) and had fewer years of education (14.3 ± 2.7) compared with NHW participants (69.5 ± 8.7 years; 15.9 ± 2.7 years; both p < 0.001). APOE ε4 carriership was slightly more prevalent among NHB participants (31.6%) than NHW participants (28.7%; p = 0.010), and genotype distributions differed significantly by race. The prevalence of diabetes (27.2% vs. 18.8%; p = 0.002) and hypertension (78.6% vs. 66.1%; p < 0.001) was higher among NHB participants, while dyslipidemia was more common among NHW participants (70.6% vs. 58.0%; p < 0.001). NHB participants also had a higher prevalence of obesity (50.1% vs. 40.1%; p < 0.001) and tobacco dependence (20.6% vs. 5.9%; p < 0.001). Across cognitive domains, NHB participants scored lower than NHW participants in memory (−0.28 ± 0.84 vs. 0.18 ± 0.89), executive function (−0.27 ± 0.86 vs. 0.17 ± 0.82), processing speed (−0.26 ± 1.05 vs. 0.15 ± 0.82), and language (−0.16 ± 0.87 vs. 0.10 ± 0.87; all p < 0.001). Neuroimaging measures showed that NHB participants had significantly higher raw WMH volumes (19.4 ± 98.3 mL) compared with NHW participants (4.3 ± 7.2 mL; p < 0.001).

Associations Between Cardiometabolic Risk Factors, Cognitive Performance, and WMH Volume

Fully adjusted linear regression models showing associations of age, APOE ε4 status, race, and cardiometabolic risk factors with cognitive performance across four domains and with WMH volume are shown in Table 2. Compared with NHW participants, NHB participants had lower scores across all cognitive domains: memory (β = −0.49, 95% CI: −0.61 to −0.36; p < 0.01), executive function (β = −0.45, 95% CI: −0.57 to −0.33; p < 0.01), processing speed (β = −0.48, 95% CI: −0.61 to −0.35; p < 0.01), and language (β = −0.27, 95% CI: −0.40 to −0.15; p < 0.01)—and had greater WMH volume (β = 1.26, 95% CI: 0.59 to 1.92; p < 0.01).

Table 2.

Associations Between Cardiometabolic Risk Factors and Cognitive Performance Across Four Cognitive Domains and WMH Volume

Predictor	Memory β (95% CI) p	Executive Function β (95% CI) p	Processing Speed β (95% CI) p	Language β (95% CI) p	WMH Volume β (95% CI) p
Age at Visit	−0.04 (−0.04, −0.03) p < 0.01	−0.03 (−0.04, −0.03) p < 0.01	−0.04 (−0.05, −0.03) p < 0.01	−0.03 (−0.04, −0.02) p < 0.01	0.10 (0.07, 0.13) p < 0.01
Education (years)	0.09 (0.07, 0.11) p < 0.01	0.11 (0.09, 0.13) p < 0.01	0.07 (0.05, 0.10) p < 0.01	0.09 (0.07, 0.11) p < 0.01	−0.09 (−0.17, −0.00) p = 0.05
APOE ε4 Carrier	−0.14 (−0.25, −0.03) p = 0.02	−0.09 (−0.20, 0.02) p = 0.10	−0.11 (−0.23, 0.01) p = 0.07	−0.12 (−0.23, −0.01) p = 0.04	0.32 (−0.15, 0.78) p = 0.18
Racial Ethnicity (NHB/ reference: NHW)	−0.49 (−0.61, −0.36) p < 0.01	−0.45 (−0.57, −0.33) p < 0.01	−0.48 (−0.61, −0.35) p < 0.01	−0.27 (−0.40, −0.15) p < 0.01	1.26 (0.59, 1.92) p < 0.01
Diabetes	−0.14 (−0.27, −0.01) p = 0.04	−0.07 (−0.19, 0.06) p = 0.28	−0.13 (−0.27, 0.01) p = 0.06	−0.15 (−0.29, −0.02) p = 0.02	0.33 (−0.22, 0.88) p = 0.24
Hypertension	0.07 (−0.05, 0.20) p = 0.26	−0.05 (−0.17, 0.07) p = 0.38	−0.09 (−0.22, 0.04) p = 0.17	−0.07 (−0.19, 0.06) p = 0.31	0.61 (0.09, 1.12) p = 0.02
Dyslipidemia	−0.02 (−0.14, 0.10) p = 0.75	0.03 (−0.08, 0.14) p = 0.59	0.09 (−0.04, 0.21) p = 0.16	0.08 (−0.04, 0.19) p = 0.21	−0.19 (−0.67, 0.30) p = 0.45
Obesity	0.09 (−0.02, 0.20) p = 0.12	−0.00 (−0.11, 0.10) p = 0.97	0.12 (0.00, 0.24) p = 0.05	−0.01 (−0.12, 0.11) p = 0.91	0.29 (−0.18, 0.76) p = 0.22
Tobacco Dependence	−0.29 (−0.47, −0.10) p < 0.01	−0.19 (−0.36, −0.01) p = 0.03	−0.23 (−0.42, −0.03) p = 0.02	−0.20 (−0.38, −0.01) p = 0.04	0.05 (−0.73, 0.82) p = 0.91

Each column presents the beta estimate with 95% confidence intervals and corresponding p-values for the respective predictor and outcome. All models adjust for age, education, APOE ε4 carriership, and racial ethnicity. Categorical predictors were coded as follows: APOE ε4 carrier (1 = carrier), Race (1 = Non-Hispanic Black [NHB], 0 = Non-Hispanic White [NHW]), and all cardiometabolic risk factors (1 = presence, 0 = absence). White matter hyperintensities volumes (WMH) models also included scanner as a covariate. Statistically significant associations (p < 0.05) are bolded.

Among cardiometabolic risk factors, diabetes was associated with lower memory (β = −0.14, 95% CI: −0.27 to −0.01; p = 0.04) and language (β = −0.15, 95% CI: −0.29 to −0.02; p = 0.02) scores, but was not related to WMH volume (β = 0.33, 95% CI: −0.22 to 0.88; p = 0.24). Hypertension was not associated with cognitive performance but was linked to greater WMH volume (β = 0.61, 95% CI: 0.09 to 1.12; p = 0.02). Tobacco dependence was associated with poorer performance in memory (β = −0.29, 95% CI: −0.47 to −0.10; p < 0.01), executive function (β = −0.19, 95% CI: −0.36 to −0.01; p = 0.03), processing speed (β = −0.23, 95% CI: −0.42 to −0.03; p = 0.02), and language (β = −0.20, 95% CI: −0.38 to −0.01; p = 0.04), but was not associated with WMH (β = 0.05, 95% CI: −0.73 to 0.82; p = 0.91). In interaction models between racial ethnicity and the cardiometabolic risk factors, there were no significant interaction associations (all p ≥0.05, results not shown).

Counterfactual Scenario using Population Intervention Models for Cognition and WMH volume

PIM estimates of cognitive outcomes under counterfactual scenarios eliminating individual cardiometabolic risk factors were calculated separately for NHW and NHB males (Figure 1A-D). Diabetes was associated with modest cognitive gains in both racial groups. Eliminating diabetes was estimated to improve memory by 0.027 (95% CI: 0.002 to 0.052) in NHW participants and by 0.039 (95% CI: 0.004 to 0.075) in NHB participants. Similar PIM effects were observed for language, with estimated gains of 0.029 (95% CI: 0.003 to 0.055) in NHW participants and 0.042 (95% CI: 0.005 to 0.083) in NHB participants. Tobacco dependency showed the most consistent and robust PIM effects across domains and racial ethnic groups. In NHW participants, full elimination of tobacco dependence was associated with cognitive gains of 0.015 (95% CI: 0.004 to 0.027) in memory, 0.010 (95% CI: 0.001 to 0.020) in executive function, 0.012 (95% CI: 0.000 to 0.026) in processing speed, and 0.010 (95% CI: 0.000 to 0.021) in language. Corresponding PIM estimates in NHB participants were larger: 0.056 (95% CI: 0.017 to 0.098) in memory, 0.037 (95% CI: 0.002 to 0.073) in executive function, 0.045 (95% CI: 0.002 to 0.095) in processing speed, and 0.038 (95% CI: −0.001 to 0.076) in language.

Fig. 1. Population Intervention Model (PIM) estimates of cardiometabolic risk factor elimination on cognition and WMH volume.

Panels A–D show the estimated cognitive gains (standardized β) associated with eliminating diabetes and tobacco for memory, executive function, processing speed, and language. Panel E shows the estimated reduction in white matter hyperintensity (WMH) volume with elimination of hypertension. Results are stratified by race and ethnicity (non-Hispanic White [NHW] and non-Hispanic Black [NHB]). Error bars represent 295% confidence intervals from 1,000 bootstrap samples for models. All models adjust for age, education, APOE ε4 carriership, and other cardiometabolic risk factors; WMH models additionally adjust for scanner.

PIM were not utilized for hypertension, dyslipidemia, or obesity across cognitive domains since there were no main effects for each risk factor. In addition, the PIM estimates of WMH volume under counterfactual scenarios eliminating hypertension in NHW and NHB males are presented in Figure 1E. Hypertension was the only cardiometabolic risk factor associated with WMH in this sample. When applying PIM, the projected reduction in WMH volume was −0.394 (95% CI: −0.769 to −0.014) in NHW participants, while among NHB participants, the estimated reduction was −0.481 (95% CI: −0.940 to −0.018).

Discussion

In this study we examined racial ethnic differences in cognitive performance and WMH as well as estimated potential benefits of eliminating modifiable cardiometabolic risks in 974 older NHW and NHB males from the HABS-HD cohort. We observed three key findings. First, tobacco dependence and diabetes emerged as consistent predictors of poorer cognitive performance. Second, hypertension was the sole cardiometabolic factor associated with higher WMH volume in this subsample. Third, population intervention modeling indicated that hypothetical elimination of tobacco dependence and diabetes could yield modest but meaningful gains in cognitive performance among NHB males. Although confidence intervals overlapped, the estimated benefits were larger for NHB males than for NHW participants. Similarly, removal of hypertension was projected to reduce WMH in both racial groups, with slightly larger estimates for NHB males. These findings highlight the differential distribution and impact of vascular risk factors in NHW and NHB males and suggest that targeted modification of specific risks could have measurable effects on brain health and cognitive aging.

Tobacco dependence was robustly associated with lower scores across all cognitive domains, aligning with literature linking smoking to accelerated cognitive decline and ADRD risk [5], [42], [43]. Mechanistically, tobacco dependence is associated with chronic inflammation, vascular injury, and oxidative stress, which may exacerbate brain aging and neurodegeneration [44]. Further, prior research focusing on the environmental affordances model has shown that males often engage in unhealthy coping strategies for stress, including tobacco dependence [45], [46], [47], [48]. Leisure activities such as cigar smoking, which is a form of tobacco dependence, are more prevalent among NHB males compared to NHW males [49], [50]. The higher prevalence of tobacco dependence in NHB males may contribute to the disproportionate cognitive impairment observed in this group. This behavioral pattern reflects a broader structural and social determinant of health, in which lifelong exposures to adverse environments and stressors can accelerate biological aging and cognitive decline [13], [51].

Diabetes was also associated with poor cognitive performance. Our results are consistent with a comprehensive literature review that has documented that diabetes is associated with AD, dementia, and reduced cognitive performance [52], [53]. Importantly, diabetes has not only been associated with an increase in AD, but also increased dementia post-stroke with the effects being observable higher in Black older adults [54]. In this sample of NHW and NHB males, diabetes was associated with the lower memory and language composite, which are also the domains that AD neuropathology primarily affects. Aligning with our PIM statistical findings, Tang and colleagues did not find that glycemic control had benefit in executive function, attention, or processing speed cognitive domains [55]. Given the expected increase in the prevalence of both type 2 diabetes [56] and cognitive impairment over the coming decades [57], it is imperative to find alternative treatments that can serve as upstream targets for conditions later in life, such as dementia. Although we remain a long way from complete elimination, alleviation of diabetes is especially important given its higher prevalence in non-White populations, underscoring pathways through which some health disparities may be partially reduced.

While hypertension was not significantly associated with explicitly poor cognitive performance, it was independently associated with WMH volume in this male-only sub cohort, consistent with our previous findings in the full HABS-HD male sample [7]. Similar associations were found for hypertension and WMH in the full sample along with a sub cohort of females [7], [8]. Further, our results reinforce prior findings that hypertension is one of the strongest predictors of WMH beyond age [58]. Previous evidence has shown that blood pressure interventions that stabilize blood pressure to normal levels, indicating better hypertension management, are associated with reduced WMH progression and volumes [59]. Although the mechanism of hypertension being associated with WMH is not fully elucidated, several hypotheses suggest that vascular injury, impaired cerebral autoregulation, and chronic hypoperfusion may underlie this relationship [60]. More specifically, progressive hypertension initiates microvascular injury, which disrupts neuroglia function, and the damage to these cells contributes to several pathological processes: enlargement of perivascular spaces, vasodilator dysfunction, microvascular rarefaction, and blood–brain barrier disruption [60]. Collectively, these changes disrupt cerebrovascular regulation, reducing cerebral blood flow, promoting neuroinflammation, and culminating in presymptomatic brain damage detectable on neuroimaging as WMH before overt clinical symptoms emerge [60]. Thus, the management of cardiometabolic risks, is imperative to alleviate structural changes in the brain.

Using PIM, we provided estimates that the elimination of modifiable risks such as tobacco dependence and diabetes could yield small but measurable cognitive benefits. The benefits of reducing or eliminating tobacco dependence among minoritized males extend beyond cognition to cardiovascular outcomes, including lower risk of stroke [61], [62], [63]. Because smoking is a shared risk factor for stroke, cognitive decline, and post-stroke dementia, our PIM findings suggest that tobacco cessation could play a key role in preventing early cognitive impairment, well before the onset of mild cognitive impairment. Unlike hypertension which has shown a paradoxical association with better cognition in some centenarians or “super agers” [64], [65], smoking presents a consistently harmful risk profile, accelerating earlier cognitive decline [5]. Evidence from large-scale studies supports this interpretation. Zhong and colleagues reported that midlife smoking was associated with a 30% higher risk of dementia across 37 prospective cohorts, whereas no excess risk was observed in former smokers [66]. The Framingham Heart Study found that individuals who began smoking in early adulthood (ages 33–44) had the highest risk of dementia in a 21-year follow-up [67]. Similarly, the ARIC study (25-year follow-up) and the Whitehall II study (32-year follow-up) demonstrated elevated dementia risks for current smokers in midlife [68], [69]. Collectively, these findings provide strong evidence that smoking tobacco confers a durable and elevated dementia risk across the adult life course, and that cessation remains one of the most effective strategies for mitigating early, accelerated, and long-term cognitive decline.

In this context, our findings of tobacco-related cognitive deficits in the overall sample, and the projected benefits of tobacco elimination particularly for NHB males, are consistent with the literature, specifically for interventions that promote cessation [70], [71], [72]. Tobacco cessation or complete abstinence may confer restorative benefits for cognitive function, as shown in cessation studies demonstrating slower rates of cognitive decline and reduced dementia risk following sustained abstinence from tobacco [70], [71], [72]. These findings are in line with evidence from smoking cessation studies suggesting that quitting may partially reverse cognitive deficits and slow decline [72], supporting our PIM findings of measurable cognitive gains under hypothetical tobacco elimination, particularly in NHB males.

In addition to smoking cessation, we observed that NHB males show the potential for cognitive gains in memory and language domains from the elimination of diabetes using the PIM statistical models. These findings align with meta-analyzed results for randomized clinical trials in which intensive glucose control was associated with a slower decline in overall cognitive performance and memory compared to conventional treatment [55]. Moreover, type 2 diabetes management medications such as glucagon-like peptide-1 (GLP-1) receptor agonists have been observed to have an association with a decreased risk of dementia (results from 26 randomized clinical trials) [73]. Hence, our PIM statistical findings align with pharmacological interventions that are targeted towards intense glucose control for type-2 diabetes.

Using PIM statistical models, our findings additionally confirmed that increased disparities in WMH, based on the main effect association of hypertension, can substantially reduce this comorbid pathology that is present in older adults [74]. Hence, alleviating hypertension not only reduces a key prognostic risk factor for dementia, but its counterfactual removal in this sample also aligns with previous literature that intense blood pressure control may be a viable target to reduce WMH [75], thereby lowering subsequent risks of dementia and cognitive impairments. Although we did not directly assess blood pressure measurements in this study, future studies could investigate how normalizing blood pressure can reduce WMH volume, hypothetically based on the main effects of the associations. However, in this study, that goes beyond the scope.

The mitigation of hypertension can alleviate WMH volume, and the complete elimination of hypertension has the potential to decrease WMH volume in both NHB and NHW, with significant gains in among NHB men. These patterns reflect the higher prevalence aligning with prior evidence of disproportionate cerebrovascular risk in NHB males. PIM analyses suggested that complete elimination of hypertension could substantially reduce WMH volume in both NHW and NHB males, with slightly larger projected benefits in NHB. To our knowledge, this is the first study to apply PIM to estimate WMH reductions under hypertension elimination scenarios in a racially diverse, male-only cohort, providing a novel public health perspective on modifiable cerebrovascular risk.

Strengths and Limitations

There are several limitations that warrant comment. The cross-sectional design precludes establishing temporal or causal relationships between risk factors, WMH volume, and cognitive outcomes. Our restriction to male participants limits generalizability to females, who may exhibit different vascular and cognitive aging trajectories, as we have shown in prior work [8], [9]. Residual confounding is possible from unmeasured variables such as diet, physical activity, socioeconomic status, and lifetime healthcare access including several modifiable dementia risk factors identified by the Lancet Commission [5]. Some measures, including tobacco dependence and cardiovascular history, relied on self-report and may be subject to recall or reporting bias. We recognize that potential misclassification of self-reported measures may lead to conservative effect estimates. Neuroimaging was conducted at a single time point, preventing evaluation of longitudinal WMH progression. Finally, statistical power was limited for some subgroup analyses, particularly for less prevalent APOE genotypes, and we modeled APOE ε4 carriership as a binary variable, which may overlook dose-dependent effects. In addition, we did not account for or consider medications.

Despite the limitations, our study has several notable strengths which should be considered. First, it leverages a large, racially diverse cohort of older male participants, allowing stratified analyses between NHW and NHB participants which is an understudied comparison given the challenges of enrolling males, and particularly NHB males, in aging research. Second, cognitive performance was assessed using a standardized neuropsychological battery spanning multiple domains, with composite scores derived using established methods [32]. Third, we applied PIM to estimate the potential reductions in WMH volume and gains in cognitive performance under hypothetical elimination of specific cardiometabolic risk factors. This approach, rarely used in Alzheimer’s disease and related dementias (ADRD) research, offers several advantages over traditional population attributable fraction estimates [19], [20], [21], [22]: (1) it explicitly models the counterfactual scenario of removing a risk factor while holding the distribution of covariates constant, (2) it can be applied within fully adjusted regression models accounting for multiple confounders, and (3) it provides interpretable effect estimates directly on the scale of the outcome, making the findings more relevant for public health translation.

Conclusions and Future directions

These results underscore the importance of targeted, culturally tailored cardiometabolic prevention and management strategies not only to improve vascular health, but also to reduce cognitive decline and dementia risk, and to address persistent racial disparities in brain aging outcomes. Our findings have strong clinical implications and open several directions for future research. First, longitudinal studies are needed to determine how tobacco dependence, diabetes, and hypertension jointly contribute to trajectories of cognitive decline and WMH progression over time. Moreover, PIM models utilizing multiple comorbid conditions would help identify which patients (for example, those that present with diabetes and tobacco dependence) would derive the greatest hypothetical benefit. Second, mechanistic studies that integrate blood-based biomarkers, neuroimaging, and neuropathology could clarify the biological pathways through which cardiometabolic risks influence brain aging, such as inflammation, vascular injury, and impaired glucose metabolism. Third, more intervention studies targeting tobacco cessation, intensive glucose control, and aggressive blood pressure management, especially combinatorial trials, should test whether modifying these risks translates into measurable gains in cognitive performance and WMH reduction in real-world clinical settings. Finally, culturally tailored prevention strategies that account for structural and social determinants of health are critically needed to address the disproportionate burden of cardiometabolic disease and cognitive decline among NHB males and other underrepresented populations.