Clinical Predictors of Postmortem Amyloid and Nonamyloid Cerebral Small Vessel Disease in Middle-Aged to Older Adults

Abstract

Background and Objectives

Sporadic cerebral small vessel disease (CSVD) is a class of important pathologic processes known to affect the aging brain and to contribute to cognitive impairment. We aimed to identify clinical risk factors associated with postmortem CSVD in middle-aged to older adults.

Methods

We developed and tested risk models for their predictive accuracy of a pathologic diagnosis of nonamyloid CSVD and cerebral amyloid angiopathy (CAA) in a retrospective sample of 160 autopsied cases from the Edinburgh Brain Bank. Individuals aged 40 years and older covering the spectrum of healthy aging and common forms of dementia (i.e., highly-prevalent etiologies such as Alzheimer disease (AD), vascular cognitive impairment (VCI), and mixed dementia) were included. We performed binomial logistic regression models using sample splitting and cross-validation methods. Demographics, lifestyle habits, traditional vascular risk factors, chronic medical conditions, APOE4, and cognitive status were assessed as potential predictors.

Results

Forty percent of our sample had a clinical diagnosis of dementia (AD = 33, VCI = 26 and mixed = 5) while others were cognitively healthy (n = 96). The mean age at death was 73.8 (SD 14.1) years, and 40% were female. The presence of none-to-mild vs moderate-to-severe nonamyloid CSVD was predicted by our model with good accuracy (area under the curve [AUC] = 0.84, sensitivity [SEN] = 72%, specificity [SPE] = 95%), with the most significant clinical predictors being age, history of cerebrovascular events, and cognitive impairment. The presence of CAA pathology was also predicted with high accuracy (AUC = 0.86, SEN = 93%, SPE = 79%). Significant predictors included alcohol intake, history of cerebrovascular events, and cognitive impairment. In a subset of atypical dementias (n = 24), our models provided poor predictive performance for both nonamyloid CSVD (AUC = 0.50) and CAA (AUC = 0.43).

Discussion

CSVD pathology can be predicted with high accuracy based on clinical factors in patients within the spectrum of AD, VCI, and normal aging. Whether this prediction can be enhanced by the addition of fluid and neuroimaging biomarkers warrants additional study. Improving our understanding of clinical determinants of vascular brain health may lead to novel strategies in the prevention and treatment of vascular etiologies contributing to cognitive decline.

Classification of Evidence

This study provides Class II evidence that selected clinical factors accurately distinguish between middle-aged to older adults with and without cerebrovascular small vessel disease (amyloid and nonamyloid) pathology.

Introduction

Sporadic cerebral small vessel disease (CSVD) is a class of pathologic processes known to affect the aging brain.¹ The 2 most common forms are nonamyloid CSVD (NA-CSVD) and cerebral amyloid angiopathy (CAA). It is estimated that about 80% of individuals older than 65 years and nearly all individuals older than 90 years show at least some clinical or radiologic evidence of CSVD.²

However, most dedicated epidemiologic and etiologic studies of CSVD rely on surrogate markers such as modern neuroimaging techniques.³ While magnetic resonance imaging (MRI)–detected white matter hyperintensities (WMH), lacunes, cerebral microbleeds (CMB), and enlarged perivascular spaces (PVS) are highly suggestive of underlying CSVD, postmortem examination remains the gold standard for assessing the presence and severity of histopathologic lesions affecting the brain.

To date, no studies appear to have examined the clinical determinants of neuropathologically confirmed NA-CSVD and CAA, most probably due to the limited availability of brain tissue data. Understanding the contribution of lifestyle and traditional vascular risk factors on these 2 distinct subtypes of CSVD would provide insights toward the development of preventive strategies for cognitive decline and dementia, as increasing evidence points toward shared histologic substrates between Alzheimer and cerebrovascular diseases,⁴ the 2 most important causes of cognitive impairment in the elderly population.

The aim of this study was to identify clinical risk factors associated with postmortem CSVD in middle-aged to older adults with and without cognitive impairment. To this end, we developed and tested risk models for their predictive accuracy of a pathologic diagnosis of NA-CSVD and CAA in retrospective samples of autopsied cases. Based on current knowledge, we hypothesized that age, hypertension, and diabetes would be the strongest predictors of NA-CSVD, whereas age and cognitive impairment would be the most significant predictors of CAA.

Methods

Participants

Our cohort was selected from the Edinburgh Brain Bank (EBB; The University of Edinburgh, Scotland). The EBB was explored to identify participants of age 40 years and older covering the spectrum of healthy aging through to common forms of dementia, focusing on Alzheimer disease (AD) and vascular etiologies, who consented to brain donation between January 2015 and August 2021. All participants meeting these criteria were included, blinded for their cognitive status and cause of death. Exclusion criteria included metabolic and toxic encephalopathies, prion diseases, atypical dementias, and other sporadic or genetic neurodegenerative diseases such as multiple sclerosis, amyotrophic lateral sclerosis, and Huntington disease.

Clinical Data Collection

Clinical data were accessed from the database of the EBB. The participant information was systematically obtained from the referring physician or any NHS electronic medical records on death. Extracted data included demographics (age at death, self-reported sex at birth), lifestyle habits (smoking, alcohol, drugs), traditional vascular risk factors (hypertension, dyslipidemia, diabetes, obesity), medical history (cerebrovascular events, coronary artery disease, peripheral artery disease, atrial fibrillation, chronic kidney disease, chronic liver disease, obstructive sleep apnea, cancer), and apolipoprotein E ε4 (APOE4) status, whenever available. Cognitive status (either dementia or no dementia) and the probable cause of dementia (AD, VCI, or mixed) were determined on documented clinical history and functional status at the time of death by treating physicians, with or without formal cognitive screening tests available. No participant was attributed a diagnosis of mild cognitive impairment in our retrospective series.

Brain Sampling and Histologic Procedures

Detailed brain sampling and histologic procedures can be found in the eMethods.

Neuropathologic Assessment

NA-CSVD

The assessment scale developed and applied to our cases for evaluation of NA-CSVD is presented in eFigure 1. Briefly, we looked at 6 H&E-stained sections: bilateral frontal white matter, central white matter, and basal ganglia. We graded arteriolosclerosis using a histologic semiquantitative scale ranging from 0 to 3 (0 = normal, 1 = mild, 2 = moderate, and 3 = severe) mainly defined as the degree of smooth muscle cells loss rather than lumen stenosis. All other NA-CSVD lesions (lipohyalinosis, fibrinoid necrosis, perivascular space dilatation, venous collagenosis, and microinfarcts) were dichotomously reported as being either absent or present. All histologic slides were rated by one observer (C.D.-T.). A sample of these slides (n = 66; 6%) was also independently assessed by a second observer (C.S.) for evaluation of inter-rater reliability. Assessors were blinded to clinical data. Any conflictual score was resolved by consensus.

Amyloid and Other Pathologies

Scores for CAA, AD pathology, and other protein deposits (Lewy bodies and pTDP-43) were extracted from the original neuropathologic reports produced by the local neuropathologists (C.S. for most cases). These were based on tissue sections from frontal, temporal, parietal, occipital, and cerebellar brain regions, assessed for the presence of leptomeningeal, parenchymal, and capillary CAA using amyloid immunohistochemistry.⁵ A semiquantitative scale (0 = absent, 1 = scant amyloid β (Aβ) deposition, 2 = some circumferential Aβ, and 3 = widespread circumferential Aβ) was applied for leptomeninges and parenchyma and a dichotomic scale (0 = absent and 1 = present) for capillaries. More detailed neuropathologic assessment is presented in the eMethods.

Statistical Analyses

All statistical analyses were conducted in RStudio (rstudio.com/; R version 4.2.1).

Descriptive Analyses

We performed a comparison of clinical characteristics of participants and neuropathologies across pathologic CSVD diagnosis (NA-CSVD only, CAA only, both, neither) and clinical groups (AD, VCI, mixed dementia and controls) using one-way ANOVA test for continuous variables, Kruskal-Wallis test by ranks for ordinal variables and Pearson χ² for binary and categorical variables. Inter-rater reliability of NA-CSVD rating was tested by calculating weighted Kappa's coefficient between the 2 observers (C.S. and C.D.-T.). We used Cohen interpretation benchmarks to evaluate the extent of agreement between raters.⁶ A p-value of <0.05 was defined to determine statistical significance.

Model Training

We developed 2 distinct multivariate models to predict NA-CSVD (here represented by arteriolosclerosis) and CAA (either leptomeningeal and/or parenchymal) using binomial logistic regression. For each of these models, we randomly selected 80% of our total study sample as our training set. The dependent variables were dichotomized into none/low (not significant) vs moderate/severe (significant) NA-CSVD and absent vs present CAA because of unbalanced distribution of our participants across degrees of pathology. This simplification is also presumed to minimize the impact of subjective pathologic scoring and to improve inter-rater reliability while providing a better clinical distinction between normal and pathologic cerebrovascular aging. Selected independent variables included age, sex, smoking, alcohol consumption, hypertension, dyslipidemia, diabetes mellitus, cerebrovascular events, coronary artery disease, peripheral artery disease, atrial fibrillation, chronic kidney disease, cancer, and cognitive status. APOE genotype and obesity were not included in our models because of the high number of missing values (60% and 38%, respectively). Drug use, chronic liver disease, and obstructive sleep apnea were also removed from the models because of the paucity of affected individuals in our sample. To assure internal validity of our models, we performed a 10-fold cross-validation such that randomly 90% of data were considered a training set and 10% as a validation set, for a total of 10 iterations per model (see eFigure 2). The results are reported as odd ratios (OR) with 95% confidence intervals (95% CI).

Model Testing

For both predictive models, our test set included the remaining 20% of our total study sample. Slight aleatory variations in test group sizes were expected with the use of the basic sample function in R, in which a vector of probability weights for obtaining the elements of the vector being sampled was defined (0.8 and 0.2 for train and test sets, respectively). Models were run with the test sets and discrimination (receiver operating characteristic [ROC] curve and area under the curve [AUC]), and classification performance (confusion matrix and accuracy, sensitivity, and specificity) was assessed. The formulas used for classification performance assessment (a to f) are presented below.

$$\left(\mathrm{a}\right)\hspace{0.17em}{\text{Sensitivity}}_{\text{NA}‐\text{CSVD}}=\frac{\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{moderate}\hspace{0.17em}\text{to}\hspace{0.17em}\text{severe}\hspace{0.17em}\text{NA}‐\text{CSVD}}{\left(\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{moderate}\hspace{0.17em}\text{to}\hspace{0.17em}\text{severe}\hspace{0.17em}\text{NA}‐\text{CSVD}\right)+\left(\text{wrongly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{none}\hspace{0.17em}\text{to}\hspace{0.17em}\text{mild}\hspace{0.17em}\text{NA}‐\text{CSVD}\right)\hspace{0.17em}}$$

$$\left(\mathrm{b}\right){\hspace{0.17em}\text{Sensitivity}}_{\text{CAA}}=\frac{\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{presence}\hspace{0.17em}\text{of}\hspace{0.17em}\text{CAA}}{\left(\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{presence}\hspace{0.17em}\text{of}\hspace{0.17em}\text{CAA}\right)+\left(\text{wrongly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{absence}\hspace{0.17em}\text{of}\hspace{0.17em}\text{CAA}\right)\hspace{0.17em}}$$

$$\left(\mathrm{c}\right)\hspace{0.17em}{\text{Specificity}}_{\text{NA}‐\text{CSVD}}=\frac{\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{none}\hspace{0.17em}\text{to}\hspace{0.17em}\text{mild}\hspace{0.17em}\text{NA}‐\text{CSVD}}{\left(\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{none}\hspace{0.17em}\text{to}\hspace{0.17em}\text{mild}\hspace{0.17em}\text{NA}‐\text{CSVD}\right)+\left(\text{wrongly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{moderate}\hspace{0.17em}\text{to}\hspace{0.17em}\text{severe}\hspace{0.17em}\text{NA}‐\text{CSVD}\right)\hspace{0.17em}}$$

$$\left(\mathrm{d}\right)\hspace{0.17em}{\text{Specificity}}_{\text{CAA}}=\frac{\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{absence}\hspace{0.17em}\text{of}\hspace{0.17em}\text{CAA}}{\left(\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{absence}\hspace{0.17em}\text{of}\hspace{0.17em}\text{CAA}\right)+\left(\text{wrongly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{presence}\hspace{0.17em}\text{of}\hspace{0.17em}\text{CAA}\right)\hspace{0.17em}}$$

$$\left(\mathrm{e}\right)\hspace{0.17em}{\text{Accuracy}}_{\text{NA}‐\text{CSVD}}=\frac{\left(\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{moderate}\hspace{0.17em}\text{to}\hspace{0.17em}\text{severe}\hspace{0.17em}\text{NA}‐\text{CSVD}\right)+\left(\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{none}\hspace{0.17em}\text{to}\hspace{0.17em}\text{mild}\hspace{0.17em}\text{NA}‐\text{CSVD}\right)}{\text{total}\hspace{0.17em}\text{test}\hspace{0.17em}\text{sample}\hspace{0.17em}}$$

$$\left(\mathrm{f}\right)\hspace{0.17em}{\text{Accuracy}}_{\text{CAA}}=\frac{(\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{presence}\hspace{0.17em}\text{of}\hspace{0.17em}\text{CAA})+(\text{correctly}\hspace{0.17em}\text{classified}\hspace{0.17em}\text{as}\hspace{0.17em}\text{absence}\hspace{0.17em}\text{of}\hspace{0.17em}\text{CAA})}{\text{total}\hspace{0.17em}\text{test}\hspace{0.17em}\text{sample}\hspace{0.17em}}$$

Missing data were rare and therefore complete case analyses were performed.

Post Hoc Exploratory Analyses

Some additional analyses were conducted a posteriori. More specifically, we applied our 2 predictive models to a sample of subjects with atypical dementia (n = 24) from the EBB. These cases included clinical diagnoses of Lewy body dementias (LBD; n = 10), frontotemporal lobar degeneration (FTLD; n = 11), posterior cortical atrophy (n = 2), and dementia of unknown etiology (n = 1). We looked at external validity of such clinical models to identify postmortem CSVD in patients presenting outside the usual spectrum of aging and highly prevalent etiologies of cognitive decline. Furthermore, we explored the potential of neuropathologic features for predicting a clinical diagnosis of dementia before death. More details on these additional analyses can be found in eAppendix 1 and eFigures 3 and 4.

Standard Protocol Approvals, Registrations, and Patient Consents

The study was approved by the ethics committee of the Centre intégré universitaire de santé et de services sociaux de la Capitale-Nationale (NSM-2021-2055). Data access was obtained under the Edinburgh Brain and Tissue Bank ethical approval from the East of Scotland Research Ethics Service REC1 (#21/ES/0087). Written informed consent for brain donation and research was obtained from all patients or their legal representatives.

Data Availability

Clinical and neuropathologic data are available on request from the corresponding author.

Results

Participants Clinical Characteristics

We identified a total of 160 autopsied individuals meeting inclusion criteria for this study (see eFigure 5). The demographic and clinical characteristics of participants are presented in Table 1 (grouped by pathologic CSVD diagnoses) and in eTable 1 (grouped by clinical diagnoses). Briefly, 40% of the individuals in our sample had a clinical diagnosis of dementia, either AD (n = 33), VCI (n = 26) or mixed dementia (n = 5), and 60% were cognitively healthy (CH, n = 96) before death. Pathologic evidence of CSVD was found in 74.4% of our cases, either NA-CSVD only (n = 36), CAA only (n = 19), or both NA-CSVD and CAA (n = 64), and 25.6% had no significant CSVD at postmortem examination (n = 41). The original study source of provenance in demented and nondemented participants is provided in eFigure 6. The mean age at death was 73.8 (SD 14.1) years, and 40% were female. Ethnicity was not documented. The mean postmortem delay for brain fixation was 70.3 (range 3–138) hours. The causes of death were highly variable, with the most frequent being intracerebral hemorrhage (25%), myocardial infarction (21%), bronchopneumonia (13%), end-stage dementia (11%), and cancer (10%). CSVD groups were older (mean age 79.4 years) with a higher proportion of female (45.4%) than the control group (mean age 57.5 years, 24.4% female). Among cases with available APOE genotype, the ε4 allele was more prevalent in those with evidence of CAA (66.7%) than other pathologic groups (22.2%). There were more past and current smokers in the CAA (52.6%) and control (51.2%) groups compared with NA-CSVD (27.8%) and mixed CSVD (17.2%). Alcohol consumption was more prevalent in CAA cases (57.9%) compared with others (32%–36%). Drug use was rarely reported in our sample but was present in 7.3% of our younger control group. Among known vascular risk factors, only hypertension and obesity differed between groups, most striking being a much higher prevalence of high blood pressure in NA-CSVD (63.9%) and mixed CSVD (50%) compared with CAA only (36.8%) and control (26.8%) groups. A history of cerebrovascular events (transient ischemic attack, ischemic stroke, or intracerebral hemorrhage) was more prevalent in NA-CSVD and mixed CSVD cases (65%), followed by CAA only cases (21.1%), compared with controls (only 2.4%). By contrast, coronary artery disease was highly represented in the control group (53.7%) and much less frequent in other pathologic groups (22.7%). Finally, there was a higher proportion of individuals with cancer in CSVD groups (26.9%) than controls (7.3%). Other medical conditions did not significantly differ between groups.

Table 1.

Demographic and Clinical Characteristics of Participants

Characteristic	Total	Pathologic CSVD diagnosis
		NA-CSVD onlya	CAA only	Both	Neither	p Value
Participants
n	160 (100)	36 (22.5)	19 (11.9)	64 (40)	41 (25.6)
Sex
Female	64 (40)	12 (33.3)	9 (47.4)	33 (51.6)	10 (24.4)	0.0317b
Age (y)
Age at death	73.8 (14.1)	79.1 (9.2)	74.5 (14.1)	81.1 (8.5)	57.5 (11.4)	<0.0001b
Clinical diagnosis
AD	33 (20.6)	4 (11.1)	6 (31.6)	23 (35.9)	0	<0.0001b
VCI	26 (16.3)	4 (11.1)	1 (5.3)	21 (32.8)	0
Mixed	5 (3.1)	2 (5.6)	0	3 (4.7)	0
Control	96 (60)	26 (72.2)	12 (63.2)	17 (26.6)	41 (100)
APOE genotype
2/2	0	0	0	0	0	<0.0001b
2/3	5 (3.1)	1 (2.8)	0	1 (1.6)	3 (7.3)
2/4	1 (0.6)	0	0	1 (1.6)	0
3/3	31 (19.4)	7 (19.4)	4 (21.1)	3 (4.7)	17 (41.5)
3/4	24 (15)	0	4 (21.1)	12 (18.8)	8 (19.5)
4/4	3 (1.9)	0	3 (15.8)	0	0
Unknown	96 (60)	28 (77.8)	8 (42.1)	45 (70.3)	13 (31.7)
Smoking
Never	108 (67.5)	26 (72.2)	9 (47.4)	53 (82.8)	20 (48.8)	0.0009b
Past	31 (19.4)	7 (19.4)	6 (31.6)	9 (14.1)	9 (22)
Current	21 (13.1)	3 (8.3)	4 (21.1)	2 (3.1)	12 (29.3)
Alcohol
None	101 (63.1)	23 (63.9)	8 (42.1)	42 (65.6)	28 (68.3)	0.0044b
Occasional	35 (21.9)	4 (11.1)	9 (47.4)	17 (26.6)	5 (12.2)
Moderate	11 (6.9)	2 (5.6)	1 (5.3)	2 (3.1)	6 (14.6)
Severe	13 (8.1)	7 (19.4)	1 (5.3)	3 (4.7)	2 (4.9)
Drugs
No	157 (98.1)	36 (100)	19 (100)	64 (100)	38 (92.7)	0.0310b
Yes	3 (1.9)	0	0	0	3 (7.3)
Vascular risk factors
HTN	73 (45.6)	23 (63.9)	7 (36.8)	32 (50)	11 (26.8)	0.0082b
DLP	21 (13.1)	7 (19.4)	2 (10.5)	7 (10.9)	5 (12.2)	0.6429
DM2	20 (12.5)	7 (19.4)	3 (15.8)	6 (9.4)	4 (9.8)	0.4524
Obesity	19/99 (19.2)	2 (5.6)	1 (5.3)	5 (7.8)	11 (26.8)	0.0050b
Medical history
CVD	70 (43.8)	23 (63.9)	4 (21.1)	42 (65.6)	1 (2.4)	<0.0001b
CAD	49 (30.6)	10 (27.8)	5 (26.3)	12 (18.8)	22 (53.7)	0.0020b
PAD	7 (4.4)	4 (11.1)	0	1 (1.6)	2 (4.9)	0.1112
AF	25 (15.6)	9 (25)	1 (5.3)	12 (18.8)	3 (7.3)	0.0870
CKD	13 (8.1)	3 (8.3)	3 (15.8)	6 (9.4)	1 (2.4)	0.3330
CHD	6 (3.8)	4 (11.1)	0	2 (3.1)	0	0.0501
OSA	0	0	0	0	0	NA
Cancer	35 (21.9)	11 (30.6)	7 (36.8)	14 (21.9)	3 (7.3)	0.0272b

Abbreviations: AD = Alzheimer disease; AF = atrial fibrillation; CAD = coronary artery disease; CHD = chronic hepatic disease; CKD = chronic kidney disease; CVD = cerebrovascular disease; CVSD = sporadic cerebral small vessel disease; DLP = dyslipidemia; DM2 = type 2 diabetes mellitus; HTN = hypertension; OSA = obstructive sleep apnea; PAD = peripheral artery disease; VCI = vascular cognitive impairment.

^aSignificant (moderate-to-severe) NA-CSVD only.

^bp Value < 0.05.

Neurodegenerative and Cerebrovascular Pathologies

Neuropathologic findings stratified by clinical diagnosis are presented in Table 2. The mean brain weight was significantly lower in dementia groups compared with CH individuals. AD and mixed dementia cases presented higher Aβ plaques and NFT pathologies, followed by VCI and then CH. Similarly, cortical Lewy bodies and pTDP-43 pathologies were more prevalent in AD and mixed cases than VCI and CH. Moderate-to-severe NA-CSVD and presence of CAA were highly represented across dementia groups, compared with CH. Among other vascular lesions, only perivascular space dilatation significantly varied between groups, with the highest prevalence seen in VCI. The distribution of individuals presenting with different levels of NA-CSVD and CAA severity according to age, sex, and hypertension status is presented in eFigures 7 and 8.

Table 2.

Neuropathologic Findings Stratified by Clinical Diagnosis

Characteristic	Total	Clinical diagnosis				p Value
		AD	VCI	Mixed	Control
Participants
n	160	33	26	5	96
Brain weight (g)	1312.2 (167.3)	1185.0 (145.7)	1278.2 (148.4)	1293.6 (173.2)	1366.2 (154.0)	<0.0001a
Diffuse Aβ plaques (Thal)	1 (0.4)	5 (5.5)	3 (1.5)	5 (4.5)	0 (0.1)	<0.0001a
NFT (Braak)	2 (0.6)	6 (6.6)	3 (2.6)	6 (6.6)	1 (0.1)	<0.0001a
Neuritic plaques (CERAD)	0 (0.0)	0 (0.2)	0 (0.1)	3 (3.3)	0 (0.0)	<0.0001a
Lewy bodies
Cortical subtype	21 (13.1)	10 (30.3)	2 (7.7)	2 (40.0)	7 (7.3)	0.0016a
Brainstem subtype	6 (3.8)	4 (12.1)	1 (3.8)	0 (0.0)	1 (1.0)	0.0359a
TDP-43	7 (4.4)	4 (12.1)	1 (3.8)	2 (40.0)	0 (0.0)	<0.0001a
Cerebral amyloid angiopathy
None	77 (48.1)	4 (12.1)	4 (15.4)	2 (40.0)	67 (69.8)	<0.0001a
Mild	27 (16.9)	7 (21.2)	6 (23.1)	0 (0.0)	14 (14.6)
Moderate	15 (9.4)	9 (27.3)	3 (11.5)	0 (0.0)	3 (3.1)
Severe	41 (25.6)	13 (39.4)	13 (50)	3 (60.0)	13 (13.5)
Nonamyloid CSVD
No arteriolosclerosis	23 (14.4)	0 (0.0)	0 (0.0)	0 (0.0)	23 (24.0)	<0.0001a
Mild arteriolosclerosis	37 (23.1)	6 (18.2)	1 (3.8)	0 (0.0)	30 (31.3)
Moderate arteriolosclerosis	50 (31.3)	17 (51.5)	11 (42.3)	4 (80.0)	18 (18.8)
Severe arteriolosclerosis	50 (31.3)	10 (30.3)	14 (53.8)	1 (20.0)	25 (26.0)
Lipohyalinosis	32 (20.0)	8 (24.2)	10 (6.3)	1 (20.0)	13 (13.5)	0.0382a
Fibrinoid necrosis	15 (9.4)	2 (6.1)	6 (23.1)	1 (20.0)	6 (6.3)	0.0473a
Perivascular space dilatation	109 (68.1)	26 (78.8)	25 (96.2)	4 (80.0)	54 (56.3)	0.0005a
Venous collagenosis	143 (55.0)	31 (93.9)	23 (88.5)	4 (80.0)	85 (88.5)	0.7339
Microinfarcts	30 (18.8)	10 (30.3)	5 (19.2)	2 (40.0)	13 (13.5)	0.1075
Mediacalcinosis	50 (31.3)	15 (45.5)	6 (23.1)	1 (20.0)	28 (29.2)	0.2218
Any ischemic lesion	40 (25.0)	11 (33.3)	9 (34.6)	2 (40.0)	18 (18.8)	0.1643
Any hemorrhagic lesion	45 (28.1)	8 (24.2)	10 (38.5)	1 (20.0)	26 (27.1)	0.6073

One-way ANOVA for brain weight; mean (SD).

Kruskall-Wallis for ABC scores; median (0.25, 0.75).

Pearson χ² test for categorical variables; count (%).

^ap Value < 0.05.

Inter-Rater Reliability for CSVD Assessment

We found moderate inter-rater agreement for the assessment of perivascular space dilatation, substantial agreement for venous collagenosis and ischemic lesions, and almost perfect agreement for arteriolosclerosis, lipohyalinosis, medial calcinosis, and hemorrhagic lesions. Weighted Cohen kappa coefficients for inter-rater reliability can be found in eTable 2.

Performance of Our Model to Predict Nonamyloid CSVD

The presence of none-to-mild vs moderate-to-severe NA-CSVD was predicted by our model with good accuracy (ACC = 84.6%, 95% CI = 69.5%–94.1%, AUC = 0.84, sensitivity [SEN] = 72%, specificity [SPE] = 95%; eTable 3 and eFigure 9). The most significant clinical predictors associated with increased risk of NA-CSVD were age (OR 1.08, 95% CI 1.01–1.18), history of cerebrovascular events (OR 19.54, 95% CI 3.32–166.17), and cognitive impairment (OR 18.30, 95% CI 3.54–139.02). More detailed results are presented in Figure 1.

Figure 1. Odds of Presenting Moderate to Severe Nonamyloid CSVD at Postmortem Examination (95% CI, p-Value).

Performance of Our Model to Predict Amyloid CSVD (or CAA)

The presence of CAA pathology was predicted by our model with high accuracy (ACC = 85.3%, 95% CI = 68.9%–95.1%, AUC = 0.86, SEN = 93%, SPE = 79%; eTable 3 and eFigure 10). Significant predictors associated with increased risk of CAA included occasional alcohol intake (OR 4.68, 95% CI 1.29–19.82), history of cerebrovascular events (OR 6.04, 95% CI 1.57–26.90), and cognitive impairment (OR 15.33, 95% CI 4.57–62.42). More detailed results are presented in Figure 2.

Figure 2. Odds of Presenting Cerebral Amyloid Angiopathy at Postmortem Examination (95% CI, p-Value).

Prediction of CSVD in Atypical Dementias

In our subset of atypical dementia cases (n = 24), our models provided poor predictive performance for both NA-CSVD (ACC = 75.0%, 95% CI 53.3%–90.2%, AUC = 0.50, SEN = 0%, SPE = 100%) and CAA (ACC = 45.8%, 95% CI 25.6%–67.2%, AUC = 0.43, SEN = 9%, SPE = 77%). ROC curves and confusion matrices are presented in eFigure 11.

Discussion

In this study, we developed models based on lifelong vascular risk factors and health variables to predict degrees of postmortem CSVD (i.e., NA-CSVD and CAA) in deceased middle-aged to older adults with or without dementia. The strongest clinical determinants included a history of cerebrovascular events and impaired cognition for both subtypes of CSVD. Additional significant predictors included age for NA-CSVD and alcohol use for CAA. The predictive accuracy was high for both models based on clinical factors (AUC of 0.84 and 0.86 for NA-CSVD and CAA, respectively) in our sample covering the spectrum of cognitively healthy aging to typical dementias. However, our models performed poorly in a smaller and heterogeneous sample of cases presenting with atypical dementia, suggesting that this population was less closely associated at the cerebrovascular level.

Clinical Predictors of Nonamyloid CSVD

Increasing age at death has been consistently associated with higher prevalence and severity of arteriolosclerosis across multiple brain regions,^7,8 and our results did not differ from those published by previous studies. In our model, a history of major cerebrovascular disease events (i.e., ischemic and hemorrhagic strokes and transient ischemic attacks) was also predictive of moderate-to-severe arteriolosclerosis. Other authors reported similar clinicopathological associations between stroke and postmortem NA-CSVD lesions such as arteriolosclerosis, lacunes, microscopic atherosclerosis, and white matter pallor.⁹ Conversely, it appears that in vivo evidence of NA-CSVD is also associated with an increased risk of both ischemic and hemorrhagic strokes,¹⁰ but these findings rely on neuroimaging markers of presumed vascular origin such as WMH on MRI instead of neuropathology. When addressing the inverse relationship, we found a significant association between clinical dementia and postmortem NA-CSVD. Other neuropathologic studies yielded similar findings with cognitive impairment (either mild cognitive impairment or dementia) being associated with more severe brain arteriolosclerosis.¹⁰

Contrary to our expectations, hypertension and diabetes were not predictive of brain arteriolosclerosis severity in our sample of middle-aged to older adults. This association between hypertension during life and histologic microinfarcts, but not arteriolosclerosis, was also reported elsewhere.⁹ This is an interesting finding since arteriolosclerosis has been traditionally considered an hypertensive arteriopathy.¹¹ However, evidence of hypertensive etiology mainly comes from other organs (e.g., the kidney), in vivo and animal studies,^7,12 or indirect evidence of cerebral microvasculature changes, rather than direct quantification of small vessel damage.¹³ Some authors argue that this association between hypertension and arteriolosclerosis may vary by age⁷ or be due to other factors such as capillary dysfunction,¹² blood pressure variability,^14,15 choice and intensity of antihypertensive treatment,^16,17 and/or the presence of genetic susceptibility genes.¹⁸ There is also some evidence that higher diastolic (but not systolic) blood pressure is linked to increased arteriolosclerosis,¹⁹ which would be consistent with the observation that more than half of cerebral perfusion occurs during diastole.²⁰ Moreover, it remains uncertain whether systemic hypertension is a risk factor for brain arteriolosclerosis or if brain arteriolosclerosis contributes to the development of hypertension via compensatory mechanisms.¹² The lack of association with diabetes in our study, another presumed risk factor for NA-CSVD, was also surprising, but had been reported elsewhere.⁹ It is possible that diabetes is only associated with arteriolosclerosis in restricted brain areas¹⁹ and that this regional pattern was not captured by our study design. Other contributing factors for the lack of association between NA-CSVD and vascular risk factors include bias inherent to being a retrospective study and uncertain accuracy of collected data from medical files.

Other traditional vascular risk factors were not associated with the severity of NA-CSVD, in line with previous studies that reported no association of NA-CSVD with cholesterol and triglycerides, high body mass index, or smoking.^19,21 Heart attack as a manifestation of coronary artery disease was demonstrated in only one study to be associated with microinfarcts, but not other histopathologic features of NA-CSVD.⁹ Moderate-to-severe obstructive sleep apnea was also identified as a clinical determinant of MRI-defined NA-CSVD,²¹ but this was not confirmed in our sample because of the lack of reporting. While midlife vascular risk factors appear to be more strongly associated with vascular pathology than late-life risk factors,^22,23 this distinction could not be considered here because of our retrospective design.

Finally, it should be noted that most studies investigating clinical contributors to NA-CSVD are looking at surrogate markers of parenchymal damage as WMH instead of postmortem vessel wall pathology.^18,23-27 Whether WMH truly mirrors the severity of arteriolosclerosis is unknown. In fact, WMH have been associated with other heterogeneous histologic changes such as gliosis, demyelination, enlarged PVS, and venous collagenosis in radiopathological correlation studies.²⁸

Clinical Predictors of Amyloid CSVD (or CAA)

In those reporting occasional alcohol consumption, we detected an increased risk of CAA pathology. This was not the case in individuals reporting moderate-to-severe alcohol intake. However, we had no available quantitative data on alcohol use, whether drinking habits were underreported, or whether former alcoholics could be distinguished from lifetime abstainers. Very few studies to date have specifically looked at this association. We found some published evidence of greater severity of CAA in demented individuals with documented current alcohol intake at the time of death.²⁹ Daily alcohol consumption was also associated with higher risk of lobar hemorrhage,³⁰ but this is probably because of increased bleeding risk and not increased CAA pathology. Conversely, other investigators found a potential benefit of moderate lifetime alcohol intake on AD by reducing parenchymal amyloid deposition as assessed by PET scan, but vessel wall deposition was not addressed and may behave differently.³¹

We also found an association between cerebrovascular disease events of all types and CAA pathology. The relationship between intracerebral hemorrhage and CAA is certainly well known,^29,32,33 but the literature is scarce regarding the role of CAA in ischemic strokes and the high incidence of ischemic events in the population with CAA may be partly explained by age-related comorbidities but also by endothelial dysfunction and impaired vasoreactivity.³⁴ Regarding our finding of increased risk of postmortem CAA in patients with clinical dementia, this is consistent with the current knowledge of cognitive impairment related to the cumulative hemorrhagic burden (microbleeds, cortical siderosis, and intraparenchymal hemorrhages) and secondary cortical thinning in patients with CAA^32,33 and the common co-occurrence in the AD population.³³

We reported no association between traditional vascular risk factors and the presence of CAA at autopsy. Two previous autopsy studies also found no association between common vascular risk factors present at midlife or late-life and CAA pathology,^22,35 and a third study reported similar results based on MRI-defined CAA.³⁶ Conflicting results were reported in one isolated study showing even lower risk of moderate-to-severe CAA in patients with hypertension and increased risk of moderate-to-severe CAA in those with diabetes and coronary artery disease, and this latter finding was only present in nondemented individuals.⁹

Of note, it is not surprising that we found shared predictors between NA-CSVD and CAA. While these results may be explained by individuals with mixed CSVD being the largest group within our study cohort, it appears that co-occurrence of multiple pathologies is also the most prevalent finding at autopsy of older individuals, with pure pathologies being excessively rare.³⁷ Therefore, it is possible, and even probable, that a single risk factor may contribute to several types of cerebral lesions.

CSVD Pathology in Atypical Dementia

Post hoc analyses of sporadic CSVD in patients with atypical dementias revealed that this heterogeneous subgroup of neurodegenerative diseases does not fit into the expected predictions from the VCI/Mixed/AD continuum based on clinical risk factors. Actually, evidence is scarce regarding CVD contribution to atypical dementias and the prevalence of copathologies (much less well described than what is known in AD).^38,39 LBD pathology seems to be inversely correlated to the degree of most vascular pathologies, including atherosclerosis and NA-CSVD. In contrast, CAA is positively correlated with deposition of brainstem and cortical Lewy bodies.³⁸ As observed in patients with AD, patients with α-synucleinopathies and concomitant CVD show relatively lower burdens of protein deposits than those without CVD for comparable cognitive impairment at the time of death,⁴⁰ suggesting that vascular pathology may lower the threshold for cognitive decline. In patients with FTLD, it has been demonstrated that severe CSVD and white matter demyelination was seen in Pick disease, indicating a possible role of vascular copathology in this specific subtype of FTLD-tau.⁴¹ Importantly, after controlling for age and sex, postmortem brain examination of LBD and FTLD due to tau and pTDP-43 cases revealed a lower prevalence of coincident cerebrovascular disease than in patients with AD, and this was even more significant in younger patients.⁴⁰ Altogether, these findings support a lower contribution of vasculopathies and secondary tissue damage in the physiopathologic processes underlying atypical dementias than in highly prevalent age-associated dementias.

Limitations

Several study limitations may arise because of our retrospective cross-sectional study design. Data collection was performed by careful examination of medical files, which led to missing information, for example, education and ethnicity were not routinely recorded.

Vascular risk factors and major cerebrovascular pathologies may differ across socioeconomic and racial groups, and it is, therefore, recommended that these medical conditions be analyzed in diverse populations to improve our ability to generalize findings and interventions for global vascular brain health.²³ This information was missing here, but we can reasonably postulate that our population was composed of a majority of White Europeans given the local ethnic composition of Edinburgh.

Health conditions were also extracted from documented clinical history. Some conditions such as lifestyle habits may be underreported, whereas others such as hypertension, dyslipidemia, and diabetes may be underdiagnosed, leading to bias and lowering of associations toward the null, which could possibly explain why we did not find any significant association between hypertension and NA-CSVD, for example. Moreover, detailed information on age at diagnosis and disease duration for vascular and other medical conditions was not available. This lack of temporal specifications could reduce the probability of detecting an effect in this type of cohort because mid-life rather than late-life vascular risk burden appear substantially more strongly associated with late-life CSVD and cognition.^22,23,36

APOE genotype was assessed in only 40% of our sample and, therefore, could not be included in our predictive models. APOE4 carrier status is a known risk factor for CVD pathology, especially CAA, and is also associated with an increased risk of AD pathology and dementia.³² Finally, no formal cognitive testing was performed in our sample selected from autopsied cases, and some individuals with mild cognitive impairment could have been wrongly classified in the cognitively healthy group.

In conclusion, in this retrospective series of middle-aged to older adults who underwent brain autopsy, a history of ischemic or hemorrhagic stroke and dementia were associated with higher odds of pathologically-confirmed moderate to severe NA-CSVD and the presence of CAA. Other significant predictors included chronological age for NA-CSVD and alcohol intake for CAA. We demonstrated that postmortem CSVD pathology can be predicted with high accuracy based on clinical factors in patients within the spectrum of AD, VCI, and normal aging. Whether this prediction can be enhanced by the addition of fluid and neuroimaging biomarkers warrant further study. Improving our understanding of clinical determinants of vascular brain health may lead to novel strategies in the prevention and treatment of vascular etiologies contributing to cognitive decline.

Appendix. Authors

Name	Location	Contribution
Caroline Dallaire-Théroux, MD, PhD	CERVO Brain Research Center; Faculty of Medicine, Université Laval; Department of Neurological Sciences, Centre Hospitalier Universitaire de Québec, Canada	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Colin Smith, MD	Academic Neuropathology, Centre for Clinical Brain Sciences, University of Edinburgh, United Kingdom	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design
Simon Duchesne, PhD	CERVO Brain Research Center; Department of Radiology and Nuclear Medicine, Faculty of Medicine, Université Laval, Quebec City, Canada	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design

Study Funding

This work was funded by the Canadian Institutes of Health Research (operating grant #159778; primary investigator S. Duchesne).

Disclosure

C. Dallaire-Théroux held a Frederick Banting and Charles Best Canada Graduate Scholarship Doctoral Award from the Canadian Institutes of Health Research (CIHR; #406235) at the time of study design, data collection and analysis. Her research journey in Edinburgh for data collection was sponsored by the Michael-Smith Foreign Study Supplements from the CIHR (#172804). C. Smith is the Director of The Edinburgh Brain Bank, which was funded by the UK Medical Research Council (MR/L016400/1) at the time of the study. All other authors report no disclosures relevant to the manuscript. Full disclosure form information provided by the authors is available with the full text of this article at Neurology.org/cp.