Estimated Pulse Wave Velocity Is Associated With a Higher Risk of Dementia in the Health and Retirement Study

Abstract

BACKGROUND

In this paper, we use the Health and Retirement Study (HRS) to examine the relationship between an estimated measure of pulse wave velocity (ePWV) and cognitive impairment with no dementia and dementia, respectively.

METHODS

We modeled the relationship between ePWV and cognitive status in 2006/2008 using data from 8,492 men and women (mean age 68.6 years) controlling for age, blood pressure, sociodemographic, and socioeconomic characteristics (sex, race and ethnicity, education, income, wealth), health behaviors (smoking and physical activity), body mass index (BMI), health status and related medication use (history of cardiovascular disease, diabetes, and stroke), and cerebrovascular disease (CVD)-related biomarkers (C-reactive protein, cystatin-C, hemoglobin A1c, total cholesterol, high-density lipoprotein [HDL] cholesterol). We assess cognitive function with the 27-item Langa-Weir Telephone Interview for Cognitive Status (TICS) scale. ePWV is derived from an equation based on participant age and resting blood pressure.

RESULTS

In a model that controlled for the constituent components of ePWV (age, age squared, systolic and diastolic blood pressure), ePWV is associated with increased odds of having cognitive impairment with no dementia (OR = 2.761) and dementia (OR = 6.344) relative to a group with no cognitive impairment or dementia. After controlling for the constituent components of ePWV, sociodemographic and socioeconomic characteristics, health behaviors, BMI, health status and medication use, and CVD-related biomarkers, ePWV remains significantly associated with dementia (OR = 3.969) but not cognitive impairment with no dementia (OR = 1.782).

CONCLUSIONS

These findings suggest that ePWV may be a novel research tool and biomarker of vascular aging that can be used in large, population-representative studies to examine cognitive aging and dementia risk.

blood pressure, cognitive aging, ePWV, hypertension, pulse wave velocity, vascular aging, vascular stiffness

Key Points

Question: Is estimated pulse wave velocity (ePWV) associated with the risk of cognitive impairment with no dementia and/or dementia in a large representative sample of middle-aged and older adults in the United States?

Findings: After controlling for age, blood pressure, sociodemographic and socioeconomic characteristics, health behaviors, body mass index, health status and related medication use, and CVD-related biomarkers, ePWV remained significantly associated with dementia but not cognitive impairment with no dementia.

Meaning: A relatively easy-to-derive measure of vascular aging—ePWV—offers novel insight into cognitive aging and may be useful in future population-based studies of dementia risk.

Vascular aging contributes to the development of Alzheimer’s Disease (AD) and Alzheimer’s Disease Related Dementia (ADRD).^1–4 With increasing age, large arteries outside of the brain lose elasticity, which exposes target organs such as the brain to increased pulsatile hemodynamic forces (i.e., increased blood pressure and blood flow pulsatility).⁵ Subsequent barotrauma and ischemia cause target organ damage, which may mediate cognitive decline and dementia risk.^6–11 Arterial stiffening has been demonstrated to be associated with tau pathology, β-amyloid plaque deposits, and cortical atrophy of brain regions involved in the pathophysiology of mild cognitive impairment (MCI) and dementia.^7–9,12 Age-associated arterial stiffening has also been demonstrated to predict cognitive decline^13–15 and transition to dementia.^16–18 As such, arterial stiffness may be useful as a biomarker of vascular aging that predicts age-related cognitive decline and/or dementia.

Carotid-femoral pulse wave velocity (cfPWV) is largely considered the referent (gold standard) measure of arterial stiffness and vascular aging. Measuring cfPWV requires dedicated and highly specialized equipment and technical aptitude, which has limited its widespread adoption into clinical practice and research.¹⁹ However, cfPWV can be indirectly estimated from two commonly measured clinical variables—age and blood pressure. Prior research has found that estimated pulse wave velocity (ePWV) demonstrates construct validity as a measure of vascular aging²⁰ and is an independent predictor of cerebrovascular events^21–24 and all-cause mortality.^21,25 Hao et al.²⁶ recently reported that ePWV independently predicts MCI and probable dementia in the Systolic Blood Pressure Intervention Trial Memory and Cognition in Decreased Hypertension (SPRINT-MIND) Study. Based on the available evidence, ePWV appears to be a promising measure of vascular aging that can be used in large-scale studies of cognitive aging that include direct measurement of blood pressure. To further validate ePWV as a research tool for the study of the relationship between vascular aging and cognitive aging, we examine associations between ePWV and cognitive impairment with no dementia and dementia, respectively, in a large, nationally representative sample of middle-aged and older (51+-year-old) US adults.

STUDY DESIGN

We analyze data from the Health and Retirement Study (HRS), as we have previously described.²⁵ The HRS began enrollment in 1992 (with participants then aged 51–61) and uses a steady-state design, replenishing the sample every 6 years with younger cohorts. Selected sociodemographic variables (i.e., sex, race, ethnicity, education) are measured at study entry. All other variables, including age and blood pressure, are measured in the year the biomarker data were collected. For this study, data were obtained from non-proxy participants over the age of 50 who had no missing values on the biomarkers, which were initially collected from half the sample in 2006 and from the other half of the sample in 2008 (n = 9,098). Participants were excluded if they were missing data on the cognitive assessment, their body mass index (BMI) was <16 kg/m², their systolic or diastolic blood pressure was <40 mm Hg, or they were categorized as “Other” race. The analytic sample for this study includes 8,492 participants.

Cognitive impairment and dementia: primary outcomes

The HRS assesses cognitive function for all participants with a 27-point scale based on the Telephone Interview for Cognitive Status (TICS), which includes: an immediate and delayed 10-noun free recall test; a serial 7 subtraction test; item recognition; and a backward count from 20 test.²⁷ In this study, we use the Langa-Weir classification of cognitive functioning, in which participants scoring ≥6 and <12 out of 27 on the TICS are categorized as having cognitive impairment but no dementia (CIND), while participants scoring <6 are categorized as having dementia (D).²⁸ Scores >12 are used to define the no cognitive impairment, no dementia (NCIND) reference group.²⁸ We used participants’ TICS measurements from either 2006 or 2008, conditional on when their biomarker data were collected.

ePWV: primary independent variable

Systolic and diastolic blood pressures are measured as the average of three consecutive readings in the sitting position with 45-second intervals between readings. For this study, we determined ePWV from the following equation²⁹:

$${\displaystyle \begin{array}{l}{\displaystyle \begin{array}{llll}{\displaystyle \mathrm{e}\mathrm{P}\mathrm{W}\mathrm{V}{\displaystyle =}}& 9.587-(0.402\times \mathrm{A}\mathrm{g}\mathrm{e})+(4.560\times 0.001\times \left(\mathrm{A}\mathrm{g}{\mathrm{e}}^2\right))& -(2.621\times 0.00001\times \left(\mathrm{A}\mathrm{g}{\mathrm{e}}^2\right)\times \mathrm{M}\mathrm{B}\mathrm{P})& {\displaystyle +(3.176\times 0.001\times \mathrm{A}\mathrm{g}\mathrm{e}\times \mathrm{M}\mathrm{B}\mathrm{P})-(1.832\times 0.01\times \mathrm{M}\mathrm{B}\mathrm{P})}\end{array}}\end{array}}$$

In this equation, age is expressed continuously in years, and mean blood pressure (MBP) is calculated from the following equation:

$${\displaystyle \begin{array}{lll}{\displaystyle \mathrm{M}\mathrm{B}\mathrm{P}=}& \mathrm{D}\mathrm{i}\mathrm{a}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{c}\mathrm{B}\mathrm{l}\mathrm{o}\mathrm{o}\mathrm{d}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{u}\mathrm{r}\mathrm{e}& +(0.4\times (\mathrm{S}\mathrm{y}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{c}\mathrm{B}\mathrm{l}\mathrm{o}\mathrm{o}\mathrm{d}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{u}\mathrm{r}\mathrm{e}-\mathrm{D}\mathrm{i}\mathrm{a}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{c}\mathrm{B}\mathrm{l}\mathrm{o}\mathrm{o}\mathrm{d}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{u}\mathrm{r}\mathrm{e})).\end{array}}$$

We used participants’ age and blood pressure measurements from either 2006 or 2008, conditional on when their biomarker data were collected.

Covariates

We estimate models that control for factors that influence vascular and cognitive aging and ADRD risk. Models include controls for sociodemographic and socioeconomic characteristics, health behaviors, BMI, health status and related medication use, and cerebrovascular disease (CVD)-related blood biomarker variables. The sociodemographic variables include: sex (male or female); a combined measure of racial/ethnic identity (non-Hispanic White, non-Hispanic Black, and Hispanic White or Black); and educational attainment (less than high school, high school graduate, associate degree/some college, college degree, graduate degree), which is generally determined early in the life course before vascular aging and cognitive decline, but can influence later-life vascular aging trajectories and cognitive reserve.^30–32 Socioeconomic attainment is measured with two additional variables: household income (in dollars) and household wealth (in dollars). Health behavior variables that influence both vascular and cognitive aging include physical activity (daily or more than once a week, once a week, one to three times a month, hardly ever, or never)³³ and current smoker (yes/no).³⁴ The model also controls for BMI (kg/m²).^35,36 The vascular aging-related health conditions include: history of heart disease (ever been told by a doctor that you have heart problems, including heart attack, ischemic heart disease, and heart failure) (yes/no); history of type 2 diabetes (yes/no); and history of stroke (yes/no). We also include measures of the use of anti-hypertensive medications (yes/no), diabetes medications (yes/no), and cholesterol medications (yes/no).

In addition to the self-reported measures identified above, we include several CVD-related blood biomarkers obtained from dried blood spots via finger prick. Each of these factors has been shown to affect vascular aging and cognitive aging,^37,38 which is why we included them in our model. These biomarkers include a ratio of total cholesterol to high-density lipoprotein (HDL) cholesterol, as a measure of lipid metabolism; glycated hemoglobin A1c, as a measure of glucose control; cystatin-C, as a measure of global kidney function; and C-reactive protein (CRP), as a measure of systemic low-grade inflammation. To account for assay and laboratory variability in the calculation of biomarker values, HRS data are released with the National Health and Nutrition Evaluation Survey (NHANES)-equivalent assay values.³⁹

Statistical analysis plan

After describing the population represented by the sample and the bivariate association between ePWV and cognitive status (NCIND, CIND, D) in 2006/2008, we present the results from multinomial logistic regression models predicting cognitive status. We present three successive models with adjustments for additional covariates in each model. The first model includes ePWV and its constituent elements—measures of systolic and diastolic blood pressure, age, and age-squared. In the second model, we add the sociodemographic variables: sex, race, ethnicity, and education. In the final model, we add the remaining variables in one step because preliminary analyses indicated that additional steps did not yield substantively different conclusions. We conducted all analyses using Stata 18.0. Descriptive results are weighted and the multivariate models include robust standard errors. As recommended by Winship & Radbill and Gelman, we did not weight the multivariate analyses because the variables used to construct the weights are included in the models.^40,41

RESULTS

Table 1 presents descriptive statistics for the population represented by the weighted analytic sample. Overall, the mean age is 68.6 years and the mean ePWV is 12.1 m/s. Approximately 80% are in the NCIND category, 15.6% are in the CIND category, and 4.2% are in the D category. Figure 1 shows a statistically significant association between ePWV and cognitive status (F = 245.79, P < 0.001). Among those in the NCIND category, ePWV is 1132 m/s, on average, while it is 1248 m/s among those in the CIND category and 1346 m/s among those in the D category.

Table 1.

Descriptive statistics of the weighted analytic sample

	Percent	Mean (std. dev.)
Cognitive status, 2006/2008
No cognitive impairment, no dementia (NCIND)	83.56
Cognitive impairment, no dementia (CIND)	13.31
Dementia (D)	3.12
ePWV		11.54 (2.23)
Systolic blood pressure		132.43 (21.12)
Diastolic blood pressure		81.09 (12.26)
Age		65.72 (9.43)
Gender
Male (reference)	43.95
Female	56.05
Race and ethnicity
Non-Hispanic White (reference)	84.38
Non-Hispanic Black	8.21
Hispanic	7.41
Education
<High school (reference)	15.73
High school graduate	53.87
Associate degree/some college	5.09
College degree	14.38
Graduate degree	10.93
Income		74,879 (117,257)
Wealth		559,994 (1,312,990)
Physical activity
More than once a week or daily	58.80
Once a week	15.83
One to three times a month	9.06
Hardly ever or never (reference)	16.31
Current smoker	14.18
Body mass index		29.43 (5.69)
Heart disease	54.43
Heart disease medication	48.71
Diabetes	18.27
Diabetes medication	14.78
Stroke	4.96
Cholesterol medication	39.8
Total cholesterol/HDL ratio		4.00 (1.22)
Glycosylated hemoglobin (HbA1c)		5.80 (0.96)
Cystatin C		1.06 (0.43)
C-reactive protein (CRP)		4.30 (7.68)
n	8,492

Figure 1.

Weighted mean estimated pulse wave velocity (ePWV) in middle-aged and older adults from the Health and Retirement Study stratified by cognitive status.

Table 2 presents the results of a multinomial logistic regression analysis predicting CIND and D, respectively, relative to NCIND. Model 1 includes ePWV and its components. Results indicate that ePWV is associated with a higher relative risk of CIND and D, respectively, compared to NCIND. In Model 2, which adds sex, race, ethnicity, and education, ePWV continues to have a large, positive, statistically significant association with D. The adjusted relative risk ratio (RRR) is 4.387. However, with the addition of the sociodemographic factors to the model, the association between ePWV and CIND becomes marginally non-significant (P < 0.10). Model 3 adds the socioeconomic attainment, health behaviors, BMI, CVD-related health condition and medication use, and CVD-related biomarker variables. In this fully adjusted model, ePWV continues to have a large, positive, statistically significant association with D. The RRR is 3.969. With the addition of these variables into the model, the association between ePWV and CIND becomes non-significant (P > 0.10).

Table 2.

Multinominal logistic regression model predicting cognitive status: 2006/2008

	Model 1				Model 2				Model 3
	CIND vs. NCIND		D vs. NCIND		CIND vs. NCIND		D vs. NCIND		CIND vs. NCIND		D vs. NCIND
	b (s.e.)	RRR	b (s.e.)	RRR	b (s.e.)	RRR	b (s.e.)	RRR	b (s.e.)	RRR	b (s.e.)	RRR
ePWV	1.0155** (0.39)	2.761	1.8475*** (0.60)	6.344	0.6693 (0.41)	1.953	1.4787* (0.61)	4.387	0.5779 (0.41)	1.782	1.3785* (0.62)	3.969
Diastolic blood pressure	−0.0600** (0.02)	0.942	−0.0889*** (0.03)	0.915	−0.0368 (0.02)	0.964	−0.0650* (0.03)	0.937	−0.0267 (0.02)	0.974	−0.0558 (0.03)	0.946
Systolic blood pressure	−0.0233 (0.01)	0.977	−0.0503** (0.02)	0.951	−0.0202 (0.01)	0.980	−0.0475** (0.02)	0.954	−0.0200 (0.01)	0.980	−0.0459* (0.02)	0.955
Age	−0.0849 (0.05)	0.919	−0.1518 (0.09)	0.859	−0.2394** (0.05)	0.842	−0.2394** (0.09)	0.787	−0.1455** (0.05)	0.865	−0.1904* (0.10)	0.827
Age2	−0.0005 (0.0007)	1.000	−0.0009 (0.001)	0.999	0.0007 (0.0008)	1.001	0.0003 (0.001)	1.000	0.00062 (0.0008)	1.001	−0.00002 (0.001)	1.000
Gender (ref = Male)					0.3009*** (0.07)	1.351	0.2207 (0.12)	1.247	0.3862*** (0.07)	1.471	0.3090* (0.13)	1.362
Race/Ethnicity (ref = Non-Hispanic White)
Black					1.2471*** (0.09)	3.480	1.5327*** (0.14)	4.631	1.0911*** (0.09)***	2.978	1.4710*** (0.15)	4.353
Hispanic					0.7030*** (0.11)	2.020	0.6934*** (0.18)	2.000	0.6386*** (0.11)	1.894	0.6568*** (0.19)	1.929
Education (ref ≤ High school)
High school					−1.1379*** (0.08)	0.320	1.8100*** (0.13)	0.164	−1.0105*** (0.08)	0.364	−1.7028*** (0.13)	0.182
Some college					−2.1357*** (0.24)	0.118	−2.2221*** (0.42)	0.108	−1.9324*** (0.25)	0.145	−2.0001*** (0.43)	0.135
College					−2.0746*** (0.15)	0.126	−2.7524*** (0.32)	0.064	−1.7219*** (0.15)	0.179	−2.4494*** (0.33)	0.086
Post-graduate					−2.0482** (0.17)	0.129	−2.6740* (0.37)	0.069	−1.6500*** (0.18)	0.192	−2.357*** (0.40)	0.095
Income ($1000)									−0.0031* (0.00)	1.000	−0.0036 (0.00)	1.000
Wealth ($10,000)									0.0017* (0.00)	1.000	0.0005 (0.00)	1.000
Physical activity (ref = Hardly ever)
More than once a week									−0.2331** (0.09)	0.792	−0.5380*** (0.15)	0.584
Once a week									−0.1255 (0.11)	0.882	−0.3525 (0.18)	0.703
One to three times a month									−0.1740 (0.13)	0.840	−0.5768* (0.24)	0.562
Current smoking (ref = No)
Yes									0.1555 (0.10)	1.168	−0.0681 (0.20)	0.934
BMI									−0.4791** (0.18)	0.619	0.4424 (0.27)	0.643
BMI2									0.0120** (0.01)	1.012	0.0083 (0.01)	1.008
BMI3									−0.00009 (0.00)	1.000	−0.00005 (0.00)	1.000
Heart disease (ref = No)
Yes									0.3482* (0.15)	1.416	0.2500 (0.26)	1.284
Heart disease medication (ref = No)
Yes									−0.3135* (0.14)	0.731	−0.3797 (0.26)	0.684
Diabetes (ref = No)
Yes									0.0107 (0.18)	1.011	0.1272 (0.33)	1.136
Diabetes medication (ref = No)
Yes									0.3439 (0.19)	1.410	0.2510 (0.35)	1.285
Stroke (ref = No)
Yes									0.5024*** (0.13)	1.653	0.8775*** (0.19)	2.405
Cholesterol medication (ref = No)
Yes									−0.0396 (0.07)	0.961	−0.2261 (0.13)	0.798
Total cholesterol/HDL ratio									0.0192 (0.03)	1.024	0.0790 (0.06)	1.075
Glycosylated hemoglobin (HbA1c)									0.0238 (0.04)	1.019	0.0727 (0.06)	1.082
Cystatin C									0.1849** (0.07)	1.203	0.1804 (0.09)	1.198
C-reactive protein (CRP)									0.0083* (0.004)	1.008	0.0018 (0.006)	1.002
Intercept	2.0042 (1.59)		3.1054 (2.94)		4.8727** (1.72)		6.3356* (3.13)		9.4296*** (2.53)		10.8685* (4.32)
n	8,492				8,492				8,492
Log Likelihood	−4,778.9549				−4,220.1609				−4,108.5232

Model 1 includes components of ePWV—age, age-squared, systolic blood pressure, and diastolic blood pressure; Model 2 controls for sex, race, ethnicity, and education; Model 3 includes income, wealth, physical activity, current smoking, BMI, heart disease, heart disease medication, diabetes, diabetes medication, stroke, cholesterol medication, Total cholesterol/HDL ratio, Glycosylated hemoglobin (HbA1c), Cystatin C, and C-reactive protein (CRP).

Abbreviations: D, dementia; CIND, cognitive impairment no dementia; NCIND, no cognitive impairment no dementia. b, beta weight.

*P < 0.05,

**P < 0.01,

***P < 0.001.

DISCUSSION

In this national study of US middle-aged and older (51+-year-old) adults, we found that ePWV is associated with an increased risk of CIND and D relative to NCIND at the bivariate level. This association remained significant when adjusting statistically for the constituent components of ePWV—age, age-squared, and blood pressure. This demonstrates that ePWV is not simply measuring the influences of age and blood pressure on cognitive function. In models that include sociodemographic, socioeconomic attainment, health behaviors, BMI, history of disease and related medication use, and other CVD-related biomarker variables, the association between ePWV and CIND is reduced to non-significance. However, the association between ePWV and D remains large, positive, and statistically significant. Thus, in the HRS, ePWV has an independent, positive predictive value for dementia, relative to a reference group with no cognitive impairment. Taken together, our findings suggest that ePWV may be a useful measure of vascular aging that provides insight into dementia risk.

Our findings support an emerging literature that documents an association between vascular aging, measured as ePWV, and ADRD risk.^26,42–44 However, our findings suggest that ePWV, which indirectly estimates arterial stiffness, may be particularly relevant for the development of dementia. Compared to CIND, dementia is marked by language, behavioral, and sensorimotor impairments that significantly impact independence and quality of life. Such factors may result from or be exacerbated by arterial stiffness. Previous studies have shown that arterial stiffness, measured directly using cfPWV, is independently associated with incident dementia over a 15-year follow-up after adjusting for age, sex, race, education, diabetes, blood pressure, and medication use.¹⁶ Arterial stiffness, measured as cfPWV, is also associated with conversion to dementia in those with CIND, independent of age, sex, educational level, blood pressure, CVD, and medication use.¹⁸ Our findings differ from those of Pase et al.¹⁷ who found that arterial stiffness, measured as cfPWV, was only associated with incident dementia in non-diabetic adults after adjusting for age, sex, education, blood pressure, and chronic disease (i.e., arterial stiffness was not associated with dementia in the total sample in the fully adjusted model). This discrepancy may be related to statistical power; Pase et al. reported n = 77 cases of dementia compared to n = 356 cases in the present study. More research is needed to unpack the complex relationship between arterial stiffness as a measure of vascular aging and ADRD.

Estimated PWV should not be used as a replacement for cfPWV—the gold standard measure of vascular aging—when cfPWV can be measured. However, our results suggest that ePWV is a useful measure of vascular aging in research studies in which blood pressure can be directly measured but cfPWV cannot.

The strengths of this study include the use of a large, representative sample of non-institutionalized US older adults, and the availability of detailed data on sociodemographic, socioeconomic, behavioral, and clinical covariates, including medical conditions and related medication use, and direct measures of blood pressure, lipid, and glucose metabolism (total cholesterol/HDL and HbA1c, respectively), systemic inflammation (CRP), and kidney function (Cystatin C). While this study has many strengths, it also has some limitations that point to directions for future research. We only used blood pressure measured in 2006 or 2008 to calculate ePWV even though blood pressure, and arterial stiffness, change over time.⁴⁵ Future studies that consider the time-varying nature of ePWV can build on the foundations laid in this study and may provide important insights into the dynamics of vascular aging and its consequences for cognitive aging and transitions to dementia.⁴⁶ Additionally, although we adjusted for a wide range of confounders, residual confounding could still exist. Despite these limitations, this study provides novel evidence that ePWV is strongly associated with elevated dementia risk in a representative sample of middle-aged and older adults in the United States, and this association persists in a model that controls statistically for a broad range of other potential influences on dementia risk. Future studies of dementia based on nationally representative data that contain biomarker data (i.e., blood pressure) should consider including ePWV as a measure of vascular aging.

ACKNOWLEDGMENTS

Funding for this study was provided by the Syracuse University Collaboration for Unprecedented Success and Excellence (CUSE) Grant Program. This research benefited from grant, P30AG066583, Center for Aging and Policy Studies, awarded to Syracuse University, in consortium with Cornell University and the University at Albany, by the National Institute on Aging of the National Institutes of Health. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The HRS (Health and Retirement Study) is sponsored by the National Institute on Aging (grant number NIA U01AG009740) and is conducted by the University of Michigan. Data from the HRS are available without cost to registered users. More information can be found at (http://hrsonline.isr.umich.edu).

CONFLICT OF INTEREST

Authors have no conflicts of interest to disclose.