Complex relationships of socioeconomic status with vascular and Alzheimer’s pathways on cognition

Graphical abstract

Highlights

• •AD and CVD often co-exists and are leading causes of age-related cognitive decline.
• •We used SEM modeling to study the effects of vascular and AD pathways on cognition.
• •Higher SES had a protective effect on cognition with some mediation through the vascular pathway.
• •There is a significant effect of vascular risk on tau deposition.
• •CVD and AD biomarkers capture the relative effects of these pathways on cognition.

Abstract

Introduction

AD and CVD, which frequently co-occur, are leading causes of age-related cognitive decline. We assessed how demographic factors, socioeconomic status (SES) as indicated by education and occupation, vascular risk factors, and a range of biomarkers associated with both CVD (including white matter hyperintensities [WMH], diffusion MRI abnormalities, infarctions, and microbleeds) and AD (comprising amyloid-PET and tau-PET) collectively influence cognitive function.

Methods

In this cross-sectional population study, structural equation models were utilized to understand these associations in 449 participants (mean age (SD) = 74.5 (8.4) years; 56% male; 7.5% cognitively impaired).

Results

(1) Higher SES had a protective effect on cognition with mediation through the vascular pathway. (2) The effect of amyloid directly on cognition and through tau was 11-fold larger than the indirect effect of amyloid on cognition through WMH. (3) There is a significant effect of vascular risk on tau deposition.

Discussion

The utilized biomarkers captured the impact of CVD and AD on cognition. The overall effect of vascular risk and SES on these biomarkers are complex and need further investigation.

1. Introduction

Cognitive decline in older adults is multifactorial, associated with cerebrovascular changes, often coexisting with Alzheimer’s Disease (AD) neuropathology (Iadecola, 2013, Scheltens et al., 2016, Bos et al., 2019). The presence of both pathologies increases the risk of cognitive impairment compared to the presence of one pathology (Vemuri and Knopman, 2016, Vemuri et al., 2015). The hallmarks of cerebrovascular disease are the presence of microvascular changes (WMH), disrupted white matter integrity, macrovascular damage (infarcts), and cerebral microbleeds (Madden et al., 2008, Cannistraro et al., 2019). The hallmarks of AD neuropathologic changes are the presence of amyloid-β plaques and neurofibrillary tangles. Vascular dysfunction has gained traction as an important contributor to AD pathophysiology, but the evidence linking vascular factors to tau is less conclusive than vascular factors to amyloid-β pathology (Albrecht et al., 2020). The relationship between age, vascular influences, and amyloid-beta and tau pathology is intricate, and has a substantial impact on cognitive function. These factors are interconnected, with age-related changes potentially interacting with vascular conditions and neurodegenerative processes, collectively shaping an individual's cognitive trajectory. (Vemuri et al., 2015, Rabin et al., 2022, Schneider et al., 2004, Schneider et al., 2007).

Previous research indicates that Social and Structural Determinants of Health (SSDoH), including socio-economic status (SES), have a significant influence on age-related cognitive decline, indicating that individuals with higher SES have a reduced risk of cognitive impairment (Plassman et al., 2010, Karp et al., 2004, Peters et al., 2009, Roe et al., 2008). SES is a multidimensional construct that encompasses various aspects of an individual's or household's economic and social well-being (Oakes and Rossi, 2003). Education and occupation are commonly used indicators of SES because they are often associated with income, lifestyle, and access to resources. Numerous factors may play a role in its effects on cognition, including healthcare access, nutrition, chronic stress levels, early brain development, educational opportunities, and cultural capital (Hazzouri et al., 2013, Wortmann, 2014). While previous research suggests that different measures of vascular risk factors and SES may provide useful tools for identifying individuals at risk for cognitive decline and dementia (Hazzouri et al., 2013, Wortmann, 2014, Kivimäki et al., 2020, Cadar et al., 2018), further research is needed to determine how SES with CVD and AD collectively predict cognitive outcomes.

The primary goal of the present study was to examine if SES, vascular risk factors, and imaging biomarkers of AD and cerebrovascular pathology are associated with cognition in a population-based sample. While previous studies have shown associations between vascular risk factors and AD neuropathology using various vascular risk factor scores (Rabin et al., 2018, Yu et al., 2022), we incorporated both a vascular risk score and neuroimaging structural components in our model. The aim of our study was to disentangle the associations between the vascular and neuroimaging variables and their influence on cognition by examining the direct and indirect effects of each of these predictors, as detailed below. We organized the variables as follows: Demographics (age and sex) were considered foundational factors and placed at level 1. SES variable, which reflects education during early life and occupation during adult life, followed demographics. Vascular risk variables were positioned following SES and demographics, as these factors typically emerge in mid- to late-life and can affect cognitive function. Amyloid deposition was placed before tau deposition and cognition. This ordering is based on extensive evidence showing that significant amyloid deposition typically occurs before tau deposition, (Aisen et al., 2017, Frisoni et al., 2022, Jack et al., 2018). This has been consistently shown in both sporadic AD studies (Jack et al., 2018) and autosomal dominant AD studies (Bateman et al., 2012). CVD biomarkers were placed after the demographics, SES and vascular risk variables (which were measured in the five-year period prior to the MRI scan). The decision to position the CVD variables after amyloid deposition was based on evidence indicating that the progression of CVD variables, particularly WMH, occurs before the onset of cognitive symptoms. (Dadar et al., 2020, Schoemaker et al., 2022). Tau pathology is often associated with later stages of disease progression compared to neurovascular pathology, as it correlates more closely with cognitive decline, severity of cognitive impairment and neurodegeneration (Laing et al., 2020), and therefore was placed at level 6.

2. Materials and methods

2.1. Selection of participants

Participants were enrolled in the Mayo Clinic Study of Aging (MCSA), a population-based study examining the epidemiology of cognitive decline and risk of mild cognitive impairment (MCI) among residents living in Olmsted County, Minnesota, as previously described (Roberts et al., 2008). Briefly, residents are randomly selected for recruitment using the Rochester Epidemiology Project medical records linkage system (St Sauver et al., 2012). Participants underwent detailed clinical visits, including neuropsychological testing, a physician examination, and blood draws every 15 months, and a subset participated in neuroimaging. The current analyses included 449 participants older than 60 who had undergone MRI, amyloid- and tau-PET (Fig. 1). The study was approved by Mayo Clinic and Olmsted Medical Center Institutional Review Boards. Written informed consent was obtained from all participants.

Fig. 1.

Flowchart illustrating the inclusion/exclusion of individuals from the MCSA in the study.

2.2. Assessment of cognitive function and clinical diagnosis

Neuropsychological testing was administered by trained psychometrists and included nine tests assessing four domains: memory, language, executive function, and visuospatial skills. To calculate the global cognition z-score, we followed the procedure previously described (Vemuri et al., 2021). First, individual test scores were converted into z-scores for cognitively unimpaired individuals older than 50 years of age. Then, a summary score was computed by taking the z-transform of the mean of the four domain z scores and then weighted back to the Olmsted County population.

A clinical consensus committee assessed each study participant to determine cognitive diagnoses blinded to the diagnosis of the previous study visit and neuroimaging results. A final diagnosis was made after considering education, occupation, visual or hearing deficits, and reviewing all other participant information. Participants who performed in the normal range and did not meet the criteria for MCI or dementia were deemed cognitively unimpaired. The MCSA design and clinical diagnoses criteria were discussed in detail by Petersen et al. (Petersen et al., 2010) and Roberts et al. (Roberts et al., 2008).

2.3. Covariates

Demographic information included self-reported age and sex. SES was evaluated based on self-reported education years and occupation (Vemuri et al., 2014) which was captured at the baseline visit. Otherwise, all other variables are associated with the clinical visit. The education/occupation is a composite score based on a previous principal component analysis, described in detail in Vemuri et al (Vemuri et al., 2014). Participants self-reported their educational attainment, while occupation was determined by the primary job individuals held for the majority of their adult years. Occupational classifications were assigned ordinal values ranging from 0 to 5 reflecting similarities in job characteristics and complexity. As previously described (Vemuri et al., 2021, Vemuri et al., 2017), we selected seven vascular risk factors known to affect cognition and vascular brain health: hypertension, hyperlipidemia, cardiac arrhythmias, coronary artery disease, congestive heart failure, diabetes mellitus, and stroke. Clinical data, including vascular risk factors, were abstracted from the medical records by a nurse. A cerebral vascular risk score was calculated with a possible range of 0–7, which represents the summation of the presence or absence of each of the specified conditions (Vemuri et al., 2017, Rocca et al., 2014).

2.4. Imaging

2.4.1. White matter hyperintensities

A trained imaging analyst segmented and edited WMH on two-dimensional FLAIR images by using a semi-automated method, as explained previously (Graff-Radford et al., 2019, Graff-Radford et al., 2019). FLAIR images were used to identify possible WMH voxels using a clustering method: a sphere is placed around each identified voxel to make it visible on 3D rendering and overlapping spheres can merge, forming clusters. The FLAIR images were then aligned with the T1-weighted image using SPM5 segmentations. Voxels associated with infarcts were removed and not included in the WMH measurement. Areas suspected as WMH masks were also removed if they occurred outside of the white matter area, if they consisted of a single isolated voxel, or if they appeared without parallel abnormality seen on FLAIR. A normalization and smoothing process was applied. The WMH was analyzed as a volume percentage of total intracranial volume.

2.4.2. White matter integrity assessment (Genu-FA)

The diffusion MRI data were processed using an in-house developed pipeline. After denoising(Veraart et al., 2016), but modified to work with interpolated data) the images, head motion and eddy current distortion was corrected using FSL’s eddy program (Andersson and Sotiropoulos, 2016, Andersson et al., 2016, Andersson et al., 2017). We corrected for Gibbs ringing as described in Kellner et al. (Kellner et al., 2016) and then skull stripped the images (Reid et al., 2018). The Rician noise bias was then removed using the noise image from denoising and the procedure outlined in (Koay et al., 2009). Diffusion tensors were estimated using nonlinear least squares fitting and used to calculate Fractional Anisotropy (FA) and Mean Diffusivity (MD) images in dipy (Garyfallidis et al., 2014). ANTS (Avants et al., 2014) was used to nonlinearly register a modified version of the JHU “Eve” WM atlas (Oishi et al., 2009) to each subject’s FA image. The atlas was modified by fusing the left and right portions of regions crossing the midplane, namely the corpus callosum, pons, and fornix. Voxels with MD > 2 x 10–3 or < 7 x 10–5 mm2/s were excluded as mostly CSF or air, respectively, and the median FA and MD were calculated in each region of interest (ROI). The cuneus, precuneus, fusiform, and lingual WM regions were excluded since they were too small for reliable registration. Fractional anisotropy measurement in the genu of the corpus callosum (Genu-FA) was chosen because it captures both variability in systemic vascular health as well as visible cerebrovascular injury in the form of WMH (Vemuri et al., 2018). Genu FA values range from 0 to 1, where 0 indicates isotropic diffusion and 1 indicates purely anisotropic diffusion (higher FA values indicate greater white matter integrity).

2.4.3. Cerebral microbleeds (CMBs)

CMBs were recognized and computed as previously described (Graff-Radford et al., 2019), in agreement with consensus criteria (Greenberg et al., 2009), as homogeneous hypointense lesions in the gray or white matter, which are distinct from vessel flow voids on T2* GRE images. All possible CMBs were marked by trained image analysts and subsequently confirmed by a vascular neurologist or radiologist experienced in reading T2* GRE who did not know the participants’ clinical information. We then calculated the total number of microbleeds as a sum of deep, lobar, and cerebellum microbleeds for each participant and included these in the analysis.

2.4.4. Infarcts

Recognition and grading of infarcts was done as previously described (Graff-Radford et al., 2020) using FLAIR MRI co-registered to the MPRAGE scan. All possible infarcts were initially identified by trained image analysts and subsequently confirmed by a vascular neurologist who was blinded to all clinical information. A total number of infarcts which was a sum of cortical and subcortical infarcts was included in the analyses (Vemuri et al., 2021).

2.4.5. PET imaging

PET imaging was acquired using previously described methods (Graff-Radford et al., 2019). Amyloid-PET imaging was performed with 11C-PiB; Tau PET was performed with AV-1451, synthesized on site with precursor supplied by Avid Radiopharmaceuticals (Avid Radiopharmaceuticals, Philadelphia, PA). Late-uptake amyloid-PET images were obtained 40–60 min and tau-PET 80–100 min after injection. CT was used for attenuation correction. Amyloid- and tau-PETs were analyzed using our in-house fully automated image-processing pipeline, where image voxel values were extracted from automatically labelled regions of interest propagated from the MCALT template (Senjem et al., 2005). The standardized uptake value ratio labelled regions of interest propagated from the MCALT template (Senjem et al., 2005). The standardized uptake value ratio (SUVR) values of amyloid- and tau-PET were determined by normalizing target regions of interest to the cerebellar crus grey matter (Jack et al., 2017). A cut-point of ≥ 1.48 SUVR was used to categorize participants as amyloid positive (Jack et al., 2017). The amyloid-PET meta region of interest was composed of prefrontal, orbitofrontal, parietal, temporal, anterior and posterior cingulate, and precuneus. Based on a voxel number-weighted average of the median tau-PET uptake in previously published regions of interest, a tau-PET meta-region of interest was formed (Jack et al., 2017) and normalized to the cerebellar crus grey median. The tau-PET meta-region of interest was composed of the entorhinal, amygdala, parahippocampal, fusiform, inferior temporal, and middle temporal regions of interest (ROI). This meta-ROI was chosen because it has been previously used in cognitively unimpaired individuals and increases with age as expected (Jack et al., 2017). Tau PET positivity was defined by a meta-ROI SUVR ≥ 1.25 as previously described (Jack et al., 2019). PET images were quantified using MRI scans of the participants. PET data were not corrected for partial volume.

2.5. Statistical analysis

This is a cross-sectional study. Participants' demographic, clinical, and imaging characteristics were summarized using means and SDs for continuous variables and frequencies and percentages for categorical variables.

Structural equation models (SEMs).

We performed path analyses (structural equation models [SEMs] with only manifest variables), using Mplus version 8.3 software as previously published (Vemuri et al., 2017). Path analyses are extensions of regression models and require ordering of the variables to be specified before running any models. The full model consisted of 7 levels, each level is assumed to come later than preceding levels in a progression: Level 1: sex and age as exogenous predictors; Level 2: education/occupation years (SES); Level 3: vascular risk score; Level 4: continuous amyloid-β deposition (PiB) SUVR; Level 5: components of vascular brain health (WMH, Genu-FA, cerebral microbleeds and infarcts); Level 6: continuous tau deposition SUVR; Level 7: global cognition. Variables in each level could be predicted by any variables in a lower level. The possible associations are direct effects (arrow directly joins variables), indirect effects (arrows pass through one or more mediators), and total effects (sum of direct and indirect). We accounted for potential learning effects through an adjustment for cognition based on the number of clinical visits, which we term 'cycles' completed by each patient. In addition, we evaluated the correlation between Genu-FA and WMH. We report regression coefficients with associated standard errors (SEs) and p values. The coefficients provide the predicted change in the outcome (higher level) variable per unit increase in the predictor (lower level) variable. SEM model fitting algorithms are sensitive to the scaling of the variables, and to aid the algorithms in converging to final estimates, variables were scaled as follows: cerebral microbleeds/10, age/10 (decades), education-occupation (SES)/10. The path analysis was pruned using backward selection and assessing goodness of fit measures (Bayesian information criterion, root mean square error of approximation [RMSEA], standardized root mean square residual [SRMR], Tucker-Lewis Index [TLI], and Comparative Fit Index [CFI] and individual p values) until all paths in the final model were significant, and the model fit observed data well.

2.6. Data availability

Data from the Mayo Clinic Study of Aging and this study are available on request.

3. Results

Data were analyzed for 449 adults with concurrently available amyloid- and tau-PET imaging from the population-based sample of the Mayo Clinic Study of Aging (MCSA). The participant characteristics at the time of imaging are shown in Table 1. Participants’ mean age was 74.5 (±8.4), 252 (56%) were male. At the time of the scans, 417 (92%) were cognitively unimpaired, 32 (7.5%) had mild cognitive impairment (MCI). As defined by amyloid- and tau-PET, 188 (42%) study participants were amyloid positive and 123 (27%) were tau-positive.

Table 1.

Characteristics table. Mean (SD) [range] listed for the continuous variables and count (%) for the categorical variables. WMH is presented as percentage of total intracranial volume (TIV). Abbreviations: SES: socio-economic status; MMSE: Mini Mental State Examination; zGlobal: Individual test scores were converted to z-scores, and a summary score was derived from the z-transformed mean of the four domain scores. WMH: white matter hyperintensities; TIV: Total Intracranial Volume.

	All Participants(n = 449)
Male, no. (%)	252 (56%)
Age, yrs	74.5 (8.4) [60.1, 95.0]
APOE4 carrier, no. (%)	134 (30%)
Education, yrs	14.8 (2.5) [6, 20]
Occupation score	3.5 (1.4) [1,5]
SES (education/occupation)	12.7 (2.4) [5.6, 17.4]
Vascular Risk Score	2.1 (1.3) [0, 6]
MMSE	28.5 (1.3) [23,30]
zGlobal	0.19 (1.03) [-3.32, 2.43]
Amyloid PET, SUVr	1.61 (0.40) [1.17, 3.34]
Amyloid Positive, no. (%)	188 (42%)
Tau PET, SUVr	1.20 (0.10) [0.94, 1.75]
Tau Positive, no. (%)	123 (27%)
WMH Percentage of TIV	1.07 (1.11) [0.07, 8.24]
WMH Volume, cc	16.2 (16.9) [1.0, 119.4]
Genu FA	0.59 (0.05) [0.37, 0.70]
Total Cerebral microbleeds	0.6 (2.4) [0, 35]
Total Infarctions	0.4 (0.9) [0, 6]
Cognitively Impaired, no. (%)	32 (7%)

The theoretical plan for the SEM analyses, including all the predictors and effects tested, is shown in Fig. 2. The results of the significant associations of the structure are shown in Fig. 3. The goodness of fit measures fit well (RMSEA: 0.039, SRMR: 0.037, TLI: 0.964, CFI: 0.980). We report unstandardized regression coefficients. Please note that the coefficients across all arrows cannot be compared, but coefficients on arrows that start with the same predictor and end with the same outcome are comparable. We have included Table 2 with all the direct effects and Supplementary Table 1, which includes all the total, direct and indirect effects observed in Fig. 3. Here we describe the effects (coefficients (SEs)) of each predictor starting with level 1:

• 1.Age. Older age had an impact on nearly all variables, including a direct effect on lower cognition (coefficient = -0.458 (0.069), p < 0.001). Specifically, the cognition score decreases by −0.458 units for every decade i.e. 10-years of increase in age. As age increases, there is an increase in vascular risk score (coefficient = 0.5 (0.069), p < 0.001), worsening of vascular brain health (Genu-FA [coefficient = -0.021 (0.002), p < 0.001], WMH [coefficient = 0.566 (0.057), p < 0.001], total number of infarcts [coefficient = 0.269 (0.047), p < 0.001] and greater amyloid-β deposition (coefficient = 0.181 (0.021), p < 0.001). This translates to about 0.5 increase in vascular risk score, 0.021 decrease in Genu-FA, 0.566% increase in WMH, a 0.269 increase in the number of infarcts and 0.181 SUVR increase in amyloid-β deposition for every decade of aging. The indirect cumulative impact of increasing age on cognition was observed through its influence on structural biomarkers of white matter health (with coefficients of −0.059 (0.026), p = 0.025 for WMH and −0.049 (0.022), p = 0.03 for Genu-FA), amyloid-β deposition (coefficient = -0.069 (0.023), p = 0.002). There was an observed indirect effect of age on the number of cerebral microbleeds and WMH through amyloid deposition (coefficient = 0.153 (0.053) p = 0.004, and coefficient = 0.077 (0.021), p < 0.001, respectively). Age indirectly influenced tau deposition through two pathways: 1) Primarily through amyloid-β to tau (coefficient = 0.033 (0.005), p < 0.001); and 2) to a much lesser degree through the vascular risk score to tau (coefficient = 0.005 (0.002), p = 0.044, total effect coefficient = 0.038 (0.005), p < 0.001).
• 2.Sex. Male (relative to female) sex had a negative direct effect on cognition (coefficient = -0.348 (0.079), p < 0.001), yet the overall total effect was attenuated (coefficient = -0.241 (0.084), p = 0.004) mainly due to an indirect positive effect on SES (coefficient = 0.082 (0.033), p = 0.014) and white matter integrity (coefficient = 0.037 (0.018), p = 0.044). Male sex was directly associated with a higher vascular risk score (coefficient = 0.398 (0.116), p = 0.001) and had an overall impact on increased cerebral microbleeds.
• 3.SES. Higher SES had a protective direct effect on cognition (coefficient = 1.426 (0.159), p < 0.001). The results showed a trend towards a smaller total indirect effect on cognition (coefficient = 0.022 (0.01), p = 0.035) through the vascular risk score, which subsequently influenced WM outcomes (Genu-FA), microbleed occurrence, and, to a lesser extent, tau deposition. SES did not directly or indirectly affect amyloid.
• 4.Vascular risk score. The vascular risk score had a total indirect effect on cognition (coefficient = -0.032 (0.01), p = 0.001). This effect was mediated through a significant indirect effect on the vascular pathway (Genu-FA, coefficient = -0.016 (0.008), p = 0.042, but not for an effect through tau (coefficient = -0.006 (0.004), p = 0.145).
• 5.Biomarkers of vascular brain health.
• a.The vascular risk factor score had a direct effect on Genu-FA (coefficient = -0.007 (0.001), p < 0.001) and on cerebral microbleeds (coefficient = 0.296 (0.085), p < 0.001), but not on WMH. Male sex had a direct effect on Genu-FA (coefficient = 0.016 (0.003), p < 0.001) but not on WMH. There was a negative correlation observed between WMH and Genu-FA (coefficient = -0.017 (0.002), p < 0.001). When comparing structural markers of white matter health, Genu-FA had a greater direct effect on global cognition compared to WMH (coefficient = 2.305 (1.03), p = 0.025 vs. −0.105 (0.045), p = 0.021, respectively).
• b.Cerebral microbleeds had a direct effect on global cognition (coefficient = -0.036 (0.016), p = 0.028), but not on tau deposition. WMH and Genu-FA did not have a direct effect on tau, but the vascular risk score had a direct effect on tau deposition (coefficient = 0.010 (0.005), p = 0.036).
• c.While age had a direct effect on infarcts (coefficient = 0.269 (0.047), p < 0.001), the infarcts component as a predictor was dropped from the analysis during the pruning process for the outcomes global cognition and tau deposition, but it was left in as a confounder for the SEM.
• 6.Biomarkers of AD neuropathology.
• a.Amyloid-β deposition negatively impacted cognition through three pathways: 1) directly with lower global cognition (coefficient = -0.381 (0.118), p = 0.001); 2) through its indirect effect on tau deposition (coefficient = -0.11 (0.055), p = 0.046); and 3) to a much lesser degree, through its indirect effect on WMH (coefficient = -0.044 (0.022), p = 0.047). When comparing the coefficients, the effect of amyloid deposition on cognition directly and through tau (coefficients = -0.381, −0.11 respectively, i.e., the AD-pathway) was 11-fold greater than the indirect effect of amyloid on cognition through WMH (coefficient = -0.044, i.e. the vascular pathway). Amyloid-β deposition had a direct effect on total number of microbleeds (coefficient = 0.848 (0.275), p = 0.002).
• b.Tau deposition had a direct effect on cognition (coefficient = -0.596 (0.294), p = 0.043). The direct effects on elevated tau included amyloid-β positivity (coefficient = 0.185 (0.015), p < 0.001) and, to a lesser degree, vascular risk score (coefficient = 0.010 (0.005), p = 0.036).

Fig. 2.

Theoretical plan of structural equation model. Arrows represent possible effects tested by the model. The AD pathway is shown by the blue boxes. Vascular pathway is shown by green. For sex: male was modeled as 1 and female as 0. Created with BioRender.com. SES: Socio-economic status; WMH: white matter hyperintensities; Genu-FA: Genu fractional anisotropy; PiB: Pittsburgh Compound-B (amyloid-PET). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3.

Final structural equation model along with significant associations shown by solid arrows. The coefficients and standard error are shown beside the arrows. The AD pathway is shown by the blue boxes. Vascular pathway is shown by green. For sex: male was modeled as 1 and female as 0. Created with BioRender.com. SES: Socio-economic status; WMH: white matter hyperintensities; Genu-FA: Genu fractional anisotropy; PiB: Pittsburgh Compound-B (amyloid-PET). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Table 2.

Direct effects.

Type of effect	Path	Estimate (std. error)	p-value
Direct	Age|Cognition	−0.458 (0.069)	<0.001
Direct	Cycle|Cognition	0.132 (0.022)	<0.001
Direct	Sex|Cognition	−0.348 (0.079)	<0.001
Direct	SES|Cognition	1.426 (0.159)	<0.001
Direct	PiB|Cognition	−0.381 (0.118)	0.001
Direct	Tau|Cognition	−0.596 (0.294)	0.043
Direct	WMH|Cognition	−0.105 (0.045)	0.021
Direct	Genu-FA|Cognition	2.305 (1.03)	0.025
Direct	Microbleeds|Cognition	−0.036 (0.016)	0.028
Direct	Age|WMH	0.566 (0.057)	<0.001
Direct	PiB|WMH	0.424 (0.108)	<0.001
Direct	Age|Genu-FA	−0.021 (0.002)	<0.001
Direct	Sex|Genu-FA	0.016 (0.003)	<0.001
Direct	Age|Infarcts	0.269 (0.047)	<0.001
Direct	Vascular risk score|Genu-FA	−0.007 (0.001)	<0.001
Direct	Vascular risk score|Microbleeds	0.296 (0.085)	0.001
Direct	PiB|Microbleeds	0.848 (0.275)	0.002
Direct	PiB|Tau	0.185 (0.015)	<0.001
Direct	Vascular risk score |Tau	0.010 (0.005)	0.036
Direct	Age|PiB	0.181 (0.021)	<0.001
Direct	Sex|SES	0.058 (0.023)	0.011
Direct	Age|Vascular risk score	0.5 (0.069)	<0.001
Direct	Sex|Vascular risk score	0.398 (0.116)	0.001
Direct	SES| Vascular risk score	−0.669 (0.241)	0.006
Correlation	Genu-FA|WMH	−0.017 (0.002)	<0.001

4. Discussion

In this large dataset of multi-modal imaging and clinical information, we investigated the complex relationships among variables of socio-economic status, vascular risk factors, AD, and CVD imaging changes and cognition. These analyses shed light on several complex connections between the various pathways (Fig. 4): (1) CVD and AD biomarkers captured the relative contribution of each of these pathways on cognition. (2) Higher SES had a protective direct effect on cognition and a mediated indirect effect through the vascular pathway. (3) Though amyloid deposition had a significant impact on WMH, the effect of amyloid directly on cognition and through tau was about 11-fold larger than the indirect effect of amyloid on cognition through WMH. (4) There was a significant impact of vascular risk factors on tau deposition that requires further investigation.

Fig. 4.

Major observed associations from SEM model. Highlighted in red are the major relationships discussed in the text. The AD pathway is shown by blue. Vascular pathway is shown by green. Created with BioRender.com. SES: Socio-economic status; WMH: white matter hyperintensities; Genu-FA: Genu fractional anisotropy; PiB: Pittsburgh Compound-B (amyloid-PET). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Cognitive aging and impairment in the population is multi-factorial. In this work, we utilized imaging biomarkers that captured Alzheimer’s and vascular processes upstream of cognitive decline. Using Amyloid-PET and Tau-PET as surrogates of AD pathology, and assuming amyloid accelerates downstream tau deposition, we modeled the AD pathway. We utilized recently proposed diffusion MRI measures that capture early CVD related changes along with WMH, a more traditional CVD imaging measure, to model the CVD pathway. Both brain white matter measures (WMH and Genu-FA) were modeled separately because each represents different aspects of CVD. Genu-FA represents early white matter damage preceding more widespread and significant white matter damage captured by WMH (Shen et al., 2022). Together, AD and CVD pathways contributed to cognitive performance.

Studies are increasingly focusing on the impact of SES on aging and dementia which is proposed to influence cognition through various interconnected mechanisms. For instance, higher SES is associated with increased cognitive reserve (Jones et al., 2011), which can potentially attenuate the impact of vascular injury or AD pathology on cognitive decline (Dufouil et al., 2003, Elkins et al., 2006). However, while some studies support this hypothesis, others have failed to replicate the findings (Soldan et al., 2020, Durrani et al., 2021). We found that the protective effect of SES on global cognition was primarily direct, i.e., higher SES was associated with higher cognition measure. However, there was a trend of indirect effects of SES on cognition mediated through vascular risk factors, which in turn influenced white matter integrity and microbleeds, and, to a lesser extent, tau deposition. These findings indicate that the known CVD and AD biomarkers additionally capture a portion of the indirect measurable effects of SES on cognition.

While the impact of AD and CVD biomarkers on cognition has been studied, the present study is unique in its model of the overlap between the AD and CVD pathways by including an early CVD imaging biomarker, vascular risk in the preceding years, AD imaging biomarkers of amyloid and tau, and a multivariable modeling approach. The observed overlap between “the two pathways” in Fig. 4 could be caused by a number of mechanisms including disruption of the blood brain barrier, dysregulation of neurovascular coupling, amyloid-β aggregation in the vessel wall resulting in cerebral amyloid angiopathy, and reduced clearance of metabolite waste (Rabin et al., 2022, Graff-Radford et al., 2019, Eisenmenger et al., 2022, Love and Miners, 2016, Laing et al., 2020, Dewenter et al., 2023). While the existing literature is unclear whether vascular disease causes, precipitates, amplifies, precedes, or simply coincides with AD (Eisenmenger et al., 2022), this work suggests that vascular risk factors could be a shared factor linked to both CVD and AD (likely through greater tau deposition). While the AD pathway had greater impact on cognition, it also had a smaller impact on development of WMH and microbleeds through amyloidosis. These effects may be attributed to processes that are not necessarily vascular by nature, such as neuroinflammation and oxidative stress (Scott et al., 2015, Abramov et al., 2004), autoregulatory dysfunction (Brickman et al., 2015), or cerebral amyloid angiopathy (Graff-Radford et al., 2019). Further research is required to unravel the biological mechanisms underlying the deposition of brain amyloid-β and the presence of WMH, as well as their implications for cognitive decline.

Tau deposition was significantly influenced by vascular risk factors. Additionally, tau deposition was not directly affected by biomarkers of vascular white matter health, a finding consistent with previous research (Graff-Radford et al., 2019). Prior studies have suggested that the effect of CVD risk factors on cognitive status may be partially mediated by tau pathology (Laing et al., 2020), through mechanisms of BBB disruption, hypoperfusion and ischemia. Specifically, tau deposition has been proposed to hinder the ability to overcome vascular damage, thereby exacerbating its impact on cognitive performance (Yu et al., 2022). This interaction may involve alterations in cerebral blood flow, BBB integrity, immune system recruitment, or genetic regulatory mechanisms (Yu et al., 2022, Laing et al., 2020). It could be argued that the association observed between vascular risk factors and tau deposition mirrors that seen with vascular risk factors and amyloid deposition, as previously suggested (Gottesman et al., 2017). However, it is possible that vascular risk factors have a more pronounced influence on tau deposition in the early stages of disease, as previously proposed (Bos et al., 2019, Vemuri et al., 2017). Several factors may contribute to this, including the potential for disrupted BBB to trigger tau pathology more than amyloid, tau’s heightened sensitivity to chronic inflammation (Newcombe et al., 2018) and the impact of ischemia on neurons, rendering them more susceptible to tau pathology (Pluta et al., 2018, Kapasi et al., 2022).

Interestingly, number of infarcts was not a significant predictor of global cognition or tau positivity. Previous research has demonstrated the impact of infarcts on the elevated risk of future cognitive decline (Benjamin et al., 2018). As suggested by the literature, the expected effect of infarcts on global cognition was likely captured by white matter damage (Rost et al., 2022). This study cohort also had a limited proportion of participants with neuroimaging confirmation of infarction or severe cerebrovascular disease (which may influence participation).

Cerebral microbleeds are small, chronic brain hemorrhages that have gained significant attention in relation to their impact on cognition (Vemuri et al., 2021, Gorelick and Farooq, 2016). In our current analysis, the number of cerebral microbleeds directly influenced global cognition. Our previous findings (Graff-Radford et al., 2019) indicated that male individuals were more likely to exhibit microbleeds. In the present analysis, sex emerged as a predictor of the number of microbleeds, with male sex exerting a detrimental effect through an increased vascular risk score. Conversely, male sex was not predictive of WMH. These findings underscore the complex and differential impact of sex on various forms of cerebrovascular pathology and their subsequent effects on cognitive outcomes.

This study offers several strengths, including well-compiled and comprehensive information from the Mayo Clinic Study of Aging. Limitations of the study include its cross-sectional design and the assumptions made during model construction, which may have resulted in potential oversight of bidirectional associations. It is possible that the impact of amyloid and tau deposition might be more pronounced compared to CVD variables in instances where the sample comprises a higher proportion of participants diagnosed with MCI. The method employed to approximate SES in this study may capture only a portion of the socio-economic factors that are likely to influence brain health. Finally, the cohort has predominantly European ancestry (99% non-Hispanic White). However, the demographics and SES characteristics of our sample closely resemble those of the broader upper Midwest region (St Sauver et al., 2012). Further work to replicate this study in diverse populations across the country and world-wide is needed.

Utilizing socio-demographics, vascular risk factors, and neuroimaging biomarkers to forecast cognitive performance within a population-based cohort offers a valuable opportunity to gain insight into the mechanisms of cognitive aging during the asymptomatic/early symptomatic phase. Additional exploration is required to investigate the underlying mechanisms by which vascular risk factors and SES interact with the progression of vascular, amyloid-β or tau pathology, particularly focusing on longitudinal and causal associations.

CRediT authorship contribution statement

Dror Shir: Writing – original draft, Conceptualization. Jonathan Graff-Radford: Writing – review & editing, Conceptualization. Angela J. Fought: Writing – review & editing, Methodology, Formal analysis, Data curation. Timothy G. Lesnick: Writing – review & editing, Methodology, Formal analysis, Conceptualization. Scott A. Przybelski: Writing – review & editing, Methodology, Formal analysis, Data curation. Maria Vassilaki: Writing – review & editing. Val J. Lowe: . David S. Knopman: Writing – review & editing. Mary M. Machulda: Writing – review & editing. Ronald C. Petersen: Writing – review & editing. Clifford R. Jack: Writing – review & editing. Michelle Mielke: . Prashanthi Vemuri: Writing – review & editing, Supervision, Methodology, Conceptualization.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: M.M. Mielke served as a consultant to Brain Protection Company, Biogen, and LabCorp and receives research support from the National Institutes of Health and the Department of Defense. She is a Senior Associate Editor for Alzheimer's and Dementia: The Journal of the Alzheimer's Association. Dr. Knopman serves on a Data Safety Monitoring Board for the Dominantly Inherited Alzheimer Network Treatment Unit study. He served on a Data Safety monitoring Board for a tau therapeutic for Biogen (until 2021) but received no personal compensation. He is an investigator in clinical trials sponsored by Biogen, Lilly Pharmaceuticals and the University of Southern California. He has served as a consultant for Roche, Samus Therapeutics, Magellan Health, Biovie and Alzeca Biosciences but receives no personal compensation. He attended an Eisai advisory board meeting for Lecanemab on December 2, 2022 but received no compensation directly or indirectly. He receives funding from the NIH. Maria Vassilaki has received research funding from F. Hoffmann-La Roche Ltd and Biogen in the past and consulted for F. Hoffmann-La Roche Ltd; she currently receives research funding from NIH and have equity ownership in Johnson and Johnson, Merck, Medtronic, and Amgen. P. Vemuri received speaking fees from Miller Medical Communications, LLC, and receives research support from the National Institutes of Health. J. Graff-Radford receives NIH funding and serves on the data and safety monitoring board board for Neurology. He has received payment for speaking at the American Academy of Neurology Annual meeting. C.R. Jack serves on an independent data monitoring board for Roche, but he receives no personal compensation from any commercial entity. He receives research support from the National Institutes of Health, the GHR Foundation and the Alexander Family Alzheimer's Disease Research Professorship of the Mayo Clinic. R.C. Petersen is a consultant for Roche, Inc., Eli Lilly and Co. Nestle, Inc. and Eisai, Inc. He receives royalties from Oxford University Press and UpToDate and receives research support from the National Institutes of Health. V. J. Lowe consults for Bayer Schering Pharma, Piramal Life Sciences, Life Molecular Imaging, Eisai Inc., AVID Radiopharmaceuticals, Elly Lilly and Company and Merck Research and receives research support from GE Healthcare, Siemens Molecular Imaging, AVID Radiopharmaceuticals and the NIH (NIA, NCI). D. Shir, T.G. Lesnick, S. Przybelski, A.J. Fought, and M.M. Machulda have no conflicts to report.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Data availability

Data will be made available on request.