Metabolic and Vascular Risk Factor Variability Over 25 Years Relates to Midlife Brain Volume and Cognition

Abstract

Background:

Metabolic and vascular risk factors (MVRF) are associated with neurodegeneration and poor cognition. There is a need to better understand the impact of these risk factors on brain health in the decades that precede cognitive impairment. Longitudinal assessments can provide new insight regarding changes in MVRFs that are related to brain imaging features.

Objective:

To investigate whether longitudinal changes in MVRF spanning up to 25 years would be associated with midlife brain volume and cognition.

Methods:

Participants were from the CARDIA study (N = 467, age at year 25 = 50.6±3.4, female/male = 232/235, black/white = 161/306). Three models were developed, each designed to capture change over time; however, we were primarily interested in the average real variability (ARV) as a means of quantifying MVRF variability across all available assessments.

Results:

Multivariate partial least squares that used ARV metrics identified two significant latent variables (partial correlations ranged between 0.1 and 0.26, p < 0.01) that related MVRF ARV and regional brain volumes. Both latent variables reflected associations between brain volume and MVRF ARV in obesity, cholesterol, blood pressure, and glucose. Subsequent bivariate correlations revealed associations among MVRF factors, aggregate brain volume and cognition.

Conclusion:

This study demonstrates that MVRF variability over time is associated with midlife brain volume in regions that are relevant to later-life cognitive decline.

Keywords

Brain volume cognition metabolic syndrome partial least squares vascular risk factors

INTRODUCTION

Dyslipidemia, obesity, diabetes, and hypertension are metabolic and vascular risk factors (MVRF) that independently and additively increase the risk of cognitive decline and dementia later in life [1]. There is a need to better understand whether these risk factors influence brain health in the decades that precede cognitive impairment. Longitudinal assessments throughout adulthood can provide new insight on clinical trials aimed at preserving brain health. Furthermore, characterizing the changes in MVRF over time may yield insight on the optimal time frames to intervene.

Cross-sectional studies illustrate associations between MVRF and neuroimaging, as reflected in studies of brain volume, regional cerebral blood flow, metabolites, neuropathology, inflammation, and neuronal activation networks [2 –7]. For example, reduced cerebral blood flow at midlife was related to dysregulated glucose homeostasis, adverse lipid profile, and obesity measures [8]. Visceral adipose fat has also been associated with microstructural changes in grey and white matter regions, as revealed from magnetic resonance imaging (MRI) [9]. In addition, higher total cholesterol is related to poorer brain function [10], greater amyloid burden, and lower hippocampal volume [11]. Greater vascular risk also relates to increased tau burden and cognitive decline in preclinical Alzheimer’s disease [12, 13].

Longitudinal investigations of MVRF and neuroimaging help to reveal important challenges regarding preserving brain health. One randomized clinical trial revealed that intensive treatment of blood pressure did not reduce the risk of cognitive impairment among older adults [14]. Another trial showed that treating blood pressure delayed the progression of white matter lesions, yet the authors also reported no improvement in cognitive and mobility outcomes [15]. Others observed loss of brain volumes among hypertensive individuals despite successful blood pressure control [16]. A large, 10-year longitudinal study showed that among various MVRFs, elevated blood pressure was associated with higher risk of dementia [17]. Moreover, greater blood pressure visit-to-visit variability was associated with poorer future performance on cognitive testing at midlife [18], as well as lower hippocampal volume and tissue integrity from diffusion MRI [19]. Higher 5-year variability in body mass index (BMI) correlated with the lower volume of the hippocampus among older adults [20]. Larger visit-to-visit fasting glucose variability in young adulthood was related to lower midlife hippocampal volume [21]. Lastly, low-density lipoprotein cholesterol (LDL-C) variability in older adults was associated with lower cognitive performance, lower cerebral blood flow, and a higher burden of cerebrovascular disease [22]. Collectively, these studies suggest complex relationships between MVRF change and brain measures, thus a comprehensive assessment of variability in a panel of MVRF with respect to multiple brain regions is warranted.

Here we report on adults that have been studied for over 25 years as part of the ongoing Coronary Artery Risk Development in Young Adults (CARDIA) study [23]. We investigate whether repeat assessments of MVRF can be used to explain between-participant differences in brain volume and cognition outcomes measures from a follow-up visit at midlife. This research extends the literature by using obesity, blood pressure, glucose, and lipids as explanatory variables in a multivariate model that considers the regional neuroimaging estimates simultaneously. Modeling data that are highly correlated is limited when restricted to univariate methods. Individual MVRFs typically do not uniquely explain variance because there can be modest or even high correlation between variables, e.g., BMI, waist circumference, fasting glucose. We further evaluated these associations in terms of cognitive performance at midlife. Our hypothesis is that elevated MVRF variability over time is related to reduced regional cortical brain volumes, particularly in dementia-related regions. Secondarily, we hypothesize that metrics that reflect aggregate brain volume and MVRF variability will relate to poor cognitive performance.

MATERIALS AND METHODS

Participants

Data were obtained from the CARDIA study which is a bi-racial (i.e., Black and White people) longitudinal study of the development and correlates of cardiovascular disease in adults who were aged 18–30 years at baseline in 1985–1986 and have been followed for several years. Between 2010 and 2011, a CARDIA subsample underwent MRI scanning as part of the year-25 assessments to characterize brain structure and function. Participants provided written informed consent at each exam, and institutional review boards from each field center and the coordinating center (The University of Alabama Birmingham Institutional Review Board, University of Minnesota Institutional Review Board, Kaiser Permanente Northern California Institutional Review Board) approved this study annually. Participants were included if they had undergone all of the following: one brain MRI and one cognitive assessment at year 25, and at least two MVRF assessments at year 25 and earlier. There were no missing data entries for the participants that were included. Thus, the total sample was N = 467 adults.

Clinical assessments

MVRF were obtained at multiple visits over 25 years as illustrated in the subplots of Fig. 1 for each variable. MVRF were characterized by: 1) BMI and waist circumference (WC); 2) diastolic blood pressure (DBP) and systolic blood pressure (SBP); 3) fasting glucose (FG: participants were instructed to fast and abstain from smoking or heavy physical exertion for 12 hours before the blood draw); 4) high-density lipoprotein cholesterol (HDL-C), LDL-C, and triglyceride (TRIG) concentrations. The maximum number of MVRF assessments was 8 (for BMI and lipids) and the minimum was 2 (for SBP and DBP).

Fig. 1

A visual display of the metabolic and vascular risk factors (MVRF) variables. The y-axis for each graph shows the calculated visit-to-visit variability (ARV: average real variability) for each variable (used in partial least squares (PLS) model 1). On the x-axis are the per participant mean estimates across time for each MVRF (referred to as “level” in PLS model 3). The inset graphs show data at each assessment as a function of time (in study years) for each MVRF. Rates of change can be deduced from the insert raw data, showing all of the observations at each visit.

We considered three approaches to characterize variability in MVRF over time. The primary approach used the average real variability (ARV) for each MVRF variable, which effectively summates the absolute visit-to-visit changes in each MVRF. This approach was proposed previously [19, 21] and defined by the following:

$\begin{matrix} Average real variability = \frac{1}{\sum_{i = 2}^{n} (t_{i} - t_{i - 1})} \\ \sum_{i = 2}^{n} (t_{i} - t_{i - 1}) \times | M_{i} - M_{i - 1} | \end{matrix}$ (1) where t_i refers to the time at visit i, M_i is the measurement at visit i, | | denotes the absolute value, and n is the total number of visits.

The second method to characterize MVRF variability was a linear rate of change. We calculated an average rate of change estimate for each parameter, which was based on the difference for the successive visits:

$\begin{matrix} Rate of change = \\ average (\frac{M_{i} - M_{i - 1}}{t_{i} - t_{i - 1}}) \end{matrix}$ (2) where t_i refers to the time at visit i, M_i is the measurement at visit i.

The third method to characterize MVRF variability was a normalized ARV. We calculated a normalized version of this variability by dividing the measure by the MVRF level over time. The goal of the normalized change model was to remove the effect of MVRF level from the variability measures and was defined as the following: $Normalized ARV = ARV / level$ (3)

Cognitive assessments

Three cognitive tests were performed at year 25 (i.e., the same visit as MRI) [24]. 1) Verbal memory was assessed using The Rey Auditory Verbal Learning Test. The memory score was computed by summing the five learning trials, and the two free recall trials, the first after a distractor list, and the second after a 30-min delay. 2) Processing speed was assessed using the Digit Symbol Substitution Test. The score was calculated as the number of correct symbols drawn in 120 s. 3) To assess executive function, we used the inhibition trial from the Stroop test. Specifically, we used the time to complete the incongruent condition in which participants are required to name the color of the ink of printed color words as fast as possible.

Brain MRI acquisition and processing

MRI included T1-weighted anatomical imaging that was acquired on 3T MRI system using a 3-dimensional MPRAGE sequence that was prescribed along the sagittal plane and with isotropic resolution, as reflected by the following parameters: voxel = 1x1x1 mm³, TR/TE/TI = 1900/2.9/900 ms, matrix = 256x256, slices = 176, FOV = 250 mm, flip angle = 90, GRAPPA = 2, and bandwidth = 170 Hz/pixel. MRI data were processed using SPM8 (https://www.fil.ion.ucl.ac.uk/spm/software/spm8/) and in-house programs developed in MATLAB (Mathworks Inc., Natick, MA). Brain volumes were extracted for a total of 93 regions of interest (ROIs) using a Talairach-based brain anatomy template and a previously developed segmentation algorithm [25]; from which 80 ROI in grey matter (cortical and subcortical) were included in our study. We adjusted these measures for age, sex, race, and intracranial volume using an analysis of covariance. These adjusted ROI volumes were then used in multivariate analyses as described below.

Statistical assessments

Clinical and cognitive variables were summarized using counts or mean/standard deviation and range. We used partial least squares (PLS) to investigate associations between the multivariable MVRF variability measures and multivariate brain volumes estimates. The following PLS models were performed:

PLS model 1: Regional brain volumes in relation to the MVRF ARV;

PLS model 2: Regional brain volumes in relation to the MVRF rates of change;

PLS model 3: Regional brain volumes in relation to the MVRF normalized variability estimates.

Fig. 2

The partial correlations between metabolic and vascular risk factors average real variability (MVRF ARV) and the volume of regions of interest (ROIs) that contribute to (A) the first latent variable, and (B) the second latent variable, obtained from the first partial least squares (PLS) model. Partial correlation values with error bars that do not cross zero are significant contributors to the latent variables, as denoted by the red color bars. The significant brain regions for the first and second latent variables are shown in (C) and (D), respectively. A positive correlation denotes an MVRF ARV that was directly related to ROI volumes marked by red color and inversely related to ROI volumes marked by blue color. For example, the first latent variable showed higher ARV was associated with lower cortical and higher subcortical volumes.

We implemented PLS in Matlab (Mathworks Inc., Natick, MA) using software provided by the Rotman Research Institute [26, 27]. Similar to a principal component analysis, PLS is a dimensionality reduction technique utilizing a singular value decomposition of the correlation matrix between the outcome and explanatory variables. Hence, unlike conventional regression analysis, collinearity of the variables can prove to be insightful, encapsulating the shared variance into latent variables. These latent variables are the linear combinations of explanatory variables (e.g., MVRF variability) that maximally explain the variance in the outcome variables (e.g., brain volume). The statistical significance of orthogonal latent variables is evaluated by non-parametric permutation testing. A latent variable was deemed significant if the permutations revealed a p-value threshold of 0.01. The bootstrap resampling determines the variables that reliably contribute to the significant latent variables [27]. As a result, ROI volumes that have a significant partial correlation with MVRF measures were identified. As described previously there is no correction for multiple comparisons for this PLS implementation [27]. We also assessed heteroskedasticity of the MVRF using the Breusch-Pagan test and ‘lmtest’ packagein R.

We also calculated bivariate correlations among MVRF factors, aggregate brain volumes, and cognitive scores. MVRF factors were obtained from a factor analysis that included MVRF measures that contribute to the significant latent variable in the PLS model. The aggregate brain volume was calculated as the weighted average of significant ROIs in each latent variable. Note that brain volumes were adjusted for age, sex, race, and intracranial volume. Statistical analyses (other than PLS) were conducted in R (R version 3.33).

RESULTS

Table 1 summarizes participant characteristics (N = 467). Figure 1 shows the relationship between variability and level over time for each MVRF. Raw MVRF data are also displayed (Fig. 1, inset), from which it was possible to see changes over time as well as the number of assessments.

Table 1

Sample characteristics in this study. The values provided here were assessed at\\ year 25 of the CARDIA study for comparison

Sample (N = 467)	Mean±SD (range) or count
Age (y)	50.6±3.4 (42–56)
Sex (F/M)	232/235
Race (Black/White)	161/306
Intracranial Volume	1221±135 (788–1643)
Body mass index (kg/m2)	28.5±5.7 (18.4–49.7)
Waist circumference (cm)	91.6±13.8 (60.7–138.5)
Diastolic blood pressure (mmHg)	72.5±10.0 (50.8–110.9)
Systolic blood pressure (mmHg)	116.9±12.8 (87.3–186.1)
Fasting glucose level (mg/dL)	96.8±24.7 (44.0–367.0)
High-density lipoprotein (mg/dL)	58.4±17.4 (28.0–152.0)
Low-density lipoprotein (mg/dL)	113.2±32.4 (35.0–282.0)
Triglycerides (mg/dL)	107.8±59.9 (27.0–399.0)
Rey Auditory Verbal Learning (memory)	61.8±15.1 (16–99)
Digit Symbol Substitution Test (processing speed)	70.6±15.9 (27–112)
Stroop test executive function time (s)	43.0±10.8 (24–120)

Brain volume and MVRF change association

PLS model 1 (MVRF ARV): This model revealed two significant latent variables. The first latent variable explained 36.5% of the variance in the correlation matrix (p = 0.001). Figure 2A shows the partial correlations for each of the MVRF ARV metrics; this latent variable was derived from 6 MVRF that contributed a meaningful partial correlation, which is based on bars with error bars that do not cross zero. Figure 2C shows the brain ROIs that reliably contributed to this latent variable according to the bootstrap resampling (see Supplementary Table 1). Figure 2B shows a second significant latent variable that explained 24.6% of the variance (p = 0.001) and consisted of 3 MVRF, namely FG, HDL-C, and LDL-C. The brain regions implicated in the second latent variable are shown in Fig. 2D (see Supplementary Table 2). In these figures, ROIs that are depicted with red color are positively associated with partial correlations while blue ROIs are inversely related to the MVRF ARV. For example, a larger insula volume was related to larger variability in MVRF whereas a smaller volume of temporal pole was related to larger MVRF ARV (Fig. 2A, C).

PLS model 2 (MVRF rate of change): This model did not produce any significant latent variables (p > 0.01).

PLS model 3 (MVRF normalized ARV): The normalized ARV estimates yielded one significant latent variable that explained 41% of variance (p < 0.001). The partial correlation and spatial pattern of this latent variable resembled the results from PLS model 1 (Supplementary Table 3). Notably, we determined that MVRFs exhibit heteroskedasticity (p < 0.01) i.e., higher variability for higher levels.

Aggregate scores for MVRF and brain volumes that correlate with cognition

PLS model 1 resulted in two latent variables that were used to address the cognition aim. The first latent variable was converted to an aggregate factor through factor analysis (MVFR Factor 1) that reflected the following sorted MVRFs (with correlation coefficients): WC (r = 0.92), BMI (r = 0.84), DBP (r = 0.39), SBP (r = 0.35), TRIG (r = 0.31), LDL-C (r = 0.23). The second latent was converted to another single factor (MVFR Factor 2) that reflected: LDL-C (r = 0.96), HDL-C (r = 0.31), FG (r = 0.28). Memory was correlated with the MVRF Factor 2 (r = –0.2) but not Factor 1. Processing speed was correlated with both MVRF factors (|r|>0.1). Similarly, executive function was correlated with both MVFR factors (r = 0.1). Aggregate brain volumes were also calculated for latent variables 1 and 2. Memory was not correlated with either aggregate brain volume, whereas processing speed was correlated with both aggregate brain volume (|r| = 0.1). Executive function only correlated with the aggregate brain volume for latent variable 1 (r = –0.1). Bivariate correlations that were expected (i.e., among cognition variables, |r|>0.4, p < 0.001) are also reported in Fig. 3.

Fig. 3

Bivariate correlation coefficients among metabolic and vascular risk factors average real variability (MVRF ARV) factors from the partial least squares (PLS) model 1, aggregate brain volume, and cognitive scores. Only significant correlations (p < 0.01, with legend provided) are reported, leaving entries blank for non-significant bivariate correlations. By definition, diagonal entries have a value of one.

DISCUSSION

This study investigated metabolic and vascular risk factors over the course of adulthood and found the variability across time was associated with between-participant differences in brain volume at midlife. Higher MVRF variability was also related to poorer cognitive performance. The novel elements of this study include the following. First, by studying a sample of adults assessed up until midlife, we demonstrate the value of longitudinal assessments prior to cognitive concerns and/or decline. Second, we demonstrate that the study of MVRF variability over time provides insight that is independent of MVRF mean over time.

The first PLS model showed that MVRF variability was related to aggregate brain volume. Previous reports have primarily focused on an individual MVRF and particular brain regions [20]. For example, higher blood pressure variability is associated with lower hippocampal volume [19], larger ventricular atrophy in older age [28], and cognitive impairment [29]. A recent CARDIA study showed that higher variability in FG was negatively associated with hippocampal volume [21]. LDL-C variability, independent of mean LDL-C levels, has also been associated with cognitive performance and cerebrovascular measures [22], a previous result that is in line with our findings. Numerous brain regions were identified in the PLS results, with regional volume estimates in the temporal lobe inversely related to higher MVRF ARV and subcortical regions positively associated with MVRF variability. Subcortical regions are implicated in the food reward behavior [30] suggesting that higher MVRF variability relates to larger volume within the brain’s rewardsystem.

Longitudinal datasets permit diverse analyses, such as mean across time or variation across time. Several studies investigated the effect of vascular risk burden in midlife on cognitive decline and dementia risk later in life. For example, a recent Framingham Heart study showed that midlife vascular risk burden was predictive of cerebrovascular neuropathology, even after adjusting for late-life vascular risk factors [31]. Another study showed that higher vascular risk factor burden was associated with lower brain volume across each age decade from 45–95 years of age, with the highest associations related to higher MVRF at a younger age [32] and emphasizing the importance of MVRF control during midlife. The healthy brain project also showed a significant association between midlife cardiovascular risk factors (including history of hypertension, hypercholesterolemia, diabetes mellitus, overweight) and late-life cognitive decline in particular memory impairment [33]. History of hypertension [34] and higher cardiovascular risk scores [12] have also been associated with higher cognitive decline laterin life.

The current study adds to the MVRF-brain health literature by investigating how brain volume and cognition are affected with respect to MVRF changes over the adulthood. Among different variability measures, we used average real variability, rather than standard deviation or coefficient of variation, because ARV accounts for the between-visit time intervals, it accounts for increase and decrease equivalently through the absolute value, it is less affected by extreme values [37], and lastly ARV has the same units of the measured parameter. It is noteworthy that the second PLS model, which used MVRF rate of change over time, did not produce any significant latent variable, i.e., visit to visit rate of change did not correlate with regional brain volume. We therefore surmise that MVRF fluctuations across time may be important in relation to brain volume alterations. We also found a significant relationship between MVRF summary measures and cognition, broadly in line with previous research that links variability of risk factors and adverse outcomes. For example, a previous meta-analysis found that long-term blood pressure variability and cholesterol measures have similar effect sizes on predicting cardiovascular events [36]; a subsequent review also implicates brain outcomes in their discussion of visit-to-visit variability of risk factors [35]. In the current study we also found that LDL-C variability contributed to both of the significant latent variables, a finding that aligns with one previous report on LDL-C variability [22].

This study has some limitations. First, the number of assessments was not the same for all the MVRF and this could influence PLS findings (e.g., BMI had the highest number of assessments while blood pressure had the fewest). Caution is warranted since the ARV may introduce some bias in the MVRF with fewer assessments [38]. Estimating variability is susceptible to propagating measurement error, for instance in the case of MVRFs that had low test-retest reliability. Also, our implementation of the normalized ARV (PLS model 3) attempted to control for the MVRF level but may have been suboptimal; specifically, some of the ARV estimates showed a higher spread in the data for higher levels, which is heteroskedasticity. We assumed that the non-parametric PLS model can accommodate these data attributes, but further investigation of modeling choices may be warranted. Other factors can influence MVRF variability and brain volume associations, which we attempted to mitigate by accounting for age, sex, and race prior to the multivariate analysis. Genetics, medications, family history, and pregnancy information were not considered in the current study. Finally, although serial assessments were used to characterize the MVRF, the MRI-based brain volumes were from a single time-point; hence it is not possible to comment on changes in the brain anatomy outcome measures in this quasi-cross-sectional study.

We found a significant association between MVRF variability over 25 years and regional brain volume as well as cognitive performance. This study demonstrates long-term MVRF exposures may afford additional sensitivity to study midlife brain volume in regions that are relevant to late-life cognitivedecline.

Footnotes

ACKNOWLEDGMENTS

The Coronary Artery Risk Development in Young Adults Study (CARDIA) is conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with the University of Alabama at Birmingham (HHSN268201800005I & HHSN268201800007I), Northwestern University (HHSN268201800003I), University of Minnesota (HHSN268201800006I), and Kaiser Foundation Research Institute (HHSN268201800004I) and NIA Intramural Research Program. Dr. Shirzadi received a postdoctoral award from the Alzheimer’s Society of Canada. We acknowledge the Dr. Sandra Black Centre for Brain Resilience & Recovery, MMI Kuwait, and the Kuwait Ministry of Health for funding support.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

van den Berg

, Kloppenborg

, Kessels

RPC

, Kappelle

, Biessels

(2009) Type 2 diabetes mellitus, hypertension, dyslipidemia and obesity: A systematic comparison of their impact on cognition.. Biochim Biophys Acta 1792, 470–481.

Dunn

, Kessler

, Feurer

, Volkow

, Patterson

, Ansari

, Li

, Marks-Shulman

, Abumrad

(2012) Relationship of dopamine type 2 receptor binding potential with fasting neuroendocrine hormones and insulin sensitivity in human obesity. Diabetes Care 35, 1105–1111.

Ottino-González

, Jurado

, García-García

, Segura

, Marqués-Iturria

, Sender-Palacios

, Tor

, Prats-Soteras

, Caldú

, Junqué

, Garolera

(2017) Allostatic load is linked to cortical thickness changes depending on body-weight status. Front Hum Neurosci 11, 639.

Beauchet

, Celle

, Roche

, Bartha

, Montero-Odasso

, Allali

, Annweiler

(2013) Blood pressure levels and brain volume reduction: A systematic review and meta-analysis. J Hypertens 31, 1502–1516.

Schneider

ALC

, Selvin

, Sharrett

, Griswold

, Coresh

, Jack

, Knopman

, Mosley

, Gottesman

(2017) Diabetes, prediabetes, and brain volumes and subclinical cerebrovascular disease on MRI: The Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). Diabetes Care 40, 1514–1521.

Cox

, Lyall

, Ritchie

, Bastin

, Harris

, Buchanan

, Fawns-Ritchie

, Barbu

, de Nooij

, Reus

, Alloza

, Shen

, Neilson

, Alderson

, Hunter

, Liewald

, Whalley

, McIntosh

, Lawrie

, Pell

, Tucker-Drob

, Wardlaw

, Gale

, Deary

(2019) Associations between vascular risk factors and brain MRI indices in UK Biobank. Eur Heart J 40, 2290–2300.

Sala

, de

Roos A

, van den Berg

, Altmann-Schneider

, Slagboom

, Westendorp

, van Buchem

, de Craen

AJM

, van der Grond

(2014) Microstructural brain tissue damage in metabolic syndrome. Diabetes Care 37, 493–500.

MacIntosh

, Shirzadi

, Atwi

, Detre

, Dolui

, Bryan

, Launer

, Swardfager

(2020) Metabolic and vascular risk factors are associated with reduced cerebral blood flow and poorer midlife memory performance. Hum Brain Mapp 41, 855–864.

Widya

, Kroft

LJM

, Altmann-Schneider

, van den Berg-Huysmans

, van der Bijl

, de Roos

, Lamb

, van Buchem

, Slagboom

, van Heemst

, van der Grond

, Leiden Longevity Study Group (2015) Visceral adipose tissue is associated with microstructural brain tissue damage. Obesity (Silver Spring) 23, 1092–1096.

10.

Meusel

LAC

, Anderson

, Parrott

, Yuen

, Tchistiakova

, MacIntosh

, Feldman

, Greenwood

(2017) Brain function is linked to LDL cholesterol in older adults with cardiovascular risk, J Am Geriatr Soc 65, e51–e55.

11.

Lee

, Parekh

, King

, Reed

, Chui

, Krauss

, Yassine

(2019) Low-density lipoprotein particle size subfractions and cerebral amyloidosis. J Alzheimers Dis 68, 983–990.

12.

Rabin

, Schultz

, Hedden

, Viswanathan

, Marshall

, Kilpatrick

, Klein

, Buckley

, Yang

, Properzi

, Rao

, Kirn

, Papp

, Rentz

, Johnson

, Sperling

, Chhatwal

(2018) Interactive associations of vascular risk and β-amyloid burden with cognitive decline in clinically normal elderly individuals findings from the Harvard Aging Brain Study. JAMA Neurol 75, 1124–1131.

13.

Rabin

, Yang

, Schultz

, Hanseeuw

, Hedden

, Viswanathan

, Gatchel

, Marshall

, Kilpatrick

, Klein

, Rao

, Buckley

, Yau

WYW

, Kirn

, Rentz

, Johnson

, Sperling

, Chhatwal

(2019) Vascular risk and β-amyloid are synergistically associated with cortical tau. Ann Neurol 85, 272–279.

14.

Williamson

, Pajewski

, Auchus

, Bryan

, Chelune

, Cheung

, Cleveland

, Coker

, Crowe

, Cushman

, Cutler

, Davatzikos

, Desiderio

, Erus

, Fine

, Gaussoin

, Harris

, Hsieh

, Johnson

, Kimmel

, Tamura

, Launer

, Lerner

, Lewis

, Martindale-Adams

, Moy

, Nasrallah

, Nichols

, Oparil

, Ogrocki

, Rahman

, Rapp

, Reboussin

, Rocco

, Sachs

, Sink

, Still

, Supiano

, Snyder

, Wadley

, Walker

, Weiner

, Whelton

, Wilson

, Woolard

, Wright

(2019) Effect of intensive vs standard blood pressure control on probable dementia: A randomized clinical trial. JAMA 321, 553–561.

15.

White

, Wakefield

, Moscufo

, Guttmann

CRG

, Kaplan

, Bohannon

, Fellows

, Hall

, Wolfson

(2019) Effects of intensive versus standard ambulatory blood pressure control on cerebrovascular outcomes in older people (INFINITY). Circulation 140, 1626–1635.

16.

Jennings

, Mendelson

, Muldoon

, Ryan

, Gianaros

, Raz

, Aizenstein

(2012) Regional grey matter shrinks in hypertensiveindividuals despite successful lowering of blood pressure. JHum Hypertens 26, 295–305.

17.

Malik

, Georgakis

, Neitzel

, Rannikmäe

, Ewers

, Seshadri

, Sudlow

CLM

, Dichgans

(2021) Midlife vascular risk factors and risk of incident dementia: Longitudinal cohort and Mendelian randomization analyses in the UK Biobank. Alzheimers Dement 17, 1422–1431.

18.

Yano

, Ning

, Allen

, Reis

, Launer

, Liu

, Yaffe

, Greenland

, Lloyd-Jones

(2014) Long-term blood pressure variability throughout young adulthood and cognitive function in midlife. Hypertension 64, 983–988.

19.

Yano

, Reis

, Levine

, Bryan

, Viera

, Shimbo

, Tedla

, Allen

, Schreiner

, Bancks

, Sidney

, Pletcher

, Liu

, Greenland

, Lloyd-Jones

, Launer

(2017) Visit-to-visit bloodpressure variability in young adulthood and hippocampal volume andintegrity at middle age: The CARDIA Study (Coronary Artery RiskDevelopment in Young Adults). Hypertension 70, 1091–1098.

20.

Bobb

, Schwartz

, Davatzikos

, Caffo

(2014) Cross-sectional and longitudinal association of body mass index and brain volume. Hum Brain Mapp 35, 75–88.

21.

Xiong

, Li

, Zhong

, Zhang

, Sun

, Zhou

, Chen

, Zhuang

, Liao

(2019) Visit-to-visit fasting glucose variability in young adulthood and hippocampal integrity and volume at midlife. Diabetes Care 42, 2334–2337.

22.

Smit

RAJ

, Trompet

, Sabayan

, le Cessie

, van der Grond

, van Buchem

, de Craen

AJM

, Jukema

(2016) Higher visit-to-visit low-density lipoprotein cholesterol variability is associated with lower cognitive performance, lower cerebral blood flow, and greater white matter hyperintensity load in older subjects. Circulation 134, 212–21.

23.

Friedman

, Cutter

, Donahue

, Hughes

, Hulley

, Jacobs

, Liu

, Savage

(1988) Cardia: Study design, recruitment, and some characteristics of the examined subjects. J Clin Epidemiol 41, 1105–1116.

24.

Launer

, Lewis

, Schreiner

, Sidney

, Battapady

, Jacobs

, Lim

, D’Esposito

, Zhang

, Reis

, Davatzikos

, Bryan

(2015) Vascular factors and multiple measures of early brain health: CARDIA Brain MRI Study, PLoS One 10, e0122138.

25.

Goldszal

, Davatzikos

, Pham

, Yan

MXH

, Bryan

, Resnick

(1998) An image-processing system for qualitative and quantitative volumetric analysis of brain images. J Comput Assist Tomogr 22, 827–837.

26.

McIntosh

, Bookstein

, Haxby

, Grady

(1996) Spatial pattern analysis of functional brain images using partial least squares. Neuroimaging 3, 143–157.

27.

McIntosh

, Lobaugh

(2004) Partial least squares analysis of neuroimaging data: Applications and advances. Neuroimage 23, 250–263.

28.

Havlik

, Foley

, Sayer

, Masaki

, White

, Launer

(2002) Variability in midlife systolic blood pressure is related to late-life brain white matter lesions: The Honolulu-Asia Aging study. Stroke 33, 26–30.

29.

Sabayan

, Wijsman

, Foster-Dingley

, Stott

, Ford

, Buckley

, Sattar

, Jukema

, van Osch

MJP

, van der Grond

, van Buchem

, Westendorp

RGJ

, de Craen

AJM

, Mooijaart

(2013) Association of visit-to-visit variability in blood pressure with cognitive function in old age: Prospective cohort study, BMJ 347, f4600.

30.

Alonso-Alonso

, Woods

, Pelchat

, Grigson

, Stice

, Farooqi

, Khoo

, Mattes

, Beauchamp

(2015) Food reward system: Current perspectives and future research needs. Nutr Rev 73, 296–307.

31.

Conner

, Pase

, Carneiro

, Raman

, McKee

, Alvarez

, Walker

, Satizabal

, Himali

, Stein

, Beiser

, Seshadri

(2019) Mid-life and late-life vascular risk factor burden and neuropathology in old age. Ann Clin Transl Neurol 6, 2403–2412.

32.

Pase

, Davis-Plourde

, Himali

, Satizabal

, Aparicio

, Seshadri

, Beiser

, DeCarli

(2018) Vascular risk at younger ages most strongly associates with current and future brain volume, Neurology 91, e1479–e1486.

33.

Yassi

, Pase

, Buckley

, Rosenich

, Watson

, Maruff

, Lim

(2022) Cardiovascular risk associated with poorer memory in middle-aged adults from the Healthy Brain Project. J Alzheimers Dis 86, 1081–1091.

34.

Gottesman

, Wu

, Coresh

, Jack

, Knopman

, Rahmim

, Sharrett

, Wagenknecht

, Walker

, Wong

, Mosley

(2020) Abstract WP479: Associations between vascular risk, amyloid burden and incident dementia: The ARIC-PET Study, Stroke 51, AWP479.

35.

Simpson

(2019) Biomarker variability and cardiovascular disease residual risk. Curr Opin Cardiol 34, 413–417.

36.

Stevens

, Wood

, Koshiaris

, Law

, Glasziou

, Stevens

, McManus

(2016) Blood pressure variability and cardiovascular disease: Systematic review and meta-analysis, BMJ 354, i4098.

37.

Yano

(2017) Visit-to-visit blood pressure variability— What is the current challenge? Am J Hypertens 30, 112–114.

38.

Mena

, Maestre

, Hansen

, Thijs

, Liu

, Boggia

, Li

, Kikuya

, Björklund-Bodegård

, Ohkubo

, Jeppesen

, Torp-Pedersen

, Dolan

, Kuznetsova

, Stolarz-Skrzypek

, Tikhonoff

, Malyutina

, Casiglia

, Nikitin

, Lind

, Sandoya

, Kawecka-Jaszcz

, Filipovsky

, Lmai

, Wang

, O’Brien

, Staessen

, International Database on Ambulatory Blood Pressure in Relation to Cardiovascular Outcomes (IDACO) Investigators (2014) How many measurements are needed to estimate blood pressure variability without loss of prognostic information? Am J Hypertens 27, 46–55.