Long‐Term Blood Pressure Variability in Cerebral Amyloid Angiopathy is Associated with Ischemic Brain Injury and Cognitive Decline

Abstract

Background

Long‐term blood pressure variability (BPV) has been proposed as a potential risk factor for dementia and cerebral small vessel disease progression. In this study, we investigate the association between BPV, brain injury, and cognitive decline in probable cerebral amyloid angiopathy (CAA).

Methods

Using a prospective memory clinic cohort, we enrolled 102 participants, including 52 with probable CAA and mild cognitive symptoms. BPV was assessed using a coefficient of variation derived from outpatient BP measurements (median 12) over 5 years before imaging with 3‐tesla research magnetic resonance imaging. We measured peak width of skeletonized mean diffusivity and neuroimaging markers of CAA, including lacunes and cortical cerebral microinfarcts. Using regression models, we evaluated the association of BPV with white matter integrity and whether CAA modified this association. We also examined the association of BPV with longitudinal cognitive decline.

Results

Systolic BPV had a dose‐dependent association with peak width of skeletonized mean diffusivity (standardized β=0.22, 95% CI: 0.06–0.39, P=0.010), independent of age, sex, mean BP, common vascular risk factors, brain atrophy, and CAA severity. The presence of probable CAA strengthened the association between BPV and peak width of skeletonized mean diffusivity (β=9.33, 95% CI: 1.32–17.34, P for interaction=0.023). Higher BPV correlated with the presence of lobar lacunes, cortical cerebral microinfarcts, and a decline in global cognition and processing speed.

Conclusions

Long‐term BPV had a dose‐dependent association with altered white matter integrity, ischemic brain injury, and cognitive decline. Controlling BPV might be a potential novel therapeutic target to prevent cognitive decline in memory clinic patients with probable CAA and mild cognitive symptoms.

Nonstandard Abbreviations and Acronyms

BPV

blood pressure variability

CAA

cerebral amyloid angiopathy

PSMD

peak width of skeletonized mean diffusivity

SVD

small vessel disease

WMH

white matter hyperintensities

Clinical Perspective.

What Is New?

• In this memory clinic cohort of patients with and without cerebral amyloid angiopathy, visit‐to‐visit blood pressure variability was independently associated with altered white matter integrity and ischemic brain injury, beyond mean blood pressure and established vascular risk factors.

• Underlying cerebral amyloid angiopathy may contribute to the association between blood pressure variability and white matter injury.

What Are the Clinical Implications?

• These findings suggest that long‐term blood pressure variability is associated with silent ischemic brain injury and cognitive decline and may represent a novel target for preventing or mitigating cerebral small vessel disease and its cognitive consequences.

Small vessel diseases (SVD) account for up to 30% of strokes and contribute to up to 50% of dementia cases, yet no effective treatment exists and preventive strategies are limited. Cerebral amyloid angiopathy (CAA) is a common and well‐characterized SVD and one of the main drivers of cognitive decline in older adults. ¹ , ² CAA is defined by amyloid‐beta deposits in leptomeningeal and cortical small vessels that cause extensive vessel wall remodeling. ³ , ⁴ The severe changes in small vessel wall physiology, including loss of smooth muscle cells, cracking, and fibrinoid necrosis may underlie the impaired vascular reactivity in CAA. ⁵ , ⁶ Overall, the accumulation of widespread CAA‐related ischemic and hemorrhagic injury, together with Alzheimer disease, are the most important substrates to dementia in older adults. ¹ , ⁷ Largely, because of advances in neuroimaging, the diagnosis of probable CAA can be made during life using the Boston criteria with high sensitivity and specificity in patients who present clinical symptoms of CAA, including cognitive impairment. ⁸ , ⁹ Despite its high prevalence in older adults, it remains unclear how the presence of CAA influences the effects of vascular risk factors, particularly elevated or variable blood pressure (BP), on brain tissue integrity.

Compelling evidence suggests that intensive BP control may slow cognitive decline and white matter disease. ¹⁰ , ¹¹ However, hypertension is not always fully explained by mean BP alone as a risk factor for vascular events, indicating that other underlying features of BP profiles, such as instability or visit‐to‐visit BP variability (BPV), also influence cerebrovascular outcomes in addition to the mean BP level. ¹² BPV, as a marker of long‐term BP fluctuations, has emerged as a risk factor for stroke ¹³ , ¹⁴ , subclinical white matter disease ¹⁵ , ¹⁶ , ¹⁷ , cognitive impairment ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , and neuropathologic changes ¹⁹ , ²¹ , with prognostic value beyond that offered by mean BP levels. ¹² , ²³ A post‐hoc analysis of SPRINT‐MIND trial has reported that participants with the highest BPV were more likely to progress from mild cognitive impairment (MCI) to dementia despite excellent mean BP control. ²² The mechanisms underlying this relationship between increased BPV and dementia and whether vascular brain pathology influences this association remain unknown.

Early white matter microstructural alterations in CAA have been associated with cognitive impairment. ²⁴ , ²⁵ , ²⁶ Since white matter hyperintensities (WMH) of presumed vascular origin are a strong predictor of dementia, it is crucial to identify and prevent the progression of WMH at an early stage. ²⁷ , ²⁸ More advanced imaging metrics, such as the diffusion magnetic resonance imaging (MRI)‐based peak width of skeletonized mean diffusivity (PSMD), can detect early alterations in white matter integrity before they are visible on standard clinical imaging. ²⁶ , ²⁹ Other markers of CAA, such as cortical cerebral microinfarcts and lacunes, are also strongly associated with cognitive decline. ³⁰ , ³¹ , ³² However, the mechanisms driving CAA‐related vascular brain remain unclear.

Our hypothesis was that increased BPV may be associated with vascular brain injury partly driven by the underlying CAA‐related vasculopathy and impaired small vessel responses to BP variability over time. Using a prospective memory clinic cohort, our objective was to determine if long‐term BPV was linked to white matter integrity, brain injury, and domain‐specific cognitive decline. We also examined whether the presence of CAA imaging markers influenced the relationship between BPV and white matter integrity.

METHODS

Data will be available upon reasonable request to the corresponding author.

Study Design

Using our single‐center prospective memory clinic CAA cohort, we employed a hybrid observational study design. This included longitudinally acquired BP data, cross‐sectional neuroimaging analysis, and a longitudinal assessment of the association between BPV and cognitive decline, following STROBE reporting guidelines.

All research procedures were approved by the Institutional Review Board and written informed consent was obtained from all participants.

Study Population

Participants with and without CAA presenting mild cognitive symptoms were recruited from the ongoing Massachusetts General Hospital CAA memory clinic cohort between August 2010 and October 2020. Detailed inclusion criteria for cohort enrollment are available in Data S1, as previously described. ²⁴ , ³³ In brief, we included non‐demented participants with available neuropsychological assessments, research MRI, and at least two BP measures over a 5‐year period. The modified Boston criteria were used to identify probable/possible CAA as described in Data S1. ³⁴ The Boston criteria for probable CAA have high specificity and positive predictive value for advanced CAA pathology in hospital‐based cohorts presenting cognitive symptoms but no prior intracerebral hemorrhage. ⁹ Participants who fulfilled possible CAA criteria were excluded from the primary study sample because of lower diagnostic certainty. Participants who did not fulfill the probable or possible CAA criteria were classified as non‐CAA. Participants with mixed cerebral microbleed pattern, defined by the concomitant presence of lobar and deep microbleeds, were excluded.

Data Collection and Follow‐Up

Participants underwent baseline clinical evaluation, research MRI, and neuropsychological examination at baseline and annually for 2 years, as previously described (Data S1). ²⁴ , ³³

Blood Pressure Assessments and Calculation of BP Variability

The BP measurements were collected retrospectively from predetermined study cohort visits and using electronic health records over 5 years before baseline imaging (Data S1), according to previously published methods in cohorts with CAA and cognitive decline. ²¹ , ³⁵ The coefficient of variation defined as the ratio of the standard deviation to the mean (SD/mean) was used to calculate systolic and diastolic BP variability for every participant. This method is widely accepted for assessing overall BP variability over time and provides more reliable information than SD about fluctuations in BP by correcting for the proportionality observed between the sample mean and variability. ¹²

Neuropsychological Assessment

Composite z‐scores adjusted for sex, age, and education were calculated for five cognitive domains (global cognition, attention/processing speed, executive function, memory, language/semantic function) using 10 individual tests, as previously described (Data S1). ²⁴ , ³³

Neuroimaging Acquisition and Analysis

All subjects completed a 3‐Tesla research MRI scan (Magnetom Prisma‐Fit or TIM‐Trio, Siemens Healthineers, Erlangen, German) using a 32‐channel head coil. The imaging was performed within 3 months of neuropsychological evaluation (median=0 days, IQR range 0–62.5). The imaging protocol is detailed in Data S1.

Conventional MRI markers of cerebral SVD were rated by an experienced neuroradiologist blinded to clinical information (M.C.Z.Z.), following consensus recommendations. ²⁴ , ³² In brief, we determined the presence, number, and topography of cerebral microbleeds, cortical superficial siderosis, the extent of visually‐rated WMH using Fazekas score, presence and number of lacunar infarcts by region (deep and lobar), and cortical cerebral microinfarcts. Based on imaging markers, a total composite CAA score was computed. ³⁶

To exclude excessive motion artifacts, we conducted a thorough visual inspection of diffusor tensor imaging as well as a quantitative evaluation using the Total Motion Index. ³⁷ We ran the fully automated PSMD script version 1.5/2020, as previously described (Figure 1). ²⁴ , ²⁶

Figure 1. White matter integrity assessment with peak width of skeletonized mean diffusivity and emerging non‐hemorrhagic markers in cerebral amyloid angiopathy.

A, Probable CAA subject with high burden of white matter hyperintensity (Fazekas 3) and increased mean diffusivity of main white matter tracts. B, CAA subject with low white matter hyperintensity burden (Fazekas 1) and preserved mean diffusivity of white main white matter tracts. C, Histogram analysis allows calculating the PSMD. Emerging non‐hemorrhagic imaging markers of cerebral amyloid angiopathy are (D) cortical cerebral microinfarcts identified on a sagittal T1‐weighted images, and (E) lobar lacunes, identified on axial FLAIR images. CAA indicates cerebral amyloid angiopathy; and PSMD, peak width of skeletonized mean diffusivity.

The advanced quantitative neuroimaging analysis protocol is described in detail in Data S1. In brief, we used FreeSurfer (version 6.0) volumetric pipeline to compute the total brain volume, total intracranial volume, and parenchymal brain fraction, as previously described. ²⁴ , ³³ , ³⁸ WMH volume was computed using the lesion prediction algorithm from the Lesion Segmentation Tool version 3.0.0 for Statistical Parametric Mapping 12 and was normalized to the intracranial volume (WMH volume/intracranial volume). ³⁹

Statistical Analysis

Descriptive data of demographic, clinical, and radiologic characteristics of participants were compared between CAA and non‐CAA groups. Normality was assessed using the Shapiro–Wilk test and visual inspection of Q–Q plots. Continuous variables were compared using independent‐samples t‐tests or Wilcoxon rank‐sum (Mann–Whitney) tests with Cohen's d or biserial‐rank correlation for effect size, depending on distribution. Categorical variables were compared using χ² or Fisher exact test, with Cramer's V or odds ratio, where appropriate. Test statistics and effect sizes are reported, as indicated.

Logarithmic transformations were used for normalized white matter hyperintensity volumes, and cerebral microbleed counts to achieve a normal distribution. In the primary analysis, we used linear regression models to evaluate the association between BP profile (mean and variability) and PSMD. Linearity of continuous predictors was assessed using multivariable fractional polynomials. Variables for which the selected fractional polynomial power was 1 were considered to have linear associations with the outcome. The model was adjusted for demographic (age, sex), vascular risk (diabetes, mean systolic BP), and imaging variables (total CAA score and brain parenchymal fraction) shown to be associated with white matter integrity and cognition. ³ , ²⁹ , ⁴⁰ In prespecified secondary analyses, linear and logistic regression models were used to evaluate the associations between long‐term systolic BPV and CAA imaging markers, for continuous and binary measures, respectively. The models were adjusted for established risk factors—specifically age, sex, diabetes, and mean blood pressure—while minimizing the risk of overfitting. Effect sizes were reported as adjusted R² and standardized regression coefficients (β) or odds ratios (OR), where approriate, together with their corresponding 95% CIs. A prespecified secondary analysis used an interaction term to verify whether the presence/absence of probable CAA modified the association between BPV and white matter integrity. We then examined whether a dose‐effect relationship existed between BPV quartiles and common SVD imaging markers. We used regression models to assess the association between BPV and longitudinal performance change in 5 cognitive domains adjusted for age, sex, and education.

Secondary sensitivity analyses included adding to the study sample possible CAA cases, adjusting the regression models for hypertension status, antihypertensive medication usage, CAA severity score, brain atrophy, and restricting analysis to participants with at least five BP measurements.

The proportion of missing data was small (ranging from 0% to 2.0% of all the covariates analyzed), including neuropsychological tests. Missing data were handled using the missing indicator approach by adding an additional category indicating missing values. Standard diagnostic methods and graphical examination of residuals were used to verify the applicability conditions in regression models. A threshold of P<0.05 was applied for significance. All analyses were performed on R (version 4.5.2).

RESULTS

We assessed 201 participants from the prospective memory clinic CAA cohort for eligibility and selected 102 eligible non‐demented participants (aged 73.4±7.4 years, 18.6% women) with available data on BP measurements, complete neuroimaging, and neuropsychological assessments (Figure 2).

Figure 2. Participant selection flowchart.

Possible CAA was defined as a single CMB in lobar regions or a single cortical superficial siderosis. BP indicates blood pressure; CAA, cerebral amyloid angiopathy; CMB, cerebral microbleeds; MGH, Massachusetts General Hospital; and MRI, magnetic resonance imaging.

Table 1 presents the demographic, clinical, and neuroimaging characteristics of the included study participants. The CAA group had numerically more outpatient BP measurements (median 12 [IQR 6–36] versus 10 [4–15] in the non‐CAA group, W=1006, P=0.049, rank‐biserial correlation r=0.195). Additionally, the CAA participants had a higher occurrence of cortical cerebral microinfarcts, more significant brain atrophy, and higher PSMD values.

Table 1.

Demographic, Clinical, and Radiologic Characteristics in Subjects With Mild Cognitive Symptoms in Probable CAA and Non‐CAA Participants

	Total (n=102)	CAA‐MCS (n=52)	Non‐CAA‐MCS (n=50)	Test statistic/effect size	P value
Demographics
Age, y	73.4±7.4	74.2±7.0	72.6±7.9	t (97.7)=1.04/Cohen's d=0.21	0.298
Sex, women	19 (18.6)	8 (15.4)	11 (22.0)	χ2 (1, N=102)=0.74/Cramer's V=0.08	0.391
Race, White	91 (90.1)	46 (88.5)	45 (91.8)	χ2 (1, N=102)=0.32/Cramer's V=0.06	0.570
Education, y	15.9±2.7	16.1±2.8	15.7±2.5	t (99.4)=0.82/Cohen's d=0.16	0.413
Vascular risk factors*
Hypertension	67 (66.3)	32 (62.7)	35 (70.0)	χ2 (1, N=101)=0.60/Cramer's V=0.08	0.441
Antihypertensive medication	64 (63.4)	32 (62.7)	32 (64.0)	χ2 (1, N=101)=0.02/Cramer's V=0.01	0.896
BMI≥30	20 (20.0)	8 (15.7)	12 (24.5)	χ2 (1, N=100)=1.21/Cramer's V=0.11	0.271
Diabetes	16 (15.8)	8 (15.7)	8 (16.0)	χ2 (1, N=101)=0.00/Cramer's V=0.00	0.966
Hypercholesterolemia	81 (80.2)	40 (78.4)	41 (82.0)	χ2 (1, N=101)=0.20/Cramer's V=0.04	0.653
Atrial fibrillation	9 (8.9)	4 (7.8)	5 (10.0)	FET; OR=0.77, 95% CI [0.14 to 3.82]	0.740
Current or former smoker	43 (42.2)	22 (42.3)	21 (42.0)	χ2 (1, N=102)=0.00/Cramer's V=0.00	0.975
BP profile
No. of BP measures	12 (26)	12 (6 to 36)	10 (4 to 15)	W=1006/r=0.195	0.049
SBP	129.2±10.2	131.1±10.0	127.2±10.2	t (99.5)=1.95/Cohen's d=0.39	0.054
DBP	72.9±6.1	73.5±6.0	72.3±6.3	t (99.3)=1.00/Cohen's d=0.20	0.321
Cognitive status at baseline
Subjective cognitive decline	11 (10.8)	3 (5.8)	8 (16.0)	FET; OR=0.32, 95% CI [0.05 to 1.46]	0.119
Mild cognitive impairment	91 (89.2)	49 (94.2)	42 (84.0)
Cognitive scores
Global cognition (MMSE)†	27.0±2.3	26.8±2.1	27.3±2.5	t (95.0)=−0.97/Cohen's d=−0.19	0.336
Executive function	−0.62±1.14	−0.73±1.19	−0.50±1.08	t (97.0)=−0.99/Cohen's d=−0.20	0.326
Attention/processing speed†	−0.15±0.82	−0.28±0.95	−0.02±0.64	t (85.7)=−1.60/Cohen's d=−0.32	0.113
Language/semantics	−0.60±1.32	−0.72±1.33	−0.48±1.32	t (98.0)=−0.88/Cohen's d=−0.18	0.378
Memory	−0.72±1.31	−0.86±1.34	−0.58±1.27	t (97.8)=−1.08/Cohen's d=−0.22	0.283
MRI markers of small vessel disease
Lobar cerebral microbleed count	0 (0–16)	16 (3 to 67)	0 (0)	W=50/r=0.890	<0.001
Presence of cortical superficial siderosis	22 (21.6)	22 (42.3)	0 (0)	FET; OR=∞, 95% CI [8.16 to ∞]	<0.001
Presence of lobar lacunes	24 (23.8)	15 (28.8)	9 (18.4)	χ2 (1, N=101)=1.53/Cramer's V=0.00	0.216
Presence of deep lacunes	16 (15.8)	10 (19.2)	6 (11.2)	χ2 (1, N=101)=0.92/Cramer's V=0.00	0.337
Presence of cortical cerebral microinfarcts	22 (22.8)	17 (33.3)	6 (12.0)	χ2 (1, N=101)=6.53/Cramer's V=0.00	0.011
nWMH volume % ICV	0.51±0.70	0.63±0.76	0.39±0.63	t (96.1)=1.76/Cohen's d=0.35	0.082
PSMD, 10−4 mm2/s	3.96±0.89	4.24±0.95	3.67±0.73	t (95.1)=3.39/Cohen's d=0.47	0.001
Brain parenchymal fraction	0.64±0.05	0.63±0.04	0.65±0.05	t (95.1)=−2.38/Cohen's d=−0.47	0.019

Values are number (%), median (interquartile range), or mean±SD.

% ICV indicates percent of intracranial volume; BMI, body mass index; BP, blood pressure; DBP, diastolic blood pressure; FET, Fisher exact test; MCS, mild cognitive symptoms; MMSE, Mini‐Mental State Examination; nWMH, normalized white matter hyperintensity; OR, odds ratio; PSMD, peak width of skeletonized mean diffusivity; r, rank‐biserial correlation; SBP, systolic blood pressure; and W, Wilcoxon rank‐sum test.

^*Missing data on vascular risk factors in 1, and BMI in participants.

^† Global cognition and processing speed scores missing in one participant.

Hypertension was diagnosed, and antihypertensive medications were administered in 66.3% and 63.4% of participants, respectively. The prevalence of hypertension and treatment regimen for hypertension did not differ between the probable CAA group and the non‐CAA group (P=0.667). Among participants taking antihypertensive medication, 1, 2, and 3 or more drugs were used in 29 (45.3%), 25 (39.1%), and 10 (15.6%), respectively.

Low, medium, and high BP variability were defined as coefficient of variation values <0.07, 0.07 to 0.12, and >0.12, corresponding to the first, median, and third tertile intervals, respectively (Figure S1). Among participants with high coefficient of variation, most systolic BP values were distributed within 15 to 35 mm Hg of the individual's mean systolic BP. In contrast, in low BP settings, systolic BP values were typically distributed within 2.5 to 12 mm Hg of the mean. Notably, higher mean BP is associated with increased coefficient of variation.

Associations Between Long‐Term BP Variability and White Matter Integrity

We found an association between PSMD and systolic (β=0.39, 95% CI [0.16–0.51], P<0.001) and diastolic BPV (β=0.20, 95% CI [0.03–0.38], P=0.022), adjusted for age (Figure 3B and 3C). In univariate analysis, PSMD was associated with systolic BPV, mean BP, age, total CAA severity score, and brain volume. Only systolic BPV but not mean BP remained a strong predictor of PSMD in a multivariable model driven by the probable CAA group (Table 2). Age was a predictor of PSMD in non‐CAA but not the probable CAA group.

Figure 3. Associations between blood pressure variability and white matter microstructural integrity in participants with and without probable CAA.

A, Interaction of probable CAA and non‐CAA participants on the association between systolic BP variability and PSMD controlling for age (PSMD=0.383+0.042*Age+9.329*[SBP_CV*Probable_CAA]). Linear regression models for (B) systolic and (C) diastolic BP variability and PSMD, adjusted for age. Shaded areas represent 95% CIs. BP indicates blood pressure; CAA, cerebral amyloid angiopathy; BP, diastolic BP; PSMD, peak width of skeletonized mean diffusivity; and SBP, systolic BP.

Table 2.

Associations Between White Matter Integrity and Blood Pressure Variability in Univariate and Multivariable Models

Variable	Univariate		Adjusted*
	β (95% CI)	P value	β (95% CI)	P value
All participants (n=102)
Systolic BP CoV	0.49 (0.32 to 0.67)	<0.001‡	0.22 (0.06 to 0.39)	0.010†
Systolic BP mean	0.25 (0.06 to 0.45)	0.010†	0.02 (−0.13 to 0.18)	0.790
Age, y	0.46 (0.29 to 0.64)	<0.001‡	0.18 (−0.02 to 0.39)	0.087
Sex, men	−0.01 (−0.21 to 0.19)	0.934	−0.01 (−0.16 to 0.14)	0.925
Diabetes	0.14 (−0.04 to 0.32)	0.141	0.13 (−0.02 to 0.28)	0.102
Total CAA score	0.40 (0.22 to 0.58)	<0.001‡	0.29 (0.14 to 0.44)	<0.001‡
BPF	−0.49 (−0.66 to −0.32)	<0.001‡	−0.21 (−0.42 to 0.00)	0.053
Probable CAA (n=52)
Systolic BP CoV	0.55 (0.32 to 0.78)	<0.001‡	0.34 (0.09 to 0.19)	0.009‡
Systolic BP mean	0.29 (0.03 to 0.56)	0.035†	0.14 (−0.08 to 0.36)	0.213
Age, y	0.32 (0.05 to 0.58)	0.022†	−0.04 (−0.32 to 0.23)	0.767
Sex, men	−0.25 (−0.52 to 0.01)	0.070	−0.01 (−0.35 to 0.08)	0.220
Diabetes	0.43 (0.18 to 0.68)	0.001‡	0.16 (−0.07 to 0.40)	0.168
Total CAA score	0.24 (−0.03 to 0.51)	0.087	0.32 (0.12 to 0.51)	0.003
BPF	−0.47 (−0.71 to −0.23)	<0.001‡	−0.40 (−0.67 to −0.12)	0.007‡
Non‐CAA (n=50)
Systolic BP CoV	0.28 (0.01 to 0.56)	0.046†	−0.03 (−0.28 to 0.21)	0.788
Systolic BP mean	0.10 (−0.18 to 0.38)	0.492	−0.19 (−0.43 to 0.05)	0.125
Age, y	0.65 (0.43 to 0.86)	<0.001‡	0.58 (0.25 to 0.91)	0.001‡
Sex, men	0.22 (−0.06 to 0.49)	0.132	0.12 (−0.10 to 0.35)	0.292
Diabetes	0.25 (−0.02 to 0.53)	0.074	0.20 (−0.04 to 0.44)	0.112
Total CAA score	0.35 (0.09 to 0.62)	0.012†	0.15 (−0.09 to 0.39)	0.226
BPF	−0.45 (−0.70 to −0.19)	0.001‡	−0.05 (−0.37 to 0.27)	0.751

The primary outcome measure was white matter integrity estimated using peak with of skeletonized mean diffusivity (PSMD). The predictive variables in simple regression models included systolic BP CoV, systolic BP mean, age, sex, diabetes, total CAA severity score, and BPF. All listed covariates were included in the adjusted multivariable model.

β indicates standardized regression coefficient; BP CoV, blood pressure coefficient of variation; BPF, brain parenchymal fraction; and CAA, cerebral amyloid angiopathy.

^*Adjusted R²=39.1% for all participants, 51.7% for probable CAA, and 41.5% for non‐CAA group.

^† P<0.05.

^‡ P<0.01.

Dose‐Dependent Association Between BP Variability and CAA Imaging Markers

Systolic BPV was associated with markers of brain injury independent of age, sex, mean BP, and diabetes (Table 3). BPV was found to have a significant association with normalized white matter hyperintensity volume (β=0.22 [0.04–0.41], P=0.021) and an even stronger association with PSMD (β=0.31 [0.11–0.47], P<0.001). We also found that BPV was associated with the presence of cortical cerebral microinfarcts (OR=1.96 [1.10–3.74], P=0.030) and lobar lacunes (OR=2.03 [1.13, 3.89], P=0.023). The highest quartile compared with the lowest quartile BPV was driving the association with PSMD (β=0.72 [0.24–1.19], P=0.004) and cortical cerebral microinfarcts (OR=7.47 [1.50–57.27], P=0.024), but not lobar lacunes, which only showed an association trend (OR=3.09 [0.78–13.82], P=0.119).

Table 3.

Association Between Long‐Term Systolic BPV and CAA Imaging Markers

	Systolic BPV by quartile							Continuous systolic BPV
	Q1	Q2		Q3		Q4
		β (95% CI)	P value	β (95% CI)	P‐value	β (95% CI)	P value	β (95% CI)	P value	aR2, %
Cerebral microbleed count*	Ref.	0.21 (−0.34 to 0.77)	0.453	0.49 (−0.09 to 1.06)	0.101	0.22 (−0.37 to 0.82)	0.464	0.11 (−0.12 to 0.33)	0.364	5.4
nWMH*	Ref.	−0.18 (−0.49 to 0.39)	0.442	−0.04 (−0.25 to 0.67)	0.882	0.44 (−0.06 to 0.93)	0.086	0.22 (0.04 to 0.41)†	0.021†	40.4
PSMD	Ref.	−0.05 (−0.52 to 0.36)	0.724	0.21 (−0.16 to 0.77)	0.200	0.72 (0.24 to 1.19)‡	0.004‡	0.31 (0.11 to 0.47)‡	<0.001‡	26.9

		OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value	aR2, %
Cortical superficial siderosis	1.00	0.80 (0.19 to 3.16)	0.748	0.86 (0.18 to 3.97)	0.850	1.59 (0.38 to 6.88)	0.521	1.21 (0.69 to 2.17)	0.508	8.2
Cortical cerebral microinfarcts	1.00	3.78 (0.75 to 28.24)	0.132	2.97 (0.54 to 23.28)	0.234	7.47 (1.50 to 57.27)†	0.024†	1.96 (1.10 to 3.74)†	0.030†	14.9
Lobar lacunes	1.00	0.37 (0.05 to 2.05)	0.277	1.23 (0.28 to 5.71)	0.786	3.09 (0.78 to 13.82)	0.119	2.03 (1.13 to 3.89)†	0.023†	23.1
Deep lacunes	1.00	5.15 (0.60 to 113.98)	0.180	2.05 (0.22 to 45.23)	0.561	7.33 (1.06 to 148.92)	0.083	1.87 (0.97 to 3.85)	0.071	23.9

BPV indicates blood pressure variability; CAA, cerebral amyloid angiopathy; PSMD, peak width of skeletonized mean diffusivity; and WMH, white matter hyperintensity.

^*Values log‐transformed. Data represent standardized beta coefficients (95% CI), odds ratio (OR) values (95% CI) and P values. All regression models adjusted for age, sex, diabetes, and mean blood pressure (n=102). Nagelkerke R² was calculated for logistic regression explained variance.

^† P<0.05.

^‡ P<0.01.

The presence of probable CAA strengthened the association between systolic BPV and PSMD (β=0.19 [0.03–0.35], P for interaction=0.023; Figure 3A). A similar analysis with normalized white matter hyperintensity volume as an outcome did not show a significant interaction between probable CAA and BPV (β=0.17 [−0.01 to 0.36], P for interaction=0.065). There was no interaction between probable CAA and BPV to predict total cerebral microbleed count, cerebral cortical microinfarcts, lobar, and deep lacunes (data not shown).

Longitudinal Cognitive Decline and BPV

In patients who had longitudinal neuropsychological data (n=59), we found that higher visit‐to‐visit BPV over 5 years predicted an annual decline in global cognitive function and processing speed (Figure 4) that remained significant after adjusting for age, sex, and mean BP.

Figure 4. Associations between 5‐year BP variability and annualized cognitive decline in participants with and without probable CAA.

Linear regression models between systolic BP variability and cognitive decline in (A) global cognition, (B) processing speed, (C) executive function, and (D) memory. Standardized β coefficients with 95% CIs in the shaded areas are presented. Note that Y‐axis scales vary for better visualization. BP indicates blood pressure; CAA, cerebral amyloid angiopathy; and MMSE, mini‐mental state examination.

Sensitivity Analyses

The excluded non‐demented participants had similar demographic and clinical characteristics to the included participants, except for a higher proportion of female sex and hypertension (Table S1). There were no changes in effect sizes in models adjusted for sex, history of hypertension, antihypertensive medication use, their number, frequency of BP measurements, CAA severity, and brain atrophy (data not shown). The effect sizes remained robust after including possible CAA participants (n=15). Overall, only 6.4% of the included BP measurements were obtained by a health care provider in a home setting. Mean BP did not differ significantly at the intra‐individual level when analyses included versus excluded home‐based measurements (t=1.306, P=0.239). Likewise, the associations observed in regression models were unchanged when home BP measurements were included or excluded (Tables S2 and S3).

DISCUSSION

In this prospective cohort study, we found that long‐term BPV predicted altered white matter integrity, particularly in the presence of probable CAA, suggesting that CAA‐related brain injury might be partly driven by increased long‐term BP fluctuations. This introduces BPV as a novel risk factor for cognitive decline in patients with CAA. We also observed a dose‐dependent association between increased BPV and cortical cerebral microinfarcts and an association between higher BPV and lobar lacunes, emerging imaging markers of CAA. ³¹ , ⁴¹ , ⁴² , ⁴³ These findings imply that BPV might be linked to a wide spectrum of silent ischemic brain injuries in cortical and subcortical regions, contributing to altered white matter integrity and potentially leading to white matter hyperintensities.

The debate on whether BPV stems from dysautonomia related to neurodegenerative mechanisms remains unresolved. Although we cannot entirely exclude the possibility of reverse causation, our rigorous adjustments for brain atrophy and CAA severity do not support the idea that BPV results from preceding brain injury. Second, the longitudinal cognitive data collected after BPV assessment suggest that greater BPV precedes and is associated with increased cognitive decline, rather than the reverse. Additionally, our study focused on participants with subjective cognitive decline and MCI, excluding those with mild dementia, reducing the likelihood that advanced neurodegeneration influenced this association. Existing literature does not support the link between autonomic dysfunction and AD or vascular dementia, ⁴⁴ making it less likely that BPV in CAA arises from dysautonomia. Finally, in light of Bradford Hill's criteria for evaluating causality, although preliminary, our findings suggest a temporally plausible (BPV measured before baseline imaging and cognitive assessment) and dose‐dependent association between BPV and white matter injury. Although with caution, we could interpret these preliminary findings as a biological gradient in favor of a causal association, ⁴⁵ warranting further studies to confirm or infirm consistency across different populations and clinical settings.

Our data also suggest that elevated BPV could be a risk factor for vascular cognitive impairment, predicting domain‐specific cognitive decline, particularly in global cognition and processing speed. The association between higher BPV and cortical cerebral microinfarcts, known to affect adjacent gray matter, white matter tracts and predict worse cognition, may contribute to this decline in cognitive function. ³⁰ , ⁴⁶ This study complements previous population‐based research by demonstrating that long‐term BPV over 5 years predicts clinically detectable cognitive decline. ¹⁶ , ²¹ Targeting this potential risk factor early in the disease course may help prevent or slow cognitive decline.

To capture white matter integrity, we used a quantitative diffusor tensor imaging technique known as PSMD, which is particularly sensitive to CAA‐related changes and has a high repeatability and reproducibility. This makes PSMD a robust tool for assessing microstructural brain injury relevant to vascular brain disease. ²⁴ , ²⁵ , ²⁶ , ²⁹ Our study adds to the growing body of literature linking BPV to subclinical brain injury, particularly providing novel evidence in CAA‐affected populations.

We discovered a dose‐dependent association between higher BPV and presence cortical cerebral microinfarcts, markers of CAA‐related vascular dysfunction. ³¹ , ⁴³ These findings support the emerging mechanistic concepts that CAA‐related brain lesions result from vascular dysfunction caused by amyloid deposition long before clinical manifestations of cerebral hemorrhage and cognitive impairment. ⁶ , ⁴² , ⁴⁷ Pathologic reports of CAA have emphasized that the replacement of vascular smooth muscle cells with a rigid ring of beta‐amyloid might be responsible for altered vascular physiology. ⁴² , ⁴⁸ The lack of association between BPV and hemorrhagic imaging markers of CAA aligns with previous research, indicating that hemorrhagic and ischemic lesions in CAA may be caused by distinct underlying CAA‐related mechanisms, where the ischemic injury is related to focally beta‐amyloid‐positive cortical vessels. ⁴³ In addition, hypoperfusion due to sudden drops (increased variability) in BP has been suggested as an additional potential cause for cortical cerebral microinfarcts. ⁴⁹

Alternative explanations include the interaction between CAA and hypertensive SVD (arteriosclerosis) at the pathophysiological level. ⁵⁰ Given the common co‐occurrence of these 2 SVD types, further studies are needed to disentangle their specific effects on brain injury.

Our findings have potential implications for preventative strategies aimed at mitigating the risk of dementia. While elevated BP is an established risk factor, our data suggest that BPV may offer an additional avenue for intervention, particularly in populations susceptible to vascular and mixed dementias. Emerging evidence suggests drug‐class‐specific effects in reducing BPV, particularly in sporadic SVD, though more research is needed to confirm these findings. ¹⁴ , ⁵¹ Given that the CAA‐related pathophysiological cascade begins decades before the onset of cognitive impairment, ⁴² targeting interventions at the earliest stages of the disease will be crucial.

Our study results are generalizable to memory clinic and CAA populations with mild cognitive symptoms, including subjective cognitive complaints and mild cognitive impairment. Our cohort included mostly White individuals with relatively high education levels and only mild cognitive symptoms, limiting the generalizability of our findings to populations with different socio‐economic background, more advanced stages of cognitive decline, and other clinical settings such as inpatient, community‐dwelling, or populations without cognitive complaints.

Our study's strengths include its prospective cohort design, a relatively high number of longitudinal BP measurements, and standardized advanced diffusion neuroimaging protocol, which enable the use of smaller sample sizes. However, limitations such as the absence of Alzheimer's disease biomarkers and the lack of APOE genotype data should be acknowledged.

Observational data suggests that e4 allele carriers may be at higher risk of neurovascular dysfunction. ⁵² Because CAA participants are known to have a high prevalence of the APOE e4 allele, that may be one of the underlying factors increasing the association of BPV‐related injury in this population. Moreover, our derivation cohort included 70% male participants. After applying the predefined inclusion criteria, the final study cohort included 4/5 male participants, due to a higher prevalence of incomplete MRI scans and missing BP data among female participants. We do not have a clear explanation for this gender‐related bias. Notably, adjustments for sex in our analyses did not reveal any associations with the study outcome based on this variable. Recent emerging data on sex differences in CAA suggest that male participants may experience earlier disease onset and have more hemorrhagic lesions, ⁵³ which could partially explain the higher male recruitment in the study's derivation cohort. Finally, we only studied diurnal BP measurements. Further research should address short‐term and circadian BP variability. Overall, our findings require further confirmation in external cohorts, including culturally and sex‐diverse cohorts, preferably with APOE genotyping, as these individuals may most likely benefit from preventive interventions targeting BPV.

CONCLUSIONS

Our findings underscore the potential role of BPV in cognitive decline and CAA‐related vascular brain injury. Cognitive decline in patients with high BPV may be driven by silent ischemic brain injury altered white matter integrity, cortical cerebral microinfarcts, and lobar lacunes. Future research should focus on specific antihypertensive treatments that might be beneficial in controlling BPV, thereby offering a promising avenue for dementia prevention.

Sources of Funding

L.S. was supported by the Swiss National Science Foundation (P2GEP3_191584, 10TS1‐_235536) under the framework the co‐fund partnership of Transforming Health and Care Systems, THCS, (GA N° 101095654 of the EU Horizon Europe Research and Innovation Programme), Alzheimer's Association award (AACSF‐22‐922 907), and the University of Geneva investigator award.

Disclosures

None.

Supporting information

Data S1. Supplemental Methods

Tables S1–S3

Figure S1