Associations of Vascular Risk and Amyloid Burden with Subsequent Dementia

Abstract

Objective:

Midlife vascular risk factors (MVRFs) are associated with incident dementia, as are amyloid β (Aβ) deposition and neurodegeneration. Whether vascular and Alzheimer disease-associated factors contribute to dementia independently or interact synergistically to reduce cognition is poorly understood.

Methods:

Participants in the Atherosclerosis Risk in Communities–Positron Emission Tomography study were followed from 1987–1989 (45–64 years old) through 2016–2017 (74–94 years old), with repeat cognitive assessment and dementia adjudication. In 2011–2013, dementia-free participants underwent brain magnetic resonance imaging (with white matter hyperintensity [WMH] and brain volume measurement) and florbetapir (Aβ) positron emission tomography. The relative contributions of vascular risk and injury (MVRFs, WMH volume), elevated Aβ standardized uptake value ratio (SUVR), and neurodegeneration (smaller temporoparietal brain regions) to incident dementia were evaluated with adjusted Cox models.

Results:

In 298 individuals, 36 developed dementia (median follow-up = 4.9 years). Midlife hypertension and Aβ each independently predicted dementia risk (hypertension: hazard ratio [HR] = 2.57, 95% confidence interval [CI] = 1.16–5.67; Aβ SUVR [per standard deviation (SD)]: HR = 2.57, 95% CI = 1.72–3.84), but did not interact significantly, whereas late life diabetes (HR = 2.50, 95% CI = 1.18–5.28) and Aβ independently predicted dementia risk. WMHs (per SD: HR = 1.51, 95% CI = 1.03–2.20) and Aβ SUVR (HR = 2.52, 95% CI = 1.83–3.47) independently contributed to incident dementia, but WMHs lost significance when MVRFs were included. Smaller temporoparietal brain regions were associated with incident dementia, independent of Aβ and MVRFs (HR = 2.18, 95% CI = 1.18–4.01).

Interpretation:

Midlife hypertension and late life Aβ are independently associated with dementia risk, without evidence for synergy on a multiplicative scale. Given the independent contributions of vascular and amyloid mechanisms, multiple pathways should be considered when evaluating interventions to reduce the burden of dementia.

Epidemiologic studies suggest that vascular risk factors, particularly when measured in midlife, are associated with risk of dementia and cognitive decline in late life.^1–3 Some of these risk factors have been implicated in Alzheimer disease (AD) specifically, when defined clinically as possible/probable AD,^4,5 and some autopsy series have suggested associations between vascular risk factors (ie, hypertension) and neuropathologic changes consistent with AD,⁶ or associations between atherosclerosis in the circle of Willis and AD neuropathologic changes.^7,8 Evidence of a more direct link between vascular risk and AD neuropathology has been strengthened with our publications from the Atherosclerosis Risk in Communities–Positron Emission Tomography (ARIC-PET) study demonstrating that a greater number of midlife vascular risk factors (of hypertension, hypercholesterolemia, obesity, smoking, and diabetes) was associated with elevated brain amyloid in later life,⁹ which, by leading hypotheses, accumulates to cause AD. Furthermore, this association might be even stronger among carriers of an APOE ε4 allele.¹⁰ It is likely, however, that even if vascular risk affects cognitive trajectories at least partially via AD-specific neuropathology, the observed effect of vascular risk on cognitive change and dementia is not entirely mediated via an effect on amyloid or other AD neuropathology.

Understanding the relative contributions of vascular risk, particularly in midlife, and brain amyloid, among individuals without dementia, on cognitive performance and its change, is critical to identifying stages at which dementia may be preventable and may further identify targets for potential prevention. The interaction of vascular disease and Alzheimer pathology is an area of intense interest, and thus imaging biomarkers of AD and other related pathology are informative in evaluating mechanisms of a vascular contribution to dementia. Magnetic resonance imaging (MRI)-defined brain volume can serve as a proxy for neurodegeneration. To evaluate some of these important biomarkers in the development of cognitive decline and dementia, we evaluated the relative contributions to cognitive decline of late life brain amyloid deposition (as measured by florbetapir positron emission tomography [PET]), vascular risk as well as small vessel disease of the brain, and brain volume loss, in the community-based ARIC-PET study. We hypothesized that there would be a multiplicative interaction between vascular risk or vascular markers and brain amyloid, with the poorest cognitive outcomes in individuals with both elevated vascular risk and brain amyloid.

Patients and Methods

Participant Sample

The ARIC study enrolled participants from 4 US communities in 1987–1989, with 6 in-person visits spanning from the 1987–1989 baseline visit through 2017 plus annual (subsequently semiannual) telephone calls. In 2011–2013 (the 5th ARIC in-person visit), the ARIC Neurocognitive Study (ARIC-NCS) was initiated, with a focus on extensive neuropsychological evaluation of participants, and the 6th ARIC visit, which took place in 2016–2017, continued this focus. ARIC visit 5 is referred to as study baseline throughout this article.

From among the participants seen at ARIC visit 5 who did not meet initial criteria for dementia, 346 participants from 3 of the 4 ARIC sites (Washington County, MD; Jackson, MS; and Forsyth County, NC) who had completed brain MRI scans at visit 5 were recruited into the ARIC-PET study, as previously described.¹¹ These individuals constitute the base population for this article. All ARIC-PET participants completed a florbetapir PET scan (2011–2014; study baseline), and were invited back for an ARIC-PET clinic visit (2014–2015).

The study was approved by all relevant institutions’ institutional review boards; all participants provided informed consent.

Cognitive Evaluation

ARIC participants seen at the first (ARIC visit 5) and subsequent ARIC-NCS visits (ie, visit 6) completed a comprehensive neuropsychological battery.¹² This battery assessed 3 primary domains (memory, language, and executive function¹³), with scores considered relative to age, race, and education-derived normative data to generate Z scores. ARIC-PET participants also had the same battery repeated in 2014–2015 (PET clinic visit). ARIC visit 5 was considered the cognitive baseline for this study; change from visit 5 to the PET clinic visit (2–3 years later) to visit 6 (>5 years after visit 5) was evaluated.

As part of the ARIC-NCS visits (visits 5 and 6), participants had informant interviews and were given, through a physician- and neuropsychologist-led adjudication process and standard diagnostic criteria, a research diagnosis of normal cognition, mild cognitive impairment, or dementia.^12,14–16 For this article, only those participants seen in person at visit 5 and therefore with an adjudicated diagnosis of their cognitive status at that visit are included. One individual was excluded based on a research diagnosis of dementia at visit 5 after having been enrolled. Participants did not have to be seen in person at subsequent visits to be included in the analytic sample.

For post-visit 5 incident dementia case identification, ongoing dementia ascertainment and surveillance in the parent ARIC cohort included but was not limited to the above-described in-person evaluation at visit 6. In addition,¹⁴ dementia cases were identified through administration of the Six Item Screener (SIS)¹⁷ annually by phone and the Ascertain Dementia 8-Item Informant Questionnaire (AD8).¹⁸ Finally, hospitalization codes and death certificates identified other cases of dementia during follow-up. Date of dementia onset was defined as the earliest of the date of SIS/AD8 interviews, date of hospitalization records with dementia diagnosis, or date of visit 6.

Magnetic Resonance Imaging

All ARIC-PET participants had 3T research MRI scans as part of ARIC-NCS.¹⁹ These scans were performed at each field center, and included the same magnetic resonance protocol (magnetization-prepared rapid acquisition gradient echo [MP-RAGE], axial T2*gradient echo, axial T2 fluid-attenuated inversion recovery [FLAIR], and axial diffusion tensor imaging), with central coordination and reading of images at the Mayo Clinic Reading Center. Brain volumes were measured using FreeSurfer,²⁰ from the MP-RAGE images, and white matter hyperintensities (WMHs) were measured from the FLAIR sequences using an in-house algorithm created at Mayo, and using a computer-aided segmentation approach. A composite region termed a “temporoparietal volume meta region of interest (ROI)” was calculated, combining the measured brain volumes of the precuneus, hippocampus, parahippocampal gyrus, entorhinal cortex, and inferior parietal lobules,²¹ based on earlier studies demonstrating associations of this group of regional volumes with AD pathology (although this region lacks specificity for AD).

MP-RAGE sequences were also used for coregistration of the florbetapir PET images. MRI and PET scans were all conducted within 1 year of each other, and ideally within 6 months of each other.

PET Imaging

Florbetapir PET images were performed at all 3 study sites using standardized procedures; 50- to 70-minute uptake scans were acquired as described elsewhere, with the cerebellum as the reference region.¹¹ All images were read and standardized uptake value ratios (SUVRs) of cerebral amyloid were quantified at the Johns Hopkins PET reading center. The primary PET measurement for this analysis was based on a global cortical SUVR, which is the weighted average (based on size) of the 9 regions most frequently involved in amyloid deposition in AD pathogenesis (precuneus, orbitofrontal cortex, prefrontal cortex, superior frontal cortex, lateral temporal lobe, parietal lobe, occipital lobe, anterior cingulate, posterior cingulate). As in prior ARIC-PET studies,^9,11 the sample median of 1.2 was used to dichotomize SUVR values, with a global cortical SUVR > 1.2 defined as florbetapir positive. When considered as an independent variable, both binarized florbetapir positivity and log-transformed global cortical SUVR were considered, as was quartile of global cortical SUVR.

Covariates

Demographic and comorbidity covariates were primarily considered at the ARIC baseline visit (visit 1), because prior studies, including some of our own, have shown that the strongest relationships between these risk factors and cognition were found when considered in midlife. For comparison purposes, however, we also considered covariate status at visit 5 (concurrent to the MRI and PET), to test whether these differences between midlife and late life risk factors persisted when considered in combination with the above-described imaging markers. Age at each visit was calculated based on self-reported date of birth at baseline, and sex, race, and educational level (<high school, high school or General Educational Development test, or >high school) were all self-reported once, at study baseline; APOE genotype was determined based on measurement in the ARIC study (TaqMan assay; Applied Biosystems, Foster City, CA) and divided into any versus no ε4 alleles. Smoking status (current vs not current) was self-reported at each visit. Hypertension was defined at each ARIC visit as systolic blood pressure > 140mmHg, diastolic blood pressure > 90mmHg (based on the average of the last 2 of 3 blood pressure measurements for each visit), or use of antihypertensive medications. Diabetes was defined as fasting glucose ≥ 126mg/dl, nonfasting glucose ≥ 200mg/dl, HbA1c ≥ 6.5 (all measured at ARIC visits), or self-report of physician-diagnosed diabetes or use of oral diabetes medications or insulin, and total cholesterol level was measured from plasma using enzymatic methods.²² Weight (in kg) and height (in m²), measured at ARIC visits, were used to calculate body mass index. Although vascular risk factors were primarily considered separately, a summative tally of the number of risk factors present in midlife and late life was calculated, considering the presence or absence of each of the following⁹: hypertension, diabetes, current smoking, hypercholesterolemia, and obesity.

Statistical Analysis

Descriptive characteristics were evaluated overall and for individuals with and without incident dementia over follow-up (using chi-squared tests and Student t tests). To evaluate the association between global cortical SUVR and cognitive decline, linear mixed models were used to account for repeated cognitive performance measurements at visit 5, the PET clinic visit, and visit 6, and to address the missing factor scores at visit 6 due to cohort attrition. The model included both a random intercept and a random slope for the time component, which quantified the rate of cognitive change, and assumed an independent correlation structure. We explored annualized cognitive change between ARIC visits 5 and 6 for the global cognitive factor score and the 3 cognitive factor scores separately. Cox proportional hazards models were used to evaluate the association between global cortical SUVR and incident dementia between visit 5 and visit 6, with visit 5 considered as time 0.

The global cortical SUVR was modeled as a categorical variable based on quartiles and as a continuous variable with log transformation to correct for skewness. The linear mixed models and the Cox models adjusted for demographics (age, sex, race, educational level, and APOE ε4 status) and vascular risk factors from either visit 1 or visit 5 (in separate models). In separate analyses, imaging measures of vascular changes in the brain (ie, WMHs) were also included, as were imaging markers of neurodegeneration (ie, temporoparietal meta ROI; although not specific to one etiology, temporoparietal volume loss is a core feature of AD, and a useful marker of all-cause losses of neurons, synapses, and other supporting elements). The volume of the temporoparietal meta ROI region was standardized and then multiplied by −1, so that the direction of associations would be consistent (with larger values being “worse”). Total intracranial volume was included in models considering brain volumes.

To evaluate for a synergistic association between vascular risk and amyloid on dementia, we tested for multiplicative interactions by introducing, in separate models, an interaction term between each vascular risk factor (as well as number of risk factors) and amyloid into regression models of incident dementia risk. Furthermore, we tested for interactions between each imaging marker (ie, WMHs and temporoparietal meta ROI) and amyloid on incident dementia. Finally, we performed a sensitivity analysis including individuals with prevalent stroke at visit 5. All statistical tests were 2-tailed. A p < 0.05 was considered statistically significant. All analyses were conducted using Stata software (v14.2; StataCorp, College Station, TX).

Results

Of the 346 individuals who completed florbetapir PET scans, we excluded 1 for having dementia at the time of PET, 2 who were non-Whites from Washington County (as is standard in ARIC, due to very small numbers), and an additional 21 missing key covariates, leaving an analytic sample of 322 individuals (Table 1). For the incident dementia analyses, an additional 13 individuals in whom no adjudicated dementia information was available and 11 with prevalent stroke at visit 5 were excluded (Fig 1). As expected with an increase in age of >20 years, the cohort developed higher vascular risk as they progressed from visit 1 to visit 5, with a more than doubling in the rate of hypertension, an increase in diabetes prevalence from 6% to >40%, and a higher number of cardiovascular risk factors overall.

TABLE 1.

Participant Characteristics in Midlife (Visit 1, 1987–1989) and Late Life (Visit 5, 2011–2013)

Characteristic	Visit 1	Visit 5
n	322
Age at study visit, yr (SD)	52.2 (5.2)	75.8 (5.3)
Female sex, n (%)	185 (57.5%)
Black race, n (%)	136 (42.2%)
Education Level, n (%)
<High school	51 (15.8%)
High school or GED	139 (43.2%)
College	132 (41.0%)
APOE ε4, any, n (%)	100 (31.1%)
Current smoker, n (%)	55 (17.1%)	16 (5.0%)
Body mass index, mean (SD)a	27.4 (4.4)	29.0 (5.3)
Hypertension, n (%)	95 (29.5%)	230 (71.4%)
Diabetes, n (%)	20 (6.2%)	130 (40.4%)
Total cholesterol, mean (SD)b	208.3 (39.0)	180.8 (39.6)
Number of CVD risk factors, median (Q1–Q3)	1.0 (1.0–2.0)	2.0 (1.0–3.0)
Number of CVD risk factors category, n (%)
0	65 (20.2%)	35 (10.9%)
1	123 (38.2%)	82 (25.5%)
≥2	134 (41.6%)	205 (63.7%)
High SUVR at visit 5, n (%)c		164 (50.9%)

^akg/m².

^bmg/dl.

^cHigh SUVR defined as a value > 1.2.

CVD = cardiovascular disease; GED = general educational development; SD = standard deviation; SUVR = standardized uptake value ratio.

FIGURE 1: Study population flow. MRI = magnetic resonance imaging.

Cortical Amyloid in Relation to Cognitive Decline and Incident Dementia

The median follow-up after visit 5 was 4.9 (Q1–Q3 range = 4.3–5.2) years, during which 36 participants developed dementia. Increasing continuous florbetapir global cortical SUVR at visit 5 was associated with increased incidence rates of dementia through visit 6, as was increasing quartile of florbetapir SUVR (Table 2). Incidence rates of dementia ranged from 22 per 1,000 person-years for individuals with global cortical SUVR values up to 1.114 (in the lowest quartile) to 74 per 1,000 person-years for individuals with SUVR values between 1.389 and 2.294. For individuals seen in person at both visits 5 and 6, annualized cognitive decline from visit 5 to visit 6 showed overall decreases in performance, with the steepest decline in those individuals with higher levels of brain amyloid at visit 5 (see Table 2).

TABLE 2.

Five-Year Dementia Incidence and Cognitive Decline by SUVR Levels at Baseline

		SUVR Quartiles
	1	2	3	4	p for linear trenda
n	76	75	74	73
SUVR value range	0.952–1.114	1.116–1.199	1.200–1.367	1.389–2.294
Dementia risk
Events/person-years, n	8/362	0/362	3/348	25/336
Incidence rate, per 1,000 person-years	22	0	9	74
Annualized cognitive decline rate
Global cognition	−0.026	−0.027	−0.018	−0.111	<0.001
Memory domain	0.018	0.017	−0.009	−0.091	<0.001
Executive function	−0.033	−0.030	−0.035	−0.100	<0.001
Language domain	−0.033	−0.035	−0.016	−0.069	0.07

^aSUVR values were log transformed.

SUVR = standardized uptake value ratio.

Contributions of Vascular Risk Factors and Brain Amyloid to Dementia Risk

Incident dementia risk after visit 5 was higher in individuals with elevated florbetapir, even when vascular risk factors were included as covariates; the effect of florbetapir (amyloid β [Aβ]) SUVR level on dementia risk was similar when these risk factors were considered at either visit 1 or at visit 5 (Table 3). However, the effect of the risk factors themselves on dementia risk differed depending on when they were measured. Independent of SUVR level, demographics, other vascular risk factors, and APOE ε4, hypertension in midlife (at visit 1) was associated with elevated risk of incident dementia in late life (see Table 3; hazard ratio [HR] = 2.57, 95% confidence interval [CI] = 1.16–5.67), although this lost statistical significance when considered in late life (at visit 5, although with a similar effect size). The relative contributions of both hypertension and elevated SUVR on dementia risk are presented in Figure 2.

TABLE 3.

Associations of Late Life SUVR Level and Midlife/Late Life Vascular Risk Factors with Dementia Risk (n = 298, 36 events)

	Visit 1 Risk Factors + SUVR		Visit 5 Risk Factors + SUVR
Variable	HR (95% CI)	p	HR (95% CI)	p
Age, per 1 year	1.19 (1.09–1.30)	<0.001a	1.19 (1.09–1.29)	<0.001a
Black race	2.34 (1.07–5.14)	0.034a	3.20 (1.34–7.66)	0.009a
Female sex	0.52 (0.24–1.14)	0.101	0.60 (0.27–1.32)	0.204
APOE ε4	0.72 (0.30–1.74)	0.468	0.93 (0.41–2.09)	0.862
Education level
<HS	1 (ref )		1 (ref )
HS or GED	0.61 (0.24–1.54)	0.291	0.30 (0.12–0.70)	0.006a
>HS	0.61 (0.24–1.54)	0.293	0.24 (0.09–0.62)	0.004a
Current smoking	1.26 (0.45–3.55)	0.664	0.17 (0.02–1.84)	0.146
Hypertension	2.57 (1.16–5.67)	0.02a	2.53 (0.87–7.39)	0.089
Diabetes	1.10 (0.33–3.70)	0.875	2.50 (1.18–5.28)	0.016a
BMI > 30	1.95 (0.95–4.01)	0.069	0.40 (0.16–0.97)	0.042a
Total cholesterol	0.63 (0.30–1.34)	0.23	0.52 (0.19–1.41)	0.198
Standardized log SUVR	2.57 (1.72–3.84)	<0.001a	2.36 (1.68–3.32)	<0.001a

Models are adjusted for age, sex, race, education level, APOE ε4 alleles, current smoking (at visit 1 or 5, respectively), elevated BMI, hypertension, diabetes, and elevated total cholesterol.

^aStatistically significant at p 〈 0.05 level.

BMI = body mass index; CI = confidence interval; GED = General Educational Development test; HR = hazard ratio; HS = high school; ref = reference; SUVR = standardized uptake value ratio.

FIGURE 2: Kaplan–Meier curves for dementia-free survival, by midlife hypertension (HTN) and late life elevated florbetapir standardized uptake value ratio (SUVR) status, with corresponding demographic-adjusted hazard ratios (HRs) and 95% confidence intervals (CIs). Ref = reference.

Diabetes, on the other hand, when considered in a model with florbetapir SUVR, demographics, other risk factors, and APOE ε4, was not an important risk factor for dementia when considered in midlife, but was strongly associated with incident dementia risk when considered in late life. Obesity demonstrated differing directions of associations in midlife versus late life; obesity in midlife was associated with nonsignificantly elevated incident dementia risk, independent of other covariates, but was associated with lower risk of dementia when considered in late life (adjusted HR = 0.40, 95% CI = 0.16–0.97; see Table 3). Neither current smoking nor hypercholesterolemia were significantly associated with dementia risk, either from midlife or late life, when considered in the model along with SUVR, although smoking had different (nonsignificant) associations with dementia when considered in midlife versus late life (see Table 3).

When the number of vascular risk factors in midlife and late life was considered, compared to individuals with 0 vascular risk factors in midlife, individuals with 1 risk factor had a nonsignificantly elevated risk of incident dementia (HR = 1.26, 95% CI = 0.35–4.61), with a further nonsignificant elevation in those with 2 or more risk factors in midlife (HR = 2.33, 95% CI = 0.68–8.03), when considered in a model with Aβ SUVR, demographics, and APOE ε4. The model considering instead visit 5 risk factor burden (and with SUVR and other covariates in the same model) was similarly negative, although with smaller effect sizes (1.00, 95% CI = 0.12–8.58 for 1 risk factor vs 0; 1.55, 95% CI = 0.20–12.13 for 2 risk factors vs 0).

In a sensitivity analysis, results for the primary analysis were similar when those individuals with prevalent stroke (n = 11) were included (Table 4).

TABLE 4.

Associations of Late Life SUVR Level and Midlife/Late Life Vascular Risk Factors with Dementia Risk, Including Patients with History of Stroke (n = 309, 37 events)

	Visit 1 Risk Factors + SUVR		Visit 5 Risk Factors + SUVR
Variable	HR (95% CI)	p	HR (95% CI)	p
Age, per 1 year	1.20 (1.09–1.31)	<0.001a	1.18 (1.08–1.28)	<0.001a
Black race	2.54 (1.16–5.58)	0.02a	3.17 (1.37–7.36)	0.007a
Female sex	0.56 (0.26–1.20)	0.135	0.62 (0.29–1.33)	0.221
APOE ε4	0.69 (0.29–1.65)	0.401	0.77 (0.34–1.73)	0.527
Education level
<HS	1 (ref)		1 (ref)
HS or GED	0.59 (0.24–1.49)	0.266	0.28 (0.12–0.65)	0.003a
>HS	0.51 (0.20–1.29)	0.156	0.20 (0.08–0.53)	0.001a
Current smoking	1.46 (0.54–3.91)	0.456	0.47 (0.09–2.48)	0.373
Hypertension	2.44 (1.12–5.29)	0.025a	2.35 (0.83–6.66)	0.107
Diabetes	1.55 (0.50–4.77)	0.447	2.25 (1.10–4.61)	0.027a
BMI > 30	1.86 (0.92–3.78)	0.086	0.43 (0.18–1.01)	0.053
Total cholesterol	0.59 (0.28–1.25)	0.17	0.58 (0.23–1.51)	0.267
Standardized log SUVR	2.55 (1.72–3.77)	<0.001a	2.28 (1.63–3.20)	<0.001a

Models are adjusted for age, sex, race, education level, APOE ε4 alleles, current smoking (at visit 1 or 5, respectively), elevated BMI, hypertension, diabetes, and elevated total cholesterol.

^aStatistically significant at p 〈 0.05 level.

BMI = body mass index; CI = confidence interval; GED = General Educational Development test; HR = hazard ratio; HS = high school; ref = reference; SUVR = standardized uptake value ratio.

Contributions of Imaging Markers to Cognitive Outcomes

When volume of WMHs and Aβ SUVR (both from visit 5) were considered in the same model, both were independent predictors of incident dementia after visit 5, but with loss of statistical significance for WMHs when mid-life vascular risk factors were added to the model (Table 5). When temporoparietal lobe ROI volume (also from visit 5), a marker of atrophy and thus a surrogate for neurodegeneration of any etiology in this study, was considered in the same model as Aβ SUVR, both were independently associated with incident dementia, and remained significantly associated with dementia risk, independent of vascular risk factors measured in middle age (see Table 6). When late life risk factors were instead included as covariates (Table 7), the covariate effects were similar to those described above.

TABLE 5.

Associations of SUVR Level and Volume of WMHs With Incident Dementia and Cognitive Change, Adjusted for Midlife Vascular Risk Factors

		Incident dementia		Cognitive decline
	N (event, if applicable)	298 (36)		322
Models	Variables	HR (95%CI)	p value	Beta (95%CI)	p value
Model 1	log SUVR	2.52 (1.83 to 3.47)	<0.001	−0.03 (−0.04 to −0.02)	<0.001
	log WMH	1.51 (1.03 to 2.20)	0.033	−0.01 (−0.02 to 0.00)	0.069
Model 2	log SUVR	2.58 (1.72 to 3.89)	<0.001	−0.03 (−0.04 to −0.02)	<0.001
	log WMH	1.21 (0.77 to 1.89)	0.411	−0.01 (−0.03 to −0.00)	0.024
	Age (per 1 year)	1.18 (1.08 to 1.29)	<0.001	0.00 (−0.00 to 0.00)	0.46
	Black race	2.27 (1.03 to 4.99)	0.041	0.01 (−0.01 to 0.03)	0.502
	Female sex	0.63 (0.24 to 1.63)	0.339	−0.02 (−0.04 to 0.01)	0.25
	APOE ε4	0.72 (0.29 to 1.77)	0.468	−0.00 (−0.02 to 0.02)	0.991
	Education level
	<high school	1 (ref )		1 (ref )
	HS or GED	0.64 (0.25 to 1.64)	0.349	−0.00 (−0.03 to 0.03)	0.878
	>HS	0.62 (0.25 to 1.57)	0.315	−0.02 (−0.04 to −0.00)	0.049
	V1 hypertension	2.58 (1.13 to 5.85)	0.024	0.02 (0.00 to 0.05)	0.047
	V1 smoking	1.29 (0.44 to 3.77)	0.639	0.01 (−0.02 to 0.04)	0.417
	V1 diabetes	1.04 (0.31 to 3.55)	0.944	−0.02 (−0.06 to 0.03)	0.454
	V1 obesity	1.83 (0.88 to 3.78)	0.105	0.01 (−0.01 to 0.04)	0.223
	V1 hypercholesterolemia	0.61 (0.29 to 1.30)	0.201	−0.01 (−0.03 to 0.01)	0.455

Model 1 only includes listed variables (i.e., log SUVR and log WMH); Model 2 also adjusted for age, sex, race, education level, APOE e4 alleles, and mid-life risk factors including current smoking (at visit 1 or 5, respectively), elevated (>30) body mass index, hypertension, diabetes, elevated total cholesterol, and total intracranial volume (measured at visit 5). SUVR and WMH were both standardized; results are per 1 SD of each imaging marker. HS = high school; GED = general educational development; SUVR = Standardized uptake value ratio; V1 = visit 1; WMH = white matter hyperintensity.

TABLE 6.

Associations of SUVR Level and Temporo-Parietal Meta ROI Volume With Incident Dementia and Cognitive Change, Adjusted for Midlife Vascular Risk Factors

		Incident dementia		Cognitive decline
Models	Variables	HR (95% CI)	p value	Beta (95% CI)	p value
Model 1	log SUVR	2.17 (1.56 to 3.02)	<0.001	−0.03 (−0.04 to −0.02)	<0.001
	temporo-parietal meta ROI volume	2.76 (1.70 to 4.47)	<0.001	−0.01 (−0.03 to 0.00)	0.145
Model 2	log SUVR	2.66 (1.72 to 4.12)	<0.001	−0.03 (−0.04 to −0.02)	<0.001
	temporo-parietal meta ROI volume	2.18 (1.18 to 4.01)	0.013	−0.02 (−0.03 to 0.00)	0.094
	Age (per 1 year)	1.12 (1.01 to 1.24)	0.034	0.00 (−0.00 to 0.00)	0.52
	Black race	1.92 (0.82 to 4.48)	0.131	0.01 (−0.01 to 0.03)	0.422
	Female sex	0.50 (0.18 to 1.39)	0.185	−0.02 (−0.05 to 0.01)	0.148
	APOE ε4	0.51 (0.19 to 1.38)	0.184	0.00 (−0.02 to 0.03)	0.91
	Education level
	<high school	1 (ref)		1 (ref)
	HS or GED	0.81 (0.30 to 2.14)	0.665	0.00 (−0.03 to 0.03)	0.962
	>HS	0.74 (0.29 to 1.91)	0.536	−0.02 (−0.04 to 0.00)	0.085
	V1 hypertension	1.97 (0.85 to 4.55)	0.111	0.02 (−0.00 to 0.05)	0.058
	V1 smoking	1.29 (0.43 to 3.83)	0.649	0.01 (−0.02 to 0.04)	0.399
	V1 diabetes	1.28 (0.38 to 4.29)	0.692	−0.03 (−0.07 to 0.02)	0.272
	V1 obesity	2.21 (1.00 to 4.87)	0.05	0.01 (−0.01 to 0.04)	0.233
	V1 hypercholesterolemia	0.71 (0.33 to 1.50)	0.366	−0.01 (−0.03 to 0.01)	0.267

Model 1 only includes listed variables (i.e., log SUVR and temporo-parietal meta ROI volume); Model 2 also adjusted for age, sex, race, education level, APOE e4 alleles, and mid-life risk factors including current smoking (at visit 1 or 5, respectively), elevated (>30) body mass index, hypertension, diabetes, elevated total cholesterol, and total intracranial volume (measured at visit 5). SUVR and temporo-parietal meta ROI region were both standardized; results are per 1 SD of each imaging marker. For temporo-parietal meta ROI, results are multiplied by −1 for a consistent direction of associations across multiple imaging markers.

HS = high school; GED = general educational development; ROI = region of interest; SUVR = Standardized uptake value ratio; V1 = visit 1.

TABLE 7.

Associations of SUVR Level, Volume of WMHs, and Temporal–Parietal Meta ROI Volume with Incident Dementia and Cognitive Change Adjusted for Late Life Risk Factors

	Incident Dementia, n = 298, 36 Events		Cognitive Decline, n = 322
Variable	HR (95% CI)	p	Beta (95% CI)	p
Models with WMHs and V5 risk factors
Log SUVR	2.33 (1.65 to 3.29)	<0.001a	−0.03 (−0.04 to −0.02)	<0.001a
Log WMHs	1.20 (0.73 to 1.96)	0.468	−0.01 (−0.03 to −0.00)	0.02a
Age, per 1 year	1.18 (1.07 to 1.29)	<0.001a	0.00 (−0.00 to 0.00)	0.667
Black race	3.03 (1.26 to 7.32)	0.014a	0.01 (−0.01 to 0.03)	0.376
Female sex	0.61 (0.23 to 1.61)	0.320	−0.02 (−0.04 to 0.01)	0.223
APOE ε4	0.95 (0.41 to 2.18)	0.896	−0.00 (−0.03 to 0.02)	0.765
Education level
<HS	1 (ref)		1 (ref)
HS or GED	0.32 (0.13 to 0.77)	0.012a	0.01 (−0.02 to 0.04)	0.649
>HS	0.24 (0.09 to 0.62)	0.003a	−0.02 (−0.04 to 0.00)	0.065
V5 hypertension	2.33 (0.78 to 6.96)	0.13	0.01 (−0.01 to 0.03)	0.407
V5 smoking	0.16 (0.01 to 1.81)	0.137	−0.05 (−0.10 to −0.01)	0.028a
V5 diabetes	2.49 (1.18 to 5.27)	0.017a	−0.01 (−0.03 to 0.01)	0.207
V5 obesity	0.40 (0.17 to 0.98)	0.045a	0.01 (−0.01 to 0.03)	0.383
V5 hypercholesterolemia	0.55 (0.20 to 1.53)	0.25	−0.00 (−0.03 to 0.02)	0.811
Models with temporoparietal meta ROI volume and V5 risk factors
Log SUVR	2.61 (1.79 to 3.81)	<0.001a	−0.03 (−0.04 to −0.02)	<0.001a
Temporoparietal meta ROI volume	2.47 (1.41 to 4.33)	0.002a	−0.01 (−0.03 to 0.00)	0.139
Age, per 1 year	1.13 (1.01 to 1.26)	0.035a	0.00 (−0.00 to 0.00)	0.853
Black race	2.52 (0.99 to 6.43)	0.053	0.01 (−0.01 to 0.03)	0.31
Female sex	0.53 (0.19 to 1.47)	0.223	−0.02 (−0.05 to 0.01)	0.112
APOE ε4	0.66 (0.27 to 1.64)	0.373	−0.00 (−0.03 to 0.02)	0.804
Education level
<HS	1 (ref)		1 (ref)
HS or GED	0.38 (0.15 to 0.95)	0.039a	0.01 (−0.02 to 0.04)	0.464
#x003E;HS	0.31 (0.11 to 0.83)	0.020a	−0.02 (−0.04 to 0.00)	0.111
V5 hypertension	2.42 (0.83 to 7.10)	0.107	0.01 (−0.02 to 0.03)	0.511
V5 smoking	0.11 (0.00 to 2.49)	0.165	−0.06 (−0.10 to −0.01)	0.024a
V5 diabetes	2.72 (1.24 to 5.95)	0.012a	−0.01 (−0.03 to 0.01)	0.269
V5 obesity	0.34 (0.13 to 0.86)	0.023a	0.01 (−0.01 to 0.03)	0.482
V5 hypercholesterolemia	0.35 (0.12 to 1.04)	0.058	0.00 (−0.02 to 0.03)	0.771

Models are adjusted for age, sex, race, education level, APOE ε4 alleles, and late life risk factors including current smoking, elevated (>30) body mass index, hypertension, diabetes, elevated total cholesterol, and total intracranial volume. SUVR, WMHs, and temporoparietal meta ROI volume were all standardized; results are per 1 standard deviation of each imaging marker.

^aStatistically significant at p 〈 0.05 level.

CI = confidence interval; GED = General Educational Development test; HR = hazard ratio; HS = high school; ref = reference; ROI = region of interest; SUVR = standardized uptake value ratio; V5 = visit 5; WMH = white matter hyperintensity; HS = high school.

When the outcome of annualized cognitive decline was considered, Aβ SUVR and WMHs were both associated with steeper decline, independent of midlife vascular risk factors, but smaller temporoparietal lobe ROI was no longer associated with cognitive decline when in a model with Aβ SUVR (see Tables 5 and 6). Individual vascular risk factors were for the most part not associated with cognitive decline, with the exception of midlife hypertension, which was borderline significantly associated with less cognitive decline over time; of note, individuals with midlife hypertension had lower cognitive scores at visit 5 (cognitive baseline), which could lead to a floor effect and explain the lack of worsening over follow-up.

Evaluations for Synergy between Aβ and Vascular Risk Factors

No statistically significant interactions were found between either midlife or late life vascular risk factors and Aβ on dementia risk between visits 5 and 6. The number of midlife or late life vascular risk factors and Aβ did not interact on dementia risk (1 vascular risk factor × Aβ, adjusted interaction HR = 1.52, 95% CI = 0.52–4.42, p = 0.44 [midlife risk factors]; adjusted interaction HR = 0.36, 95% CI = 0.03–5.12, p = 0.45 [late life risk factors]; 2 risk factors × Aβ, adjusted interaction HR = 1.20, 95% CI = 0.49–2.90, p = 0.69 [midlife risk factors]; adjusted interaction HR = 0.38, 95% CI = 0.03–5.15, p = 0.47 [late life vascular risk factors]). Similarly, there was no evidence of interaction on the multiplicative scale between WMHs and Aβ (adjusted interaction HR = 0.81, 95% CI = 0.54–1.23, p = 0.33), although results were more suggestive of a potential interaction for neurodegeneration and Aβ on dementia risk (adjusted interaction HR = 0.73, 95% CI = 0.51–1.04, p = 0.08). Results were similar for analyses of cognitive decline after visit 5.

Discussion

In this study of individuals from 3 US communities that were without dementia at the time of amyloid PET imaging, amyloid level by PET as well as vascular risk status independently contributed to subsequent risk of dementia. Specifically, midlife hypertension remained important for late life development of dementia, independent of amyloid level, and late life diabetes was important for incident dementia, independent of amyloid level. Importantly, relationships described above were additive, without evidence of an interaction on a multiplicative scale.

In addition to vascular risk factors, WMHs, which may represent a downstream sequela of these risk factors, were no longer significantly associated with dementia when vascular risk factors themselves were also considered, emphasizing the importance of considering earlier factors that may have contributed to these late life imaging markers. Temporoparietal lobe ROI, however, a non-specific marker of neurodegeneration, did contribute to dementia risk, independent of amyloid PET and vascular risk factors. This result is consistent with earlier studies showing that atrophy independently contributes to cognition and longitudinal cognitive decline, independent of amyloid and tau.²³

These results emphasize the importance of considering pathologies of several different types, including potentially vascular etiologies, when evaluating risk for dementia or cognitive decline. This is supported further by pathologic studies that have shown a high proportion of mixed pathologies, particularly in diverse populations similar to the ARIC cohort, with more frequent mixed pathologies reported previously in Black decedents,²⁴ and particularly in community-based (vs clinical) cohorts.²⁵

Other studies have reported similar additive effects of vascular and amyloid pathologies on cognition, with most failing to find evidence of a multiplicative interaction, similar to what we report, although few have the longitudinal follow-up to consider midlife vascular risk, which in the ARIC study not only has been shown to be especially important for cognition¹ but also has been found to be directly associated, itself, with late life brain amyloid.⁹ Among patients with AD, the presence of infarcts appears to be associated with worse cognition,²⁶ and 2 autopsy studies have demonstrated that measures of cerebrovascular disease and AD each independently predicted pre-mortem dementia status.^27,28 Another study reported that “vascular brain injury,” defined by infarction and WMHs, had a greater influence on cross-sectional cognitive function than did extent of amyloid, without evidence of an interaction between amyloid and vascular risk²⁹; in the Mayo Clinic Study of Aging, a similar additive but not multiplicative effect of vascular risk and brain amyloid on cognition was reported.³⁰ Others, however, have found evidence for a synergistic worsening in cognition among individuals with both elevated vascular risk and elevated amyloid burden.³¹ In the Harvard Aging Brain Study, cognitive decline was steepest in individuals with higher vascular risk as well as greater amyloid burden by Pittsburgh compound B PET, with evidence of not only independent effects but also interaction on the multiplicative scale.³²

Vascular contributions to dementia are defined in various ways in the existing literature, but the majority of these above-cited studies consider their contributions later in life, and often concurrent to the amyloid characterization. Our prior work has emphasized the importance of midlife as opposed to late life vascular risk,^1,33 and thus, studies that only evaluate late life vascular risk factors may underestimate a vascular contribution to cognitive endpoints. Here, we found that only midlife hypertension, and not late life hypertension, was significantly independently associated with dementia, but interestingly, diabetes, which has been shown to be important for dementia even in later life,³⁴ was only important when considered in late life. We note, however, that even for diabetes, we only evaluated incident dementia cases after visit 5, when diabetes was defined, making it unlikely that these associations are due to reverse causation. The lack of an earlier life association may simply reflect insufficient power due to smaller numbers of diabetes cases in midlife, or may point to either a prediabetes process (or even genetic or environmental factors)³⁵ that influences dementia risk, or an impact on cognition from hyperglycemia in individuals with preclinical neurodegenerative conditions.

We found evidence consistent with other published literature³⁶ that the relationship between late life obesity and dementia differed from that with midlife obesity, independent of amyloid levels, with the suggestion of an inverse effect (less risk of dementia) in association with late life obesity; this so-called “obesity paradox” may be a result of reverse causation, with lower body weight resulting from dementia, rather than being a risk factor for dementia. In addition to consideration of midlife versus later life vascular risk factors, some prior studies have evaluated vascular disease, such as microvascular disease of the brain (ie, WMHs and infarcts), as opposed to risk factors. In this study, although WMHs (in later life, at the time of the amyloid PET) independently contributed to dementia, this association lost significance when midlife vascular risk factors themselves were added to the model, suggesting that using these measures alone may also underestimate the full extent of vascular impact on dementia.

One potential reason for inconsistencies in studies to date, including our own, that attempt to evaluate for interaction between vascular and amyloid burden on cognition is the limitation by survivorship biases, in that individuals who survive to an older age and are recruited into a study with imaging and regular follow-up are less likely to have a high burden of multiple brain pathologies. Therefore, the population that does have high levels of multiple pathologies (vascular, neurodegenerative, and amyloid, for instance) and is still in a study like ARIC-PET is likely a sample that has already proven successful survival and protection from dementia, likely due to other contributing factors related to cognitive reserve or vascular risk factor control that are not well captured here or in other studies evaluating these relationships.

Although this study considers 2 of the identified components of the proposed ATN (amyloid/tau/neurodegeneration) framework for AD³⁷ (amyloid and neurodegeneration), the results add to the existing literature by emphasizing the importance of another component, the “V” of vascular disease. In response to this framework, it has been proposed that vascular dysfunction be considered as an important additional contributor to AD,³⁸ and that consideration of biomarkers include biomarkers of vascular dysfunction, such as markers of small vessel disease of the brain. Our results further support the independent and important role of “V” as a potentially added component to this framework, given its independent contributions, beyond that of amyloid. The lack of statistical evidence for a synergistic association, as evidenced by lack of a statistically significant interaction between amyloid and vascular risk, does not eliminate the importance of vascular risk in dementia or even AD specifically, but rather the observed additive association further emphasizes the contributions made by vascular risk factors to dementia even in the presence of AD-specific neuropathology. Furthermore, it is possible that an interaction still exists but was not detected due to our relatively small sample size.

ARIC-PET is an observational study, and thus is limited by the potential for unmeasured confounding, as there may be factors that contribute to late life dementia, including vascular, amyloid, tau, and other neurodegenerative pathologies, that are not well captured here. The lack of any measure of tau is a limitation. In addition, there is a possibility of a survival bias, and the dementia adjudication and surveillance process, although thorough, is imperfect and could result in some ascertainment bias. Additionally, the absolute number of dementia events was relatively small, and thus it is possible that this study was underpowered to detect not only an interaction between vascular risk or vascular injury and amyloid but also to detect independent associations with risk factors and dementia or cognitive decline; this may have been especially true for certain risk factors in midlife, such as diabetes, when there were relatively few participants having these risk factors. A better understanding of the relative contributions of vascular risk, vascular disease, and neurodegenerative processes to cognition would be better evaluated with a greater chronological separation between imaging and incident dementia, as our brain imaging was conducted at visit 5, with incident dementia cases identified in the subsequent 5+ years; knowing that vascular risk factors are particularly important in middle age for later life cognition, it is likely that imaging markers may take some time to exert their influence, if present, on cognition and dementia risk.

Vascular and amyloid mechanisms of cognitive decline and dementia both play important roles in the aging brain, and both are likely to have long latency for neurocognitive effects, but our study did not confirm evidence of synergistic interactions between the two, although it may have been limited by a relatively small sample size. We do, however, note that each has an independent role in the development of dementia, and therefore any attempts to treat or prevent dementia or AD specifically should consider both pathways of brain injury. Furthermore, midlife vascular risk remains a critical factor in determining an individual’s risk for dementia, and these earlier markers of vascular risk may need to be considered in future studies evaluating the relative contributions of pathologies leading to neurodegenerative conditions. Our findings further support the broad heterogeneity in dementia causation, further emphasizing the likely need to target multiple pathways in ultimate efforts for prevention.