Retinal perfusion is linked to cognition and brain MRI biomarkers in Black Americans

Abstract

INTRODUCTION

We investigated whether retinal capillary perfusion is a biomarker of cerebral small vessel disease and impaired cognition among Black Americans, an understudied group at higher risk for dementia.

METHODS

We enrolled 96 Black Americans without known cognitive impairment. Four retinal perfusion measures were derived using optical coherence tomography angiography. Neurocognitive assessment and brain magnetic resonance imaging (MRI) were performed. Multiple linear regression analyses were performed.

RESULTS

Lower retinal capillary perfusion was correlated with worse Oral Symbol Digit Test (P < = 0.005) and Fluid Cognition Composite scores (P < = 0.02), but not with the Crystallized Cognition Composite score (P > = 0.41). Lower retinal perfusion was also correlated with higher free water and peak width of skeletonized mean diffusivity, and lower fractional anisotropy (all P < 0.05) on MRI (N = 35).

DISCUSSION

Lower retinal capillary perfusion is associated with worse information processing, fluid cognition, and MRI biomarkers of cerebral small vessel disease, but is not related to crystallized cognition.

1. BACKGROUND

Vascular contributions to cognitive impairment and dementia (VCID) is one of the most common causes of dementia, second only to Alzheimer's disease (AD). In addition to cerebral large vessel pathologies, small vessel pathologies are also common in dementia. Consensus statements from several organizations have highlighted the importance of developing biomarkers of small vessel VCID for diagnosis and therapeutic interventions. ¹ , ² , ³

Imaging measures of retinal perfusion have been explored as potential biomarkers of VCID. ⁴ Retinal and cerebral microvasculature share similar embryological, anatomical, physiological, and pathological characteristics. This similarity creates an opportunity to study cerebral microvasculature through retinal imaging. ⁴ , ⁵ , ⁶ Retinal microvascular pathologies detectable by retinal photography have been associated with cognitive impairment, ⁷ , ⁸ and cerebrovascular diseases such as stroke. ⁹ Optical coherence tomography angiography (OCTA) is an imaging modality recently approved by the Food and Drug Administration (FDA). Unlike fluorescein angiography, which requires intravenous dye injection, OCTA allows for noninvasive high‐resolution in vivo imaging of the retinal capillaries, objective quantification of retinal perfusion changes, and early detection of vascular changes before clinical manifestations, and therefore has the potential as a useful screening tool for VCID. Despite its limitations such as artifacts (eg, the projection artifact seen in deep retinal layers), OCTA has been increasingly incorporated into clinical practice. It has been used to detect early signs of microvascular abnormalities in the retina in systemic vascular diseases such as diabetes mellitus and hypertension. ¹⁰ , ¹¹ OCTA studies have found increased foveal avascular zone area and decreased vessel density in preclinical AD. ¹² , ¹³ , ¹⁴ , ¹⁵ Similarly, in a study of elderly Caucasians, retinal perfusion metrics were found to be lower among cognitively normal carriers of the apolipoprotein E (APOE) ε4 allele, ¹⁶ a well‐known genetic risk factor for AD and cardiovascular disease. However, only a few studies have examined the application of OCTA in VCID. Our team found that lower retinal perfusion was associated with abnormalities in cerebral perfusion and cognitive impairment among Latinx subjects at high risk for cerebral small vessel disease. ¹⁷ Based on these findings, it is possible that OCTA may broadly capture the microvascular manifestations of different vascular risk factors (eg, diabetes, hypertension, and APOE ε4), and therefore serve as a biomarker of VCID.

RESEARCH IN CONTEXT

1. Systematic review: The authors searched available publications using traditional (eg, PubMed) sources. The association of retinal capillary blood perfusion with cognitive function and MRI biomarkers of cerebral small vessel disease is not yet well studied. Relevant studies are cited in the paper.

2. Interpretation: Our results indicate that lower retinal capillary perfusion is associated with worse information processing speed and fluid cognition scores, and MRI biomarkers of small vessel vascular contributions to cognitive impairment and dementia (VCID) in community‐dwelling older African Americans. These findings suggest that optical coherence tomography angiography‐derived capillary perfusion measures can be used as biomarkers for screening small vessel VCID, particularly among underdiagnosed, underserved, high‐risk populations such as Black Americans, and as a biomarker for monitoring disease progression and responses to treatment.

3. Future directions: Further studies with larger, more diverse samples, prospective follow‐up of participants, and more comprehensive cognitive assessment are needed to validate our findings.

To further study the hypothesis that a change in retinal capillary perfusion is an early biomarker of cerebral small vessel disease and impaired cognition, we investigated the association of retinal capillary perfusion with cognitive measures mostly affected in small vessel VCID among older, community‐dwelling adults without known cognitive impairment. We also explored the association of retinal capillary perfusion with magnetic resonance imaging (MRI)‐based measures of white matter integrity and cerebral perfusion, such as free water (FW) and peak width of skeletonized mean diffusivity (PSMD), which have been shown as sensitive biomarkers of small vessel VCID. All of our participants are Black/African Americans (hereafter, Black Americans), an understudied minority group at high risk of diabetes, hypertension, cardiovascular disease, and potentially VCID. Not only is dementia approximately 1.5 to 2 times more prevalent in Black Americans than White Americans, ¹⁸ but Black Americans are also more likely to be underdiagnosed with cognitive impairment, less likely to receive adequate care, and are underrepresented in dementia studies. ¹⁹

2. METHODS

2.1. Subjects

Participants of this study were recruited through convenience sampling from subjects who previously enrolled in the African American Eye Disease Study (AFEDS) for retinal imaging. AFEDS is a population‐based study of over 6000 Black American residents of Inglewood, California. ²⁰ Community‐dwelling Black Americans over the age of 40 without any history of cognitive impairment were recruited. Participants were enrolled at the Roski Eye Institute, University of Southern California (USC). Self‐reported history of physician‐diagnosed ophthalmic, neurological, and cardiovascular diseases and treatments, and other general health conditions were obtained. Cognitive assessment and MRI imaging were later added to the study as supplemental investigations to expand the scope of the study and generate new, promising leads for dementia, especially VCID. Exclusion criteria included inability to consent, any history of cognitive impairment or any cognitive complaint by the participants or their family members, history of neurological disorders such as stroke, history of psychiatric disease, history of ophthalmic disorders affecting the acquisition of retinal images including low vision, severe medial opacities, high‐grade refractive errors, and prior intraocular surgery, and contraindications to MRI imaging. The study conformed to the principles of the Declaration of Helsinki. It was approved by the USC Institutional Review Board. Written informed consent was obtained from all participants before recruitment.

2.2. OCTA imaging

Fovea‐centered 3 × 3 mm² OCTA images were captured from both eyes of each participant using a commercially available swept source OCTA device (PLEX Elite 9000; Carl Zeiss Meditec, Dublin, CA) at the USC Roski Eye Institute. The device has a 100‐kHz A‐scan rate, ∼5 μm axial, and ∼16 μm transverse resolution in tissue, and uses an optical microangiography algorithm incorporating the variations of intensity and phase information between repeat B‐scans for detection of blood flow. ²¹ The OCTA imaging methods have been described previously and were similar to the published protocols from the MarkVCID study. ²² , ²³ Briefly, images with a manufacturer‐reported signal strength of 8 or higher were included. Images with artifacts including motion, shadow, tilt, decentration, or segmentation error were excluded. If multiple images were available for one eye, the best quality image with minimal artifacts was selected. The OCTA assessment was blinded to the cognitive and cerebrovascular status of the subject. The superficial retinal layer was segmented using the commercially available manufacturer software. The OCTA images were converted into binarized images using a combination of a global threshold, hessian filter, and adaptive thresholding ²⁴ ; the binarized images were then skeletonized. A previously validated semiautomated software written in MATLAB (R2018b; MathWorks, Inc., Natick, MA, USA) was used to quantify measures of retinal perfusion. ¹⁶ , ¹⁷ , ²² , ²⁴ , ²⁵ , ²⁶ Vessel area density (VAD) measures the total area of the retinal image that is occupied by blood flow signal. Vessel skeleton density (VSD) measures the overall perfused vessel length within an OCTA image. Vessel area flux (VAF) and vessel skeleton flux (VSF) are novel measures that approximate the number of red blood cells moving through retinal vessel segments per unit time, and are quantified based on non‐binarized images. ²² , ²⁵ Measures were averaged if quality images were available for both eyes.

2.3. Cognitive assessment

Neurocognitive assessments were conducted using the standardized cognitive function battery tests of the NIH Toolbox for the Assessment of Neurological and Behavioral Function (NIH‐TB), ²⁷ , ²⁸ using the NIH Toolbox iPad App (Glinberg & Associates, Inc) by trained administrators. Summary scores of Fluid Cognition Composite and Crystallized Cognition Composite were derived from subtests. The Fluid Cognition Composite score measures executive function, information processing speed, working memory, and episodic memory. The Crystallized Cognition Composite score assesses verbal knowledge and language skills and is influenced by education and past learning experiences. The uncorrected standard scores that compare the participants to the NIH‐TB nationally representative normative data were used. A score at or near 100 suggest average cognitive ability, a score around 85 suggests below‐average cognitive ability, and a score of 70 or lower suggests significant cognitive impairment. ²⁹ In addition, the NIH‐TB Oral Symbol Digit Test, a measure of simple information processing speed, was administered. ³⁰ These cognitive function tests have been shown to be valid and reliable. ²⁷ , ²⁸

2.4. MRI imaging

MRI was incorporated into our investigation after the start of the study; therefore, fewer participants were recruited for MRI scanning (Figure 1). MRI images were acquired on a Siemens 3T Prisma MRI scanner with a 32‐channel head coil at the Center for Image Acquisition of USC Stevens Neuroimaging and Informatics Institute. MRI scans included 3D T1‐weighted magnetization‐prepared rapid gradient‐echo imaging, 3D T2‐weighted fast spin echo fluid‐attenuated‐inversion‐recovery; 2D diffusion‐weighted spin‐echo echo‐planar imaging/diffusion tensor imaging (DTI); and 3D gradient and spin‐echo pseudo‐continuous arterial spin labeling (pCASL). MarkVCID consortium imaging‐based biomarker toolkits were used to calculate mean FW, PSMD, and ARTS (ARTerioloSclerosis) score. ²³ , ³¹ A custom MATLAB program was used to calculate the global mean cerebral blood flow (CBF) from the pCASL data. ³²

FIGURE 1.

Flowchart of participant recruitment and data acquisition.

Specifically, FW represents the fraction of water molecules that are unrestricted by cerebral tissue structures. ³³ FW has been strongly associated with cognitive decline (especially processing speed and executive function) in patients with cerebral small vessel disease. ³⁴ , ³⁵ PSMD is a new DTI‐derived measure calculated as the difference between the 95th and 5th percentile of MD values in the skeletonized MD maps. PSMD measures the distribution of diffusivity and was shown to be strongly associated with lower processing speed in small vessel VCID. ³⁶ The ARTS score is a classifier initially trained on ex‐vivo brain MRI‐derived white matter hyperintensities (WMH) volume and four regional fractional anisotropy (FA) values, in addition to age and gender, against autopsy histopathologic arteriolosclerosis findings. The in‐vivo classifier was then derived based on the relationship between the ex‐vivo and in‐vivo MRI values. The ARTS score was shown to predict the presence of moderate or severe arteriolosclerosis in older adults free of dementia, and a higher ARTS score was associated with more decline in cognitive functions (especially processing speed) affected in small vessel VCID. ³⁷

2.5. Statistics

Statistical analyses were performed in R version 4.1.1 (R Core Team, 2021; R Foundation for Statistical Computing, Vienna, Austria) using packages Stats version 4.1.1, and Emmeans version 1.6.3; and Python version 3.9.9 using packages SciPy version 1.7.0, and Statsmodels version 0.13.2. The correlation between the OCTA‐derived measures of vascular perfusion and cognitive function tests was evaluated using both univariate and multivariable linear regression analysis adjusted for age, gender, and level of education, which are known to be associated with both cognition and retinal vascular measures. To demonstrate the magnitude of the OCTA‐cognition association, we reported the differences in means and estimated marginal means (least square means) of cognitive scores with the OCTA measures categorized into tertiles. In the subset of the subjects with MRI data, we also explored the correlation between the OCTA measures and MRI measures using univariate and multivariable linear regression analyses. All P‐values were two‐sided. The level of statistical significance was set at 0.05. To minimize type II errors in this exploratory investigation, we did not perform adjustment for multiple testing following common suggestions. ³⁸ , ³⁹

3. RESULTS

A total of 101 participants were initially recruited. Five subjects were excluded due to past history of stroke. Please see Figure 1 for a flowchart of study participants and data. Table 1 shows the characteristics of the 96 study participants included in our data analyses. OCTA images of sufficient quality were available for 82 participants, five of whom had images available for one eye only (included in the analyses). Cognitive assessment scores were available for 85 subjects. The mean Oral Symbol Digit Test score was 62.8 ± 17.8. The age‐specific distribution of the Oral Symbol Digit Test scores was similar to the corresponding NIH normative data. ³⁰ MRI data were available for 42 subjects. Seventy‐four participants had both OCTA and cognitive measures, and 35 participants had both OCTA and MRI measures available for analysis (Figure 1).

TABLE 1.

Characteristics of study participants.

	All study participants with any OCTA, cognitive measures, or MRI	Participants with both OCTA and cognitive measures
Variable	Mean ± SD or N (%)
Number of subjects	96	74
Age	66.4 ± 9.6 years	66.4 ± 8.4
Male gender, No. (%)	31 (32.3%)	25 (33.8%)
Level of education, No. (%)
High school graduate	20 (20.8%)	16 (21.6%)
Some college education	31 (32.3%)	24 (32.4%)
College graduate	28 (29.2%)	22 (29.7%)
Graduate degree	16 (16.7%)	12 (16.2%)
Diabetes mellitus, No. (%)	22 (22.9%)	20 (27.0%)
Hypertension, No. (%)	66 (68.8%)	53 (71.6%)
Systolic blood pressure, mmHg	144.6 ± 24.4	148.5 ± 24.9
Diastolic blood pressure, mmHg	83.8 ± 12.0	85.5 ± 12.2
Oral Symbol Digit Test score	62.8 ± 17.8	62.6 ±  17.8
NIH Toolbox Fluid Composite Score, uncorrected	81.4 ± 12.1	80.6 ±  11.8
NIH Toolbox Crystallized Cognition Composite Score, uncorrected	101.8 ± 9.1	102.0 ± 8.8
VAD	0.440 ± 0.015	0.440 ± 0.015
VSD	0.148 ± 0.008	0.148 ± 0.008
VAF	0.156 ± 0.015	0.156 ± 0.015
VSF	0.055 ± 0.006	0.055 ± 0.006
FW	0.21 ± 0.03	0.21 ± 0.03
PSMD	2.46 ± 0.39	2.53 ± 0.40 *
FA	0.51 ± 0.02	0.51 ± 0.02
WMH	4.96 ± 6.68	5.50 ± 7.26
ARTS	−0.83 ± 0.63	−0.75 ± 0.64
CBF	41.38 ± 12.84	39.59 ± 11.12

Abbreviations: ARTS, ARTerioloSclerosis; CBF, cerebral blood flow; FA, fractional anisotropy; FW, free water; MRI, magnetic resonance imaging; OCTA, optical coherence tomography angiography; PSMD, peak width of skeletonized mean diffusivity; VAD, vessel area density; VAF, vessel area flux; VSD, vessel skeleton density; VSF, vessel skeleton flux; WMH, white matter hyperintensities.

^* P = 0.037 for difference between participants with both OCTA and cognitive measures and those without both measures. All other variables were not significantly different between those two groups of participants.

We evaluated the association of four OCTA‐based retinal vascular measures with cognition measures in both univariate and multivariable models (Figures 2 and S1). Even after adjusting for age, gender, and level of education, lower OCTA perfusion measures, especially VAD and VAF (Figure 2), were associated with worse Oral Symbol Digit Test scores (P < 0.001, Std. beta = 0.42, 95% confidence interval [CI] [0.19, 0.65]; and P = 0.005, Std. beta = 0.36, 95% CI [0.11, 0.60], respectively) and Fluid Cognition Composite scores (P = 0.022, Std. beta = 0.28, 95% CI [0.04, 0.52]; and P = 0.005, Std. beta = 0.33, 95% CI [0.10, 0.56], respectively). No statistically significant correlation was found between the OCTA (VAD and VAF) measures and the Crystallized Cognition Composite scores (P = 0.41, Std. Beta = 0.10, 95% CI [−0.15, 0.36]; and P = 0.65, Std. Beta = 0.06, 95% CI [−0.19, 0.31], respectively). Our results stayed consistent even after further adjustment for diabetes and hypertension status in our multivariable models. Three of the participants had diabetic retinopathy in at least one eye. Similarly, our results stayed consistent after removing these subjects from the analyses (data not shown).

FIGURE 2.

The relationship between representative optical coherence tomography angiography‐derived measures of retinal perfusion, VAD (vessel area density) and VAF (vessel area flux), and NIH Toolbox cognitive function battery test scores: Oral Symbol Digit Test, Fluid Cognition Composite, and Crystallized Cognition Composite scores. The univariate linear regression lines and 95% confidence intervals are shown. The P‐values for the univariate linear regression analysis and the multivariable regression analysis, adjusting for age, gender, and education, are reported. *P < 0.05, **P < 0.01, ***P < 0.001.

To quantify the magnitude of the association between OCTA measures and cognitive test scores, we estimated the unadjusted and adjusted means of cognitive function test scores between the different tertiles of the OCTA measures. The crude Oral Symbol Digit Test score was 21.4 (± 6.9) lower (P = 0.009) in subjects in the lowest VAD tertile than in subjects in the highest VAD tertile, and 25.7 (± 7.8) lower (P = 0.005) in subjects in the lowest VAF tertile than in those in the highest VAF tertile. Similarly, the crude Fluid Cognition Composite score was 15.7 (± 3.8) lower (P < 0.001) in subjects in the lowest VAD tertile than in subjects in the highest tertile, and 19.5 (± 4.7) lower (P < 0.001) in subjects in the lowest VAF tertile than in those in the highest tertile. No statistically significant difference was seen in the means of Crystallized Cognition Composite scores between the OCTA tertiles. We also estimated marginal means (least square means) of cognitive test scores for individuals in different tertiles of OCTA measures (Figures 3 and S2), after controlling for differences in age, gender, and level of education. The Oral Symbol Digit Test score was 20.8 (± 7.1) lower (P = 0.013) in subjects in the lowest VAD tertile than in subjects in the highest VAD tertile, and 25.1 (± 7.9) lower (P = 0.007) in subjects in the lowest VAF tertile than in those in the highest VAF tertile. Similarly, the Fluid Cognition Composite score was 8.0 (± 2.9) lower (P = 0.02) in subjects in the lowest VAD tertile than in subjects in the middle tertile, and 7.2 (± 3.0) lower (P = 0.05) in subjects in the lowest VAF tertile than in those in the middle tertile. No statistically significant difference was seen in the estimated marginal means (least square means) of Crystallized Cognition Composite scores between the OCTA tertiles (Figures 3 and S2).

FIGURE 3.

Estimated marginal means (least square means) of cognitive function test scores derived from the multivariable regression analysis (controlling for age, gender, and level of education), with the optical coherence tomography angiography perfusion measures, VAD (vessel area density) and VAF (vessel area flux), categorized into tertiles. The difference between the lowest and highest group means is shown. *P < 0.05, **P < 0.01.

We also explored the multivariable association between OCTA‐derived perfusion measures and MRI‐based biomarkers of cerebral small vessel disease (Figures 4 and S3). Despite a relatively smaller sample size (N = 35) available for this analysis, we found that lower OCTA measures, especially VAD, VAF, and VSF, were associated with higher levels of FW (P = 0.003, Std. Beta = −0.45, 95% CI [−0.73, −0.17]; P = 0.003, Std. beta = −0.46, 95% CI [−0.75, −0.17]; and P = 0.002, Std. beta = −0.48, 95% CI [−0.77, −0.19], respectively) and PSMD (P = 0.032, Std. beta = −0.32, 95% CI [−0.61, −0.03]; P = 0.007, Std. beta = −0.41, 95% CI [−0.69, −0.12]; and P = 0.005, Std. beta = −0.42, 95% CI [−0.70, −0.14], respectively), and lower levels of FA (P = 0.070, Std. beta = 0.30, 95% CI [−0.03, 0.63]; P = 0.015, Std. beta = 0.41, 95% CI [0.09, 0.73]; and P = 0.017, Std. beta = 0.41, 95% CI [0.08, 0.74], respectively). However, the association of the OCTA measures with the ARTS score, CBF, and WMH, although trending in the expected direction, did not reach statistical significance (Figures S4 and S5). Representative OCTA images and their corresponding MRI images are shown in Figure 5.

FIGURE 4.

The relationship between optical coherence tomography angiography‐derived measures of retinal perfusion, VAD (vessel area density) and VAF (vessel area flux), and magnetic resonance imaging measures: FW (mean free water), PSMD (peak width of skeletonized mean diffusivity), and mean skeletonized FA (fractional anisotropy). The univariate linear regression lines and 95% confidence intervals are shown. The P‐values for the univariate linear regression analysis and the multivariable regression analysis, adjusting for age, gender, and education, are reported. *P < 0.05, **P < 0.01, ***P < 0.001.

FIGURE 5.

Optical coherence tomography angiography (OCTA) images representative of high (top row), and low (bottom row) OCTA perfusion measures, and their corresponding FW (free water), FA (fractional anisotropy), and FLAIR (fluid attenuated inversion recovery) images from brain magnetic resonance imaging scans.

4. DISCUSSION

This study evaluated the relationship between OCTA‐derived measures of retinal capillary perfusion with cognitive function and MRI measures known to be affected in small vessel‐related VCID. This investigation focused on community‐dwelling Black Americans, a population at higher risk of developing VCID. We found that (1) lower retinal capillary perfusion measures are correlated with worse information processing speed and fluid cognition, and (2) lower retinal perfusion measures are also correlated with worse MRI measures of FW (mean FW), PSMD, and FA. These associations suggest that retinal perfusion measures reflect both functional and structural changes associated with VCID in the brain, and therefore might be a useful screening and monitoring tool for cerebral small vessel disease.

Our results indicate that altered retinal perfusion may be a biomarker of early changes in cognition resulting from cerebral small vessel disease. Cerebral small vessel disease is associated with a decline in information processing speed and executive function. ⁴⁰ , ⁴¹ , ⁴² Changes in these cognitive functions can be measured through the NIH‐TB Oral Symbol Digit Test and Fluid Cognition Composite score. ²⁷ Processing speed is one of the earliest measures that aging affects, with a decline starting in the thirties. ⁴³ The symbol digit test, a measure of simple processing speed, has been reported to be sensitive in detecting patients with small vessel VCID, even when cognitive deficits are subtle. ⁴⁴ We found that the Oral Symbol Digit Test score was at least 20 points lower among individuals in the lowest VAD/VAF tertile than those in highest tertile. This difference is greater than the age‐related decline expected from age 40 to 49 (mean ± standard deviation: 81.38 ± 29.36) and 60 to 69 (63.57 ± 26.42). ³⁰ Our observation of impaired simple processing and fluid cognition among individuals with lower levels of retinal perfusion suggests that these OCTA‐based measures of change in retinal perfusions may be used for early detection of cerebral small vessel disease. On the other hand, crystallized cognition, as measured by the NIH‐TB Crystallized Cognition Composite score based on vocabulary and language tests, is not usually affected in cerebral small vessel disease. Consistently, crystallized abilities, as expected, were not associated with retinal perfusion measures in our study. This absence of correlation with cognitive functions not affected in small vessel VCID suggests that these OCTA‐based retinal capillary measures might be biomarkers specific for small vessel VCID.

Consistent with the above‐mentioned associations with cognition, we found that abnormal retinal perfusion was associated with early structural changes in the brain associated with cerebral small vessel disease. Individuals with lower retinal perfusion were found to have higher FW, PSMD, and lower FA, all of which have been implicated in cerebral small vessel disease. FW, a measure of unrestricted water diffusion by tissue structures, has been shown to be a sensitive biomarker of small vessel brain injury even in relatively healthy, younger adults. ⁴⁵ A higher baseline FW has also been associated with lower cognitive scores and accelerated cognitive decline, including executive function. ³⁴ , ⁴⁶ PSMD, a diffusion tensor‐based measure that represents the distribution of diffusivity along white matter tracts, has been robustly associated with established biomarkers of cerebral small vessel disease, ⁴⁷ but not with neurodegenerative disease. ³⁶ Higher PSMD has been associated with lower processing speed ³⁶ and executive function. ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ On the other hand, in our study, OCTA‐derived retinal perfusion measures were not significantly associated with other MRI measures, including CBF, WMH, and ARTS score, although the trends are in the expected direction. The ARTS score is a classifier trained to predict the presence of moderate or severe arteriolosclerosis based on MRI measures of WMH and diffusion anisotropy. Our study may be limited in statistical power to detect these associations due to our small sample size of MRI imaging (N = 42) and relatively low severity of WMH in this sample of cognitively asymptomatic individuals.

In addition to conventional measures of retinal capillary density, we also investigated OCTA‐derived measures of flux (VAF and VSF), which are more sensitive to subclinical changes in capillary blood flow and have been demonstrated to detect blood flow changes that precede frank pathological changes in capillary structure. ²² , ²⁵ Our results show a strong correlation between these flux measures and both cognitive and MRI measures. These associations appeared to be more consistent and have greater statistical significance than those of retinal density measures (VAD and VSD). This is in line with our previous findings that OCTA‐based flux measures are more sensitive than density measures in detecting a change in retinal capillary perfusion, ²² and that blood pressure and diabetes status are correlated with flux but not density measures earlier in the disease process. ²⁵ Consistently, it has been shown that in mild cognitive impairment (MCI), changes in retinal blood flow can occur without a significant change in vessel diameter. ⁵² , ⁵³ Our results suggest that other novel OCTA measures of retinal perfusion can be further explored as enhanced biomarkers of small vessel VCID.

Our study has several strengths that differentiate it from limited studies on the subject, including (1) recruiting Black Americans, who are understudied in cognitive impairment and dementia research; (2) investigating flux as a novel measure of retinal blood flow; (3) adopting novel MRI metrics of cerebral small vessel diseases; and (4) focusing on specific cognitive and MRI measures that are more likely to be affected in VCID. Despite these strengths, this study has some limitations. The cross‐sectional design of this study limits the causal inference from our data. We did not however evaluate the causality of cognitive decline in VCID, or assess the clinical risk predictors of VCID, but rather evaluated biomarkers that can potentially capture the microvascular manifestations resulting from these risk factors. Our study had a relatively small number of available MRI images, limiting the statistical power of tests with MRI data. Nevertheless, even with this small sample size, we found a significant correlation between retinal perfusion measures and MRI measures that are affected early in small vessel VCID. Another limitation is that we did not conduct comprehensive cognitive assessments to diagnose MCI or dementia and identify the causes (eg, APOE genotyping), to rule out other etiologies of cognitive impairment such as AD among our participants. However, VCID likely played a major role, based on the following: (1) this study sample had a high prevalence of vascular risk factors (approximately 70% having hypertension and over 20% having diabetes); (2) MRI measures (eg, PSMD) strongly associated with VCID were also affected among participants with poorer retinal perfusion; and (3) the observed association of the measures of retinal small vessels with cognitive performance measures suggests that a change in the small vessels contributes to the cognitive decline observed. Furthermore, our study population is fairly well educated. However, our data show that our study participants' Oral Symbol Digit Test and Crystallized Cognition Composite scores were comparable to those from the NIH‐TB's nationally representative normative samples. Therefore, despite these limitations, our results may be applicable to a similar urban population of older community‐dwelling Black Americans, who, like our participants, are likely to have undiagnosed cognitive impairment/dementia. On the other hand, the generalizability of our results to other populations/races/ethnicities is unknown at this time. Lastly, we did not adjust for multiple testing in our analyses, because this was a small study aimed at generating new, promising hypotheses for further research. Nonetheless, further studies with larger, more diverse samples, prospective follow‐up of participants, comparison with other ocular measures that have been associated with AD and related dementias (eg, retinal nerve fiber layer thickness), and more comprehensive cognitive assessment would help validate our findings.

5. CONCLUSION

Our results indicate that OCTA‐derived retinal capillary perfusion measures might be useful biomarkers of small vessel VCID in community‐dwelling individuals at high risk of developing VCID, with the potential for early detection, monitoring of disease progression, and monitoring the response to treatments. As a noninvasive, rapid, easy‐to‐administer, and FDA‐approved imaging modality, OCTA may be particularly useful for screening underdiagnosed, underserved, high‐risk populations such as Black Americans.

CONFLICT OF INTEREST STATEMENT

Farzan Abdolahi: No commercial relationship. Victoria Yu: No commercial relationship. Rohit Varma: No commercial relationship. Xiao Zhou: No commercial relationship. Ruikang Wang: Commercial relationships: Patent—Kowa Inc, consultant/contractor—Insight Photonic Solutions, Patent—Carl Zeiss Meditec, consultant/contractor—Carl Zeiss Meditec. Lina M. D'Orazio: No commercial relationship. Chenyang Zhao: No commercial relationship. Kay Jann: No commercial relationship. Danny J. Wang: No commercial relationship. Amir H. Kashani: Consultant/contractor—Carl Zeiss Meditec, recipient—Carl Zeiss Meditec. Xuejuan Jiang: No commercial relationship. Author disclosures are available in the Supporting information.

CONSENT STATEMENT

Written informed consent was obtained from all participants before recruitment.

Supporting information

Supporting Information

Supplementary information

Abdolahi F, Yu V, Varma R, et al. Retinal perfusion is linked to cognition and brain MRI biomarkers in Black Americans. Alzheimer's Dement. 2024;20:858–868. 10.1002/alz.13469