Arterial stiffness and risk of dementia in late life: The ARIC neurocognitive study

Abstract

INTRODUCTION

Associations of arterial stiffness, measured through pulse wave velocity (PWV), with incident dementia may be driven by age‐related structural changes to the arterial wall or by elevated blood pressure.

METHODS

Dementia‐free White or Black participants in the community‐based Atherosclerosis Risk in Communities (ARIC) cohort study with measures of PWV were examined. The associations of carotid‐femoral PWV (cfPWV) and total, structural, and load‐dependent PWV with dementia were estimated using Poisson regression models and Cox proportional hazards models. Dementia cases were defined by physician‐adjudicated review, telephone interviews, hospitalization records, and death certificates.

RESULTS

Of 4818 participants included in the study population (mean age: 75 years; 41% male, 21% Black), 1067 (22%) developed dementia over a median follow‐up of 9.0 years. Higher cfPWV, total PWV, structural PWV, and load‐dependent PWV were associated with greater dementia risk.

DISCUSSION

Both structural and pressure‐related arterial stiffness mechanisms may underlie the link between PWV and dementia.

Highlights

• Higher cfPWV is associated with increased dementia incidence.

• Total, structural, and load‐dependent PWV are associated with dementia incidence.

• Age and APOE ε4 status modify associations between some PWV measures and dementia.

1. INTRODUCTION

By 2060, the burden of dementia, including Alzheimer's disease (AD) and vascular dementia, is expected to double in the United States due to increases in life expectancy and population aging. ¹ Modifiable vascular risk factors, such as hypertension and diabetes, play a critical role in contributing to dementia risk across the life course. Evidence suggests that between 14% and 44% of dementia cases are directly attributable to risk factors such as elevated low‐density lipoprotein (LDL) cholesterol, diabetes, hypertension, smoking, and obesity. ² , ³

Increased arterial stiffness, a marker of vascular aging, results from a reduction in arterial elastin content, deposition of collagen, and arterial wall calcification. ⁴ Although this process occurs in part as a result of common vascular risk factors for dementia (i.e., hypertension), ⁵ age‐related structural remodeling of the arterial wall independent of other vascular risk factors also significantly contributes to arterial stiffness. ⁶ Carotid‐femoral pulse wave velocity (cfPWV), a measure of arterial stiffness, may be inversely associated with brain health due to resulting cerebral microvascular dysfunction, reduced blood flow, and structural brain damage, ⁷ , ⁸ but the mechanisms underlying this association are unclear. ⁹ Examination of the components of pulse wave velocity (PWV), including structural PWV, which reflects remodeling of the arterial wall, and load‐dependent PWV, which reflects the effects of elevated blood pressure, may elucidate the mechanisms linking arterial stiffness with dementia. Recent studies have found that cfPWV and structural PWV are associated with cognitive impairment ¹⁰ , ¹¹ and markers of brain pathology typically associated with dementia. ¹¹ , ¹² However, evidence of longitudinal associations between PWV measures and incident dementia is sparse. ⁹ , ¹⁰ Studies that have estimated longitudinal associations of PWV with incident dementia have primarily examined small, racially homogeneous samples and used less robust methods of addressing informative attrition. ¹³ , ¹⁴

In this study, we used data from the Atherosclerosis Risk in Communities (ARIC) Neurocognitive Study (ARIC‐NCS), a community‐based cohort study of predominantly Black or White adults, to explore associations of arterial stiffness, measured through cfPWV, with subsequent incident dementia. We hypothesized that higher cfPWV would be associated with increased incidence of dementia over approximately 9 years of follow‐up, independent of vascular risk factors. We also investigated associations of total, structural, and load‐dependent PWV with incident dementia to provide insight into the arterial stiffness mechanisms driving the relationship between cfPWV and dementia risk and to strengthen evidence for arterial stiffness as a dementia risk factor independent of other established vascular risk factors. In addition, we hypothesized that higher total, structural, and load‐dependent PWV would each be associated with increased incidence of dementia. Finally, we examined whether the relationship between PWV measures and incident dementia varied by age, race, sex, education, baseline hypertension status, or the presence of apolipoprotein E (APOE) ε4 alleles.

2. METHODS

2.1. Cohort

The ARIC Study is a prospective community‐based cohort study originally designed to investigate the etiology of atherosclerotic disease in a middle‐aged, predominantly Black or White population. ¹⁵ Between 1987 and 1989, 15,792 participants aged 45 to 64 years were recruited by field centers in four US communities (Washington County, Maryland; Forsyth County, North Carolina; Minneapolis, Minnesota; and Jackson, Mississippi). The initial assessment was followed by three triennial follow‐up assessments: Visit 2 (1990 to 1992, n = 14,348), Visit 3 (1993 to 1995, n = 12,887), and Visit 4 (1996 to 1998, n = 11,656). Visit 5 was conducted 15 years later, from 2011 to 2013 (n = 6538) and succeeded by Visit 6 (2016 to 2017, n = 4214), Visit 7 (2018 to 2019, n = 3589), Visit 8 (2020, n = 3226), and Visit 9 (2021 to 2022, n = 2105). The protocol for each visit was approved by the Institutional Review Boards at Johns Hopkins University, Wake Forest University, University of Mississippi Medical Center, and the University of Minnesota. Written informed consent was obtained from each participant or their legally authorized representative.

Assessments in the current investigation ranged from June 2011 to December 2022, with Visit 5 (2011 to 2013) serving as the baseline. At the baseline, measures of arterial stiffness were obtained, and a comprehensive neuropsychological battery was administered for the ARIC‐NCS. Among the 6538 participants assessed, 43 participants who self‐reported race as Asian (n = 12), Indian (n = 6), or Black living in Minnesota (n = 9) or Maryland (n = 16) were excluded due to small numbers. Participants who had a body mass index ≥ 40 kg/m² (n = 274) were excluded because of measurement accuracy concerns with cfPWV with higher body mass indices. Participants with an unknown level of education (n = 11) or with a missing or invalid cfPWV measurement (n = 1243) were also removed from the analytic sample. Among the 4967 participants remaining, 149 were excluded after being diagnosed with dementia at or before the Visit 5 baseline. The resulting analytic sample included 4818 dementia‐free participants (eFigure 1 in the Supplement).

RESEARCH IN CONTEXT

1. Systematic review: We reviewed published articles on associations between central arterial stiffness measured through cfPWV and incident dementia, as well as associations of novel metrics of total, structural, and load‐dependent PWV with indicators of brain health.

2. Interpretation: With its larger sample size and extended follow‐up compared to other prospective studies of the association between central arterial stiffness and dementia risk, this investigation provides evidence that central arterial stiffness is associated with increased risk of dementia and that the link between arterial stiffness and dementia risk may be attributable to age‐related structural remodeling of the arterial wall and increases in stiffness resulting from elevated blood pressure.

3. Future directions: Future work should determine whether arterial stiffness, in combination with other metrics, could serve as a useful tool for dementia risk stratification.

2.2. Exposure

The measurement protocol for arterial stiffness is documented in prior publications. ¹⁶ , ¹⁷ Briefly, cfPWV was measured at Visit 5 using an automated waveform analyzer (VP‐1000 Plus, Omron Healthcare, Kyoto, Japan) ¹⁸ and a segmometer (Rosscraft, Surray, Canada) after participants were supine for 5 to 10 min (eMethods in the Supplement). Higher values of cfPWV indicate greater arterial stiffness.

The exposure variable of cfPWV was explored using two analytic approaches: (1) standardized continuous (one standard deviation [SD] change of 3.1 m/s) to assess a dose–response relationship between cfPWV and dementia incidence and (2) dichotomized to indicate elevated (≥13.4 m/s, corresponding to the 75th percentile of cfPWV measurements in the study population) versus non‐elevated cfPWV, given the absence of an established clinical threshold. To examine the components of PWV attributable to age‐related structural changes separately from those attributable to the effects of blood pressure, ⁹ , ¹¹ , ¹⁹ exposure variables of total PWV, structural PWV, and load‐dependent PWV, derived from cfPWV, systolic blood pressure (SBP), and diastolic blood pressure (DBP) measurements, ¹⁹ were explored. Total PWV was derived from measurements of cfPWV, DBP, and the logarithmic mean blood pressure of the participant. Structural PWV was derived from measurements of cfPWV, DBP, and the logarithmic mean blood pressure for a SBP of 120 and a DBP of 80. Load‐dependent PWV was calculated by subtracting the structural PWV from the total PWV. Full details of the computation of these metrics are given in the eMethods in the Supplement. The same two analytic approaches were used for these exposure variables, with 1 SD of total, structural, and load‐dependent PWV representing a difference of 3.8, 3.5, or 0.9 m/s, respectively, and elevated total, structural, and load‐dependent PWV defined as ≥15.8, ≥15.5, and ≥0.7 m/s, respectively, corresponding to the 75th percentile of each measurement in the study population.

2.3. Outcome

The primary outcome was incident dementia, ascertained using an established protocol ²⁰ based on the National Institute on Aging‐Alzheimer's Association criteria ²¹ , ²² and the Diagnostic and Statistical Manual of Mental Disorders (DSM‐5). ²³ Incident dementia was determined from expert adjudication of in‐person neurocognitive exam results, phone‐based interviews and assessments, informant interviews, medical records, and death certificates after Visit 5. Full details of the dementia ascertainment process are provided in the eMethods in the Supplement.

2.4. Covariates

Multiple time‐invariant covariates were incorporated into the current investigation based on their established association with dementia. ²⁴ , ²⁵ Race, sex, and education were obtained via self‐report at Visit 1, and smoking status and weekly physical activity were assessed at Visit 5. The date of birth was used to calculate the age in years at Visit 5. APOE ɛ4 genotyping (0 or ≥1 alleles) was performed using the TaqMan assay (Applied Biosystems, Foster City, CA, USA) on blood samples obtained at Visit 1. Heart rate, mean arterial pressure (MAP), the use of hypertension medication, and the presence of diabetes were included as time‐varying cardiovascular risk factors for dementia. ²⁴ Further description of covariates and additional measurements that were reported in Table 1 but not utilized in analytic models are described in the eMethods in the Supplement.

TABLE 1.

Characteristics of study population at baseline by elevated or non‐elevated cfPWV: ARIC study, 2011 to 2013 (n = 4818).

	All N = 4818	Non‐Elevated cfPWV (<13.4 m/s) N = 3409	Elevated cfPWV (≥13.4 m/s) N = 1099	p
Age, mean (SD), years	75.2 (5)	74.6 (4.8)	77 (5.2)	<0.001
Female sex, n (%)	2844 (59)	2066 (60.6)	623 (56.7)	0.02
Black race, n (%)	1030 (21.4)	657 (19.3)	333 (30.3)	<0.001
Education, n (%)
Less than high school degree	588 (12.2)	360 (10.6)	197 (17.9)	<0.001
High school degree	2044 (42.4)	1438 (42.2)	474 (43.1)
At least some college	2186 (45.4)	1611 (47.3)	428 (38.9)
One or more APOE ε4 alleles, n (%)	1314 (28.1)	940 (28.4)	290 (27.1)	0.39
Ever smoker, n (%)	2575 (57.2)	1829 (57.2)	566 (55.5)	0.33
Heart rate, mean (SD), beats/min	64.8 (10.6)	64.1 (10.2)	67.5 (11.3)	<0.001
Body mass index, mean (SD), kg/m2	27.9 (4.5)	27.9 (4.4)	27.6 (4.7)	0.04
Physical activity, mean (SD), min/week	194.2 (187.6)	204.4 (188.7)	158.3 (177.9)	<0.001
Diabetes, n (%)	1212 (25.4)	732 (21.7)	405 (37.2)	<0.001
Mean arterial pressure, mean (SD), mmHg	87.2 (11.4)	86.1 (11)	90.6 (12.1)	<0.001
Systolic blood pressure, mean (SD), mmHg	129.9 (17.6)	127.5 (16.5)	137.8 (18.8)	<0.001
Diastolic blood pressure, mean (SD), mmHg	66.2 (10.4)	65.8 (10.2)	67.3 (10.9)	<0.001
Hypertension, n (%)	3460 (72.4)	2335 (69)	896 (82.1)	<0.001
Hypertension medication, n (%)	3483 (72.5)	2367 (69.6)	882 (80.4)	<0.001
Total cholesterol, mean (SD), mg/dL	183.6 (41.7)	184.6 (41.1)	181.8 (42.8)	0.05
HDL cholesterol, mean (SD), mg/dL	52.9 (14)	53.2 (13.7)	52.3 (14.8)	0.07
LDL cholesterol, mean (SD), mg/dL	105.9 (34.6)	106.8 (34.2)	104.1 (35.3)	0.03
Global cognition	0.2 (0.8)	0.27 (0.83)	−0.1 (0.9)	<0.001
Memory domain	0.1 (0.8)	0.18 (0.77)	−0.1 (0.8)	<0.001
Language domain	0.1 (0.8)	0.2 (0.81)	−0.1 (0.8)	<0.001
Executive function domain	0.1 (0.9)	0.24 (0.86)	−0.2 (0.9)	<0.001
Total pulse wave velocity, m/s	13.8 (3.7)	12.1 (2.2)	18.9 (2.7)	<0.001
Structural pulse wave velocity, m/s	13.6 (3.5)	12.1 (2.1)	18.2 (2.7)	<0.001
Load‐dependent pulse wave velocity, m/s	0.2 (0.9)	0.0 (0.7)	0.7 (0.1)	<0.001
Follow‐up dementia‐free, No. (%)	3751 (77.9)	2744 (80.5)	739 (67.2)	<0.001

Note: Study baseline defined as the year in which a comprehensive neuropsychological battery was administered for the first time in the Atherosclerosis Risk in Communities Neurocognitive Study. Stratified statistics are based on complete cases for Visit 5 cfPWV. Overall statistics are based on all participants included in the analytic sample. Univariate differences in study variables were assessed using chi‐squared test for binary variables, Cochran‐Armitage trend tests for ordinal variables, and t‐tests for continuous variables. Dementia diagnosis during follow‐up was determined by adjudicated review, telephone interviews, informant interviews, hospitalization records, and death certificates. Diagnosis date was based on the last examination or phone‐based assessment. If a dementia diagnosis was ascertained from an informant interview, hospitalization record, or death certificate, the date was defined as 180 days prior to the documented incident or interview. All measurements presented are described in either the Methods or the supplemental eMethods.

Abbreviations: APOE, apolipoprotein E; cfPWV, carotid‐femoral pulse wave velocity; HDL, high‐density lipoprotein; LDL, low‐density lipoprotein.

2.5. Statistical analysis

Analyses were conducted in SAS 9.4 (SAS Institute, Cary, NC, USA). Differences in the baseline characteristics of participants by elevated or non‐elevated cfPWV were evaluated utilizing chi‐squared tests, Cochran‐Armitage trend tests, and t‐tests. Multiple imputation by chained equations ²⁶ was employed to impute missing values when a participant in the analytic sample completed a visit but had one or more covariates missing or was missing the Visit 5 cfPWV measurement but had the Visit 6 or Visit 7 cfPWV measurement. Kaplan‐Meier curves of dementia cases stratified by elevated cfPWV, total PWV, structural PWV, and load‐dependent PWV were generated. Poisson regression models with robust error variance were used to estimate unconditional dementia incidence rates with 95% confidence intervals (CIs) per 1000 person‐years. A second model estimated incidence rates after adjusting for multiple time‐invariant covariates including age, sex, race‐center, education, and the presence of APOE ε4 alleles. A third model added heart rate, MAP, the use of hypertension medication, and the presence of diabetes as time‐varying cardiovascular covariates.

Cause‐specific Cox proportional hazards regression models were employed to estimate the prospective association between PWV measures at Visit 5 and the onset of dementia. The initial model calculated unconditional hazard ratios (HRs) and 95% CI. The second and third models incorporated the same covariates included in the second and third Poisson regression models. The Efron method was used to handle ties. The assumption of linearity was evaluated by examining Martingale residuals, and the proportional hazards assumption was assessed by inspecting Schoenfeld residuals. Neither modeling assumption was violated. Measurements of time‐varying covariates from Visits 6 through 9 were incorporated into the regression models. Sensitivity analyses performed incorporated stabilized inverse probability of censoring weights (IPCW) ²⁷ , ²⁸ , ²⁹ or used elevated cfPWV redefined based on age‐specific normal values derived from a population with no cardiovascular risk factors ³⁰ (eMethods in the Supplement).

Stratified analyses tested for multiplicative and additive effect modification by median baseline age, race, sex, education, baseline hypertension status (regardless of antihypertensive medication use), and the presence of APOE ε4 alleles. Statistical significance was defined as p < 0.05. Kaplan‐Meier curves stratified by elevated cfPWV and each subgroup were plotted to visualize effect modification.

3. RESULTS

3.1. Cohort characteristics

Among the 4818 participants in the analytic sample, 59.0% (2844/4818) were female and 21.4% (1030/4818) were Black. Participants with elevated cfPWV measurements (Table 1) were more likely to be older, male, Black, and report a lower level of education. Heart rate and MAP were higher and the prevalence of diabetes, hypertension, and the use of hypertension medication was greater among participants with elevated cfPWV.

The median (interquartile interval) follow‐up was 9.0 (6.7–9.9) years. The maximum follow‐up was 11.3 years. Within the subsample of 1,067 participants diagnosed with dementia during the study period, 34.7% (370/1067) were evaluated during a follow‐up neurocognitive examination and 49.3% (526/1067) were assessed during telephone interviews. The remaining cases of dementia were ascertained from hospitalization records (14.5%, 155/1067) and death certificates (1.5%, 16/1067). Kaplan‐Meier curves (Figure 1) and Poisson regression models (Table 2) revealed that dementia incidence rates were greater among participants with elevated levels of cfPWV, total PWV, structural PWV, and load‐dependent PWV. These contrasts decreased in magnitude after adjusting for demographic and cardiovascular covariates for all PWV measures.

FIGURE 1.

Kaplan‐Meier curves depicting percentage of dementia‐free participants by pulse wave velocity measures: ARIC Study (2011 to 2022) (n = 4818). Elevated cfPWV, total PWV, structural PWV, and load‐dependent PWV are defined as greater than or equal to 13.4, 15.8, 15.5, and 0.7 m/s, respectively. Multivariate imputation by chained equations was employed to impute missing cfPWV measurements. Dementia diagnosis was determined by adjudicated review, telephone interviews, informant interviews, hospitalization records, and death certificates. Diagnosis date was based on the last examination or phone‐based assessment. If a dementia diagnosis was ascertained from an informant interview, hospitalization record, or death certificate, the date was defined as 180 days prior to the documented incident or interview. cfPWV, carotid‐femoral pulse wave velocity; PWV, pulse wave velocity.

TABLE 2.

Multivariable‐adjusted Poisson‐based incidence rates (95% CI) per 1000 person‐years for dementia by non‐elevated or elevated pulse wave velocity measures: ARIC Study 2011 to 2022 (n = 4818).

	Model 1 Unconditional	Model 2 +Demographics	Model 3 +Cardiovascular
cfPWV
Non‐elevated (<13.4 m/s)	26.5 (24.6 to 28.6)	23.1 (21.3 to 25.0)	23.1 (21.3 to 25.1)
Elevated (≥13.4 m/s)	51.2 (46.2 to 56.8)	31.7 (27.9 to 36.0)	30.5 (26.8 to 34.8)
Total PWV
Non‐elevated (<15.8 m/s)	29.2 (27.1 to 31.4)	25.4 (23.5 to 27.5)	25.5 (23.5 to 27.6)
Elevated (≥15.8 m/s)	53.9 (48.7 to 59.6)	33.5 (29.5 to 38.0)	31.6 (27.7 to 36.1)
Structural PWV
Non‐elevated (<15.5 m/s)	29.1 (27.0 to 31.4)	25.6 (23.7 to 27.7)	25.6 (23.7 to 27.8)
Elevated (≥15.5 m/s)	54.2 (49.0 to 60.0)	32.9 (28.9 to 37.4)	30.9 (27.1 to 35.3)
Load‐dependent PWV
Non‐elevated (< 0.7 m/s)	31.5 (29.3 to 33.8)	25.5 (23.5 to 27.7)	25.1 (23.1 to 27.3)
Elevated (≥0.7 m/s)	44.7 (40.0 to 49.9)	32.5 (28.7 to 36.7)	32.6 (28.4 to 37.4)

Note: Poisson regression models with robust error variance were used to model dementia incidence rates by PWV exposures. Demographics models adjusted for age, sex, race‐center, education, and APOE ε4 alleles as time‐invariant covariates. Cardiovascular models additionally adjusted for heart rate, mean arterial pressure, hypertension medication, and diabetes as time‐varying covariates. Multivariate imputation by chained equations was employed to impute missing covariates and missing cfPWV measurements. Elevated PWV measures were defined as ≥75th percentile of measurements in the study population. Dementia diagnosis was determined by adjudicated review, telephone interviews, informant interviews, hospitalization records, and death certificates.

Abbreviations: APOE, apolipoprotein E; cfPWV, carotid‐femoral pulse wave velocity; CI, confidence interval; PWV, pulse wave velocity.

3.2. PWV and dementia

Higher levels of cfPWV were associated with greater risk of dementia (Figure 2). For every 1 SD (3.1 m/s) higher cfPWV, the unadjusted hazard of dementia was 37% greater (95% CI: 1.30, 1.45). This association was attenuated but remained statistically significant with the addition of demographic (HR: 1.13; 95% CI: 1.06, 1.20) and cardiovascular (HR: 1.11; 95% CI: 1.04, 1.18) covariates (Figure 2). In unconditional models, participants with elevated cfPWV (≥13.4 m/s) were approximately twice as likely (HR: 2.09; 95% CI: 1.85, 2.37) (Figure 2) to develop dementia compared to individuals without elevated cfPWV. The associations in the adjusted models remained statistically significant. Similar effect sizes (eFigure 2 in the Supplement) were evident in sensitivity analyses in which elevated cfPWV was redefined utilizing age‐specific reference values. ³⁰ In analyses of total, structural, and load‐dependent PWV, every 1 SD (3.8 m/s) higher total PWV was associated with a 10% higher hazard of dementia (95% CI: 1.03, 1.17), every 1 SD (3.5 m/s) higher structural PWV was associated with a 7% higher hazard of dementia (95% CI: 1.01, 1.14), and every 1 SD (0.9 m/s) higher load‐dependent PWV was associated with a 13% higher hazard of dementia (95% CI: 1.06, 1.21) after adjustment for demographic and cardiovascular covariates (Figure 3). When total, structural, and load‐dependent PWV were defined as elevated versus non‐elevated, elevated total PWV (≥15.8 m/s) was associated with a 29% higher hazard of dementia (95% CI: 1.13, 1.48), elevated structural PWV (≥15.5 m/s) was associated with an 23% higher hazard of dementia (95% CI: 1.07, 1.40), and elevated load‐dependent PWV (≥0.7 m/s) was associated with an 38% higher hazard of dementia (95% CI: 1.19, 1.61) in fully adjusted models (Figure 3). Parameter estimates from IPCW models were nearly identical to estimates from unweighted models (eFigures 3, 4 and 5 in the Supplement).

FIGURE 2.

Multivariable‐adjusted hazard ratio estimates (95% CI) of dementia by carotid‐femoral pulse wave velocity: ARIC Study (2011 to 2022) (n = 4818). Multivariable cause‐specific Cox proportional hazards regression models were used to model the associations between cfPWV and dementia. Demographics models adjusted for age, sex, race‐center, education, and APOE ε4 alleles as time‐invariant covariates. Cardiovascular models additionally adjusted for heart rate, mean arterial pressure, hypertension medication, and diabetes as time‐varying covariates. Multivariate imputation by chained equations was employed to impute missing covariates and missing cfPWV measurements. One standard deviation of cfPWV is equal to 3.1 m/s. Elevated cfPWV is defined as ≥13.4 m/s. Dementia diagnosis was determined by adjudicated review, telephone interviews, informant interviews, hospitalization records, and death certificates. Censoring based on the last completed examination or phone‐based assessment. If a dementia diagnosis was ascertained from an informant interview, hospitalization record, or death certificate, the date was defined as 180 days prior to the documented incident or interview. cfPWV, carotid‐femoral pulse wave velocity; CI, confidence interval.

FIGURE 3.

Multivariable‐adjusted hazard ratio estimates (95% CI) of dementia by total, structural, and load‐dependent pulse wave velocity: ARIC Study (2011 to 2022) (n = 4818). Multivariable cause‐specific Cox proportional hazards regression models were used to model the associations between total, structural, and load‐dependent PWV and dementia. Demographics models adjusted for age, sex, race‐center, education, and APOE ε4 alleles as time‐invariant covariates. Cardiovascular models additionally adjusted for heart rate, mean arterial pressure, hypertension medication, and diabetes as time‐varying covariates. Multivariate imputation by chained equations was employed to impute missing covariates and missing cfPWV measurements. One standard deviation of total, structural, or load‐dependent PWV is equal to 3.8, 3.5, or 0.9 m/s, respectively. Elevated total, structural, and load‐dependent PWV were defined as greater than or equal to 15.8, 15.5, and 0.7 m/s, respectively. Dementia diagnosis was determined by adjudicated review, telephone interviews, informant interviews, hospitalization records, and death certificates. Censoring was based on the last completed examination or phone‐based assessment. If a dementia diagnosis was ascertained from an informant interview, hospitalization record, or death certificate, the date was defined as 180 days prior to the documented incident or interview. CI, confidence interval; LD‐PWV, load‐dependent pulse wave velocity; PWV, pulse wave velocity; S‐PWV, structural pulse wave velocity; T‐PWV, total pulse wave velocity.

The median age at baseline functioned as a statistically significant moderator on the multiplicative scale such that stronger relative associations of standardized cfPWV, elevated cfPWV, standardized total PWV, standardized structural PWV, and elevated structural PWV with dementia were observed among younger participants (<74 years old) in unweighted models (eFigure 6 in the Supplement). Stratified analyses indicated that a 1 SD higher cfPWV was associated with a 35% increase (95% CI: 1.19, 1.53) in the fully adjusted hazard ratio among younger participants but only a 15% increase (95% CI: 1.07, 1.23) among older participants. Similar patterns were observed for standardized total and structural PWV. Elevated cfPWV yielded HRs for dementia of 1.95 (95% CI: 1.49, 2.56) and 1.50 (95% CI: 1.29, 1.74) for younger and older participants, respectively, with a similar pattern observed for structural PWV. In the weighted models, the multiplicative interaction for elevated cfPWV was no longer significant (eFigure 7 in the Supplement). The smaller relative associations between these PWV measures and incident dementia among older participants may be attributable to higher underlying absolute dementia risk among older participants (eFigure 8 in the Supplement), resulting in attenuated relative effect estimates on the multiplicative scale.

The presence of one or more APOE ε4 alleles functioned as a statistically significant moderator on the multiplicative scale such that stronger relative associations were observed among participants with no APOE ε4 alleles across all PWV measures in the unweighted models (eFigure 9 in the Supplement). For example, stratified analyses indicated that every 1 SD higher cfPWV was associated with a 16% increase (95% CI: 1.08, 1.25) in the fully adjusted hazard ratio for dementia among participants with zero APOE ε4 alleles but no increase (HR: 1.01; 95% CI: 0.91, 1.12) among participants with one or two APOE ε4 alleles. Likewise, elevated cfPWV yielded HRs for dementia of 1.54 (95% CI: 1.30, 1.81) and 1.12 (95% CI: 0.89, 1.42) among participants with zero and one or two APOE ε4 alleles, respectively. Following application of IPCW, the interactions for standardized structural PWV and elevated structural PWV were no longer significant (eFigure 10 in the Supplement). The presence of one or more APOE ε4 alleles was also a statistically significant moderator of the relationship between standardized cfPWV, standardized total PWV, and standardized load‐dependent PWV on the additive scale in unweighted models but not weighted models (eFigures 9 and 10 in the Supplement). Notably, Kaplan‐Meier curves showed a lower percentage of dementia‐free survival after approximately 10 years of follow‐up among those with elevated cfPWV, regardless of APOE ε4 ε4 allele status, with the lowest dementia‐free survival among those with one or two APOE ε4 alleles and elevated cfPWV (eFigure 11 in the Supplement).

Race, sex, education, and baseline hypertension status did not exhibit consistent patterns of statistically significant interactions on either the multiplicative or additive scale in the fully adjusted models (eFigures 12–23 in the Supplement). Point estimates for the risk of dementia were consistently higher among Black participants and frequently higher among participants who had obtained a high school degree or its equivalent.

4. DISCUSSION

In this study, greater arterial stiffness measured through cfPWV was prospectively associated with increased incidence and a greater hazard of all‐cause dementia over a median follow‐up of 9 years. This association remained statistically significant even after adjustment for cardiovascular risk factors such as heart rate, MAP, use of antihypertensive medication, prevalent diabetes, and weighting to address potential differential attrition. Examination of the relationships between the components of PWV and incident dementia revealed that both age‐related structural PWV and blood pressure‐related load‐dependent PWV underlie the relationship between arterial stiffness and dementia risk, signifying a need to target both the structural aspects of arterial stiffness and blood pressure control for dementia prevention. These results expand on previous reports from the ARIC study that elevated cfPWV is cross‐sectionally associated with a higher prevalence of dementia in late life ⁴ by demonstrating a longitudinal association and providing evidence of the mechanisms of increased arterial stiffness that are associated with dementia risk.

Arterial stiffness increases primarily through two mechanisms: age‐related remodeling of the arterial wall, which leads to degradation of elastin content and calcification, ³¹ , ³² and increased pressure load resulting from elevated blood pressure, which promotes the accumulation of excess collagen. ³¹ , ³² , ³³ Evidence from human and animal studies suggests that elevated arterial stiffness is inversely associated with brain health due to increased transmission of pulsatile energy to the microvasculature of the brain. ³¹ , ³² The resulting increased pulsatile load damages cerebral microvessels and reduces cerebral perfusion, ⁸ , ³² , ³⁴ , ³⁵ which in turn contributes to impaired white matter microstructural integrity. ³⁴ , ³⁵ , ³⁶ Disruption of white matter integrity results in axonal disconnection, myelin dysfunction, and neuroinflammation, facilitating neurodegeneration. ³⁷ As neurodegeneration progresses, the resulting loss of functional neural circuits manifests clinically as cognitive decline and dementia. ³⁷ Prior work in ARIC demonstrating that cfPWV is associated with biomarkers of cerebrovascular disease, such as white matter hyperintensity volume ⁷ association with dementia, ³⁸ supports this proposed pathway. ⁹ Building on this evidence, the present findings demonstrate that both structural PWV, which reflects remodeling of the arterial wall, and load‐dependent PWV, which reflects the hemodynamic effects of elevated blood pressure, ⁹ , ¹¹ , ¹⁹ are associated with dementia risk, with the strongest associations observed for load‐dependent PWV.

Prior studies indicated that midlife cfPWV was associated with dementia incidence later in life. Research from the Framingham Heart Study Offspring Cohort found that each SD increment in cfPWV in midlife was associated with a 27% higher hazard (HR: 1.27, 95% CI: 1.06, 1.53) of dementia over 15 years of follow‐up. ³⁹ A study of 532 participants in the Cardiovascular Health Study Cognition Study (mean age 77.8 years, 22.0% Black, 41.0% male) followed over 15 years reported that individuals in the highest quartile of cfPWV had a 57% higher hazard (HR: 1.57, 95% CI: 1.01, 2.42) of dementia compared to those in the lowest quartile of cfPWV. ¹⁴ However, in studies with shorter follow‐up, the association between cfPWV and incident dementia is rarely statistically significant. ¹³ , ⁴⁰ , ⁴¹ For example, in a study of 2767 participants of the Rotterdam Study (mean age 70.7 years, 41.9% male), the association between a SD increment of cfPWV and the hazard of incident dementia was not statistically significant (HR: 0.91, 95% CI: 0.75, 1.10) over 4.4 years of follow‐up. ⁴⁰ Some studies found that adjustment for other vascular risk factors in late life mitigated the association between cfPWV and dementia. In a study of 1101 participants of the Framingham Offspring Study (mean age 69 years, 54% women), higher cfPWV was significantly associated with incident all‐cause dementia in models adjusted for age, sex, education, and the presence of APOE ε4 alleles but not in models additionally adjusted for cardiovascular risk factors. ¹³ The present study expands on previous research by utilizing a large sample size and a long follow‐up period to demonstrate that elevated cfPWV in late life is associated with increased dementia incidence independent of other cardiovascular risk factors such as hypertension, diabetes, heart rate, and MAP.

Few other studies examined the separate components of PWV in relation to dementia. Among 6147 racially/ethnically diverse participants (mean age 62.7 years, 47.3% male) in the Multi‐Ethnic Study of Atherosclerosis (MESA), total PWV was associated with dementia over a median follow‐up of 14.3 years (HR: 1.06; 95% CI: 1.01, 1.12), but structural and load‐dependent PWV were not. ⁹ However, participants in this study had a more favorable cardiometabolic risk profile (diabetes prevalence: 25.4% in ARIC vs 12.4% in MESA; hypertension prevalence: 72.4% in ARIC vs 44.3% in MESA) and a lower mean age compared to our sample. As arterial stiffness is driven primarily by age‐related arterial stiffening and prolonged exposure to elevated blood pressure, ⁹ , ¹¹ , ¹⁹ these differences may have contributed to the discrepant findings. Additional research is needed to confirm our findings that both structural mechanisms and pressure‐related mechanisms underlying arterial stiffness are highly relevant to dementia risk.

The observed moderating effects of age and the presence of APOE ε4 alleles on the association between many PWV measures and dementia risk highlight populations that may be particularly vulnerable to the adverse effects of arterial stiffness on brain health. The increased HRs associated with higher cfPWV, total PWV, and structural PWV among participants younger than 74 years may indicate that younger populations are more susceptible to the adverse cognitive effects associated with arterial stiffness, apart from load‐dependent arterial stiffness. HRs indicate a heightened vulnerability to higher cfPWV, total PWV, structural PWV, and load‐dependent PWV among participants with zero APOE ε4 alleles compared to participants with one or two APOE ε4 alleles, in contrast to findings from Poels et al. that the presence of APOE ε4 alleles does not modify the association between central arterial stiffness and dementia. ⁴⁰ These results may be explained by the elevated baseline dementia risk associated with APOE ε4 allele carriage, which may make detecting additional risk attributable to higher PWV measures more difficult, resulting in the attenuated HRs observed in this group. However, Kaplan‐Meier curves show that, compared with individuals without elevated cfPWV, individuals with elevated cfPWV have a lower percentage of dementia‐free survival regardless of APOE ε4 allele status (eFigure 10 in the Supplement), indicating that interventions to lower central arterial stiffness may benefit all populations, independent of APOE genotype.

This study has several limitations. First, this study may have limited generalizability since the study population included only White or Black participants, and Black participants were primarily recruited from one of the four study sites. Given the 25 years of follow‐up prior to the assessment of PWV, this study may also be vulnerable to informative attrition bias. However, the application of IPCW and multiple imputation by chained equations increases the robustness of the results. ²⁶ , ²⁸ , ²⁹ Finally, due to the lack of information on dementia etiology in the ARIC Study, we are unable to examine whether associations between PWV measures and incident dementia vary across dementia subtypes in clinically meaningful ways.

The strengths of this study include the use of a cfPWV, a gold‐standard measure of arterial stiffness, ³¹ , ⁴² and inclusion of novel measures of total, structural and load‐dependent PWV to provide insights into the components of arterial stiffness that are associated with dementia risk. ⁹ , ¹¹ The ARIC Study also comprehensively measures cognitive outcomes through the administration of multiple cognitive tests, cognitive adjudication by a team of physicians and neuropsychologists, and medical record surveillance to ensure that all cases of dementia are identified. Lastly, in comparison to prior prospective studies, this study includes a larger sample size and extended follow‐up.

5. CONCLUSION

Arterial stiffness, measured through PWV, is longitudinally associated with increased incidence and a greater risk of all‐cause dementia. Associations between arterial stiffness and dementia risk may be driven by both age‐related structural remodeling of the arterial wall and increases in arterial stiffness due to elevated blood pressure levels, especially in older populations with a high burden of cardiometabolic risk factors. While screening for arterial stiffness is not currently standard in clinical practice, future work should continue to examine the associations between arterial stiffness and dementia risk to inform whether arterial stiffness, in combination with other metrics, could serve as a useful tool for dementia risk stratification.

CONFLICT OF INTEREST STATEMENT

Knopman serves on a Data Safety Monitoring Board for the Dominantly Inherited Alzheimer Network Treatment Unit study sponsored by Washington University St. Louis, the SMART‐HS clinical trial (University of Kentucky), the CRANE study (University of Michigan), and for Roche TRONTINEMAB studies WN45443 + WN45447. He receives personal compensation for all. He has served as a consultant for AriBio and Alzeca Biosciences but receives no personal compensation. Matsushita receives research funding from Resolve to Save Lives, Novo Nordisk, Merck, and Huinno and personal fees from Abbott and Fukuda Denshi outside of the submitted work. Barbee, Pike, Wei, Tanaka, Kucharska‐Newton, Sharrett, Hughes, Mosley, Meyer, and Palta report no disclosures of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All ARIC participants provided written informed consent prior to participation.

Supporting information

Supporting Information

Supporting Information