Structural and Metabolic Retinal Changes Associated With Mild Cognitive Impairment in Type 2 Diabetes

Type 2 diabetes (T2DM) is associated with mild cognitive impairment (MCI), which is characterized by objective cognitive deficits on standard neuropsychological tests without affecting daily living (1). The World Health Organization predicts a rapid increase in the number of individuals with T2DM and incident cognitive impairment, with an annual conversion rate from MCI to dementia of around 10% (2) and almost twofold higher risk of developing dementia in individuals with T2DM compared with age-matched individuals without diabetes (3). Because diagnostics of MCI rely on extensive neuropsychological testing, there is a need to identify those at increased risk of cognitive impairment to ensure early diagnosis in order to manage modifiable risk factors (e.g., metabolic control) to slow progression of cognitive loss and to provide patients with appropriate support.

Diabetes itself causes structural and vascular changes to the retina even in the absence of diabetic retinopathy (DR) (4). This is considered related to retinal neurodegeneration and microangiopathy, resulting in damage to the neurovascular unit and, ultimately, leading to instability of the retina–blood barrier (5). Because the retina and the brain share similar embryologic origin, retinal imaging might be useful to detect early pathology in the cerebral vasculature and parenchyma, because retinal and cerebral neurovascular degeneration often run in parallel (6). This may provide researchers and clinicians the opportunity to use noninvasive retinal imaging–based markers for identifying individuals with high risk for cognitive impairment and dementia.

Retinal sensitivity and eye fixation assessed by microperimetry have been proposed as noninvasive retinal tools for identifying and monitoring MCI in people with T2DM (7,8). Other retinal structural and metabolic markers have been less investigated within the T2DM population. A single cross-sectional study of 137 participants with T2DM reported venular tortuosity to be higher in persons with cognitive impairment compared with healthy control individuals, but the study used a telephone interview only to assess cognitive status (9). Therefore, the aim of the present study was to determine whether there was a difference in retinal imaging–based neuronal and vascular markers in individuals with T2DM with or without MCI.

This was a prospective, cross-sectional study including individuals with T2DM identified using the Funen Diabetes Database (FDDB), a regional diabetes database in the area of southern Denmark established in 2003 (10). The database provides information on relevant clinical information, including, for example, age, type of diabetes, BMI, blood pressure, DR screening results, and information on comorbidities, such as acute coronary syndrome and cerebrovascular incidents. Potential eligible participants were invited, using a secure digital mailbox (e-Boks; https://www.nets.eu/developer/Nets%20share/ebokschannel/Pages/default.aspx), distributed for each citizen in Denmark. Participants were invited to take part in this study between 25 November 2020 and 25 February 2022, and all examinations were performed at Odense University Hospital. A prescreening based on eligibility criteria (inclusion and exclusion criteria) by telephone was performed to streamline recruitment. Criteria for eligibility were diagnosis of T2DM, age older than 65 years, duration of diabetes of at least 5 years, and ability to provide informed consent. Individuals were excluded if they had a diagnosis of stroke or any neurodegenerative disease; if retinal neurodegeneration was already known to be present due to previous laser photocoagulation, glaucoma, or diabetic macular edema or other eye disorders affecting vision besides possible complications of DR; refractive error ≥6 diopters; media opacities precluding retinal imaging; or severe systemic illness or personal circumstances that would prevent the participant from fulfilling the study protocol. If deemed eligible, individuals were allowed to participate with both eyes under the assumption that retinal changes may not be seen symmetrically.

A medical and ophthalmological history as well as demographics (year of birth, sex, education) were obtained. Current use of antidiabetic agents and blood pressure medication was recorded. Duration of diabetes was obtained from the FDDB and cross-checked with medical records. If a discrepancy was seen between medical records and the FDDB, the earliest year of diagnosis was used to calculate diabetes duration. Marital status was categorized as having a partner or being single. We measured weight and height and calculated BMI as well as abdominal circumference. Smoking status was registered as never, former, or current smoker. Brachial arterial blood pressure was measured with participants sitting with an Omron M6 monitor (HEM-7001-E; Hoofddorp, the Netherlands), and mean arterial pressure was calculated as BP_d + (BP_d + BP_s)/3, where BP_d and BP_s are the diastolic and systolic blood pressures, respectively. In addition, fasting (minimum 4 h) venous blood samples were drawn to collect information on blood glucose levels, lipid profile, and renal function. Participants collected a morning urine sample for measurement of albuminuria; this was defined as normo-albuminuria if the measurement was less than 30 mg/g, microalbuminuria if it was between 30 and 300 mg/g, and macro-albuminuria if it was greater than 300 mg/g. The Geriatric Depression Scale-15 was used to detect presence of depressive symptoms.

All participants underwent a stepwise neuropsychological evaluation. First, the Montreal Cognitive Assessment (MoCA) test was administered by a certified physician. Those with a MoCA score below 26 were considered to have potential cognitive impairment and were subsequently tested with a neuropsychological test battery (NTB). The NTB comprised 13 tests covering five cognitive domains: processing speed (Digit Symbol Substitution Test and Trail-Making Test A); attention and executive functioning (Trail-Making Test B, letter fluency, and backward digit span); memory (Rey Auditory Verbal Learning Test direct, delayed, and recognition; Rey-Österrieth Complex Figure Test delayed and forward digit span); visio-construction (Rey-Österrieth Complex Figure Test copy); and language (Boston Naming Test and Category fluency). Test scores were converted to percentile scores according to reference data (11–16). Individuals were considered to have MCI if the average score of one or more domains was below the 15th percentile or if more than half of the test scores in a single domain were below the 5th percentile (17).

We performed multiple imaging technologies. Our primary outcome was difference in retinal vessel saturation, and secondary outcomes included differences in retinal vascular and structural parameters. Each retinal parameter is described in the following sections.

Best-corrected visual acuity (BCVA) was obtained prior to pupil dilation and the undertaking of any imaging evaluations (discussed below), using Early Treatment Diabetic Retinopathy Study charts at 4 m (Precision Vision). We performed a dilated-eye examination using tropicamide 10 mg/mL and phenylephrine 10%, including 45° seven-field color fundus photography (TRC-50DX fundus camera; Topcon, Tokyo, Japan). DR grading was undertaken by a certified grader according to the International Clinical Diabetic Retinopathy Disease Severity Scale (18), with each eye categorized as level 0 (no DR), 1 through 3 (mild, moderate, and severe nonproliferative DR, respectively), or 4 (proliferative DR).

Central retinal thickness was measured using spectral domain optical coherence tomography (OCT; Spectralis OCT Family Acquisition Module, V6.9a; Heidelberg Engineering). OCT imaging acquisition settings were high speed and volume scan with Automated Real Time 9. The automated segmentation protocol of the Spectralis OCT (Heidelberg Eye Explorer V1.10.2.0) was used to measure the macular ganglion cell layer (mGCL) and macular retinal nerve fiber layer (mRNFL) thicknesses. The automated segmentation was manually corrected for any misalignments. The scans were divided according to the Early Treatment Diabetic Retinopathy Study (ETDRS) map into 1, 3, and 6 mm rings. The inner ring was defined as the central thickness subfield, and the outer rings were divided into four zones designated as the superior, nasal inferior, and temporal zones.

Retinal structures, including vessel width, density, tortuosity, and fractal dimension, were assessed using Vessel Assessment and Measurement Platform for Images of the Retina (VAMPIRE-Web; Universities of Dundee and Edinburgh, UK) (19). In brief, 45°, disc-centered, color fundus photos were used. Three concentric zones were semiautomatically generated around the optic disc, labeled zone A, B, and C at 0.0–0.5, 0.5–1.0, and 0.5–2.0 disc diameters from the disc margin, respectively. The software color labeled venules blue and arterioles red. In cases where the software mislabeled retinal vessels, detected one vessel as two, or labeled a hemorrhage, the annotation was manually corrected. The retinal vessel width was calculated as the central retinal artery and vein equivalent in zone B (20). Vessel density, tortuosity, and fractal dimension were measured in zone C (21).

Macula-centered 4.5 × 0.4.5 mm OCT-angiography (OCT-A) images were obtained with a TRC-50DX fundus camera (Topcon, Tokyo, Japan). MATLAB (MathWorks, Natick, MA) was used for quantification of retinal microvasculature from OCT-A images, as previous described (22). In short, markers of interest were 1) the area of the foveal avascular zone (FAZ), given by the total numbers of pixels within the region; 2) nonperfusion regions defined as dark areas from a binarized image larger than 0.02 mm; 3) vessel density within the ETDRS grid (1 and 3 mm), calculated as the areas not defined as nonperfusion regions over total area of the interested region; and 4) fractal dimension calculated using a box-counting method in a skeletonized image. All the markers were measured in the superficial capillary plexus (SCP) and deep capillary plexus (DCP). We excluded scans with image quality below 40 according to Topcon software.

Retinal-vessel oxygen saturation measurements were performed in 50° disc-centered images using the Oxymap Model T1 (23). In short, two concentric circles were semiautomatically placed around the optic disc with diameters 1.5 times and 3 times the optic disc diameter. The largest venule and arteriole in each quadrant were annotated in length between 50 and 200 um (24). We performed both mean and quadrant arterial and venular saturation analyses. Images with quality below 5 were excluded according to the Oxymap software.

The study was performed according to the tenets of the Helsinki Declaration, and relevant permissions were obtained from the Region of Southern Denmark’s record of data processing activities (Journal no. 20/34731) and from the Danish National Committee on Health Research Ethics (S-20200050). The study was registered at with ClinicalTrials.gov before initiation (identifier NCT04610749). Written informed consent was obtained from all participants.

Study sample characteristics are presented in Table 1. Categorical data are presented with numbers and percentages, and numerical data are presented with medians and interquartile range (IQR). Given that data from both eyes of participants were included, mixed-model regression analysis with cluster robust SE was used to test for difference in retinal markers between persons with and without MCI. We used an unadjusted crude model and a multivariable model adjusting for age, sex, mean arterial pressure, and presence (yes/no) of DR. Because of the interaction between DR and MCI, we reported results for the entire group as well as stratified in groups with and without overt DR (Table 5 as discussed in Results). In a secondary analysis, we calculated area under the curve (AUC) for retinal parameters that were associated with MCI adjusted for age, sex, mean arterial pressure, and presence of DR (Supplementary Material). Missing data were considered missing at random and included abdominal circumference (n = 1), total cholesterol (n = 1), LDL cholesterol (n = 2), estimated glomerular filtration rate (n = 1), and albumin excretion rate (n = 1). There were no missing values for data available for regression analysis. P values < 0.05 were considered statistically significant. The study included multiple analyses and, thus, is at risk for spurious findings. We performed Bonferroni corrections based on 83 initial analyses. Results that were still statistically significant after corrections are annotated in the tables. Abovementioned statistics were performed with STATA, version 17.0 (StataCorp LLC, College Station, TX).

Sample size was based on previous results regarding retinal vessel saturation and prevalence of MCI in T2DM. Power was determined by conducting a statistical power analysis using PROC POWER in SAS. With 200 included persons using a significance level of 5%, the power would be 86%. With a reduction to 120 included participants, the power would be reduced to 80% at a significance level of 5%.

Data are available upon reasonable request to the author.

Overall, 134 participants (n = 245 eyes) were included in this study (Fig. 1). Upon reviewing MoCA scores, 65 participants scored below 26 points, of whom 38 participants (28% of the overall group) were categorized with MCI based on NTB results. The number of total impaired cognitive domains varied with 17, 11, 8, and 2 participants impaired in 1, 2, 3, and 4 domains, respectively. The most frequently impaired domain was memory (n = 18; 47%), and the least frequently impaired was language (n = 8; 21%) (Fig. 2).

Compared with individuals without MCI, participants with MCI had completed fewer years of education (9.0 years vs. 11.0 years; P = <0.001) and were more often taking glucose-lowering treatment, excluding insulin (97% vs. 83%; P = 0.028) (Table 1). Persons with MCI had statistically significantly worse BCVA (84.0 vs. 84.5; P = 0.035) compared with individuals without MCI, but there was no difference between history of cataract surgery or presence of any overt DR. There were no differences between presence of DR (yes/no) and MCI (32% vs. 21%; P = 0.23). In addition, sex, age, diabetes duration, marital status, BMI, smoking, blood pressure, or history of ischemic heart disease was comparable between the two groups. There were no differences detected in HbA_1c, cholesterol levels, estimated glomerular filtration rate, or albumin excretion rate between those with or without MCI.

In a mixed regression model adjusted for age, sex, mean arterial pressure, and presence of DR, mRNFL was thinner in the superior zone (inner region; 22.3 µm vs. 24.1 µm; P = 0.007), nasal zone (inner region; 20.5 µm vs. 21.7 µm; P = 0.049), and nasal zone (outer region) (43.3 µm vs. 46.5 µm; P = 0.032) in individuals with MCI compared with those without MCI. This was also seen in the mGCL, with thinning in the inner region (P = 0.029), inner superior (P = 0.029), nasal (P = 0.010), and inferior (P = 0.027) zone in individuals with MCI compared with those without MCI. Likewise, participants with MCI had central macular thinning in the inner superior (P = 0.018) and nasal (P = 0.033) zones when compared with those without MCI, after adjustment of age, sex, and presence of DR (Table 2).

There was no statistically significant difference in vessel width for either arterioles or venules, vessel tortuosity, density, or fractal dimension in the crude or adjusted model between groups (with or without MCI) (Table 3). Likewise, there was no statistically significant association between OCT-A markers, including FAZ, macular vessel density, or fractal dimension, in the SCP or DCP and presence or absence of MCI (Table 4).

Compared with those without MCI, participants with MCI had lower venular saturation of oxygen in the upper nasal quadrant (61.0% vs. 56.4%; P = 0.028) in the adjusted model, but this was no longer statistically significant after stratification for presence of DR in the model (Table 5). There were also no statistically significant differences in mean retinal arterial or venular oxygen saturation levels between both groups. In the adjusted model for age, sex, mean arterial pressure, and presence of DR, the AUC varied between 0.65 and 0.68 for retinal parameters for which we found a statistical significant association (Supplementary Material).

In this study, 28% of participants with T2DM who fulfilled our criteria of MCI had retinal neurovascular alterations based on the presence of a decreased thickness in the mRNFL as well as lower venular oxygen saturation in the upper nasal quadrant. We did not find any statistically significant differences in retinal vessel caliber, tortuosity, density, fractal dimension, or OCT-A markers.

The prevalence of MCI in people with T2DM in this study was lower than that found in a recent meta-analysis, which reported prevalence of 45%, with rates ranging from 22% to 68% (25). That systematic review included 10 studies from Asia and two from Europe and reported a lower overall prevalence of MCI in European studies (37% vs. 46%). However, this was not directly comparable to our study, because the European studies used the Mini-Mental State Examination (MMSE) and MoCA to assess MCI, whereas MoCA was used as screening for possible cognitive dysfunction in the present study, supplemented by a robust NTB (26,27). In general, MMSE and MoCA are used as screening tests for cognitive impairment, which may have resulted in false-positive MCI cases in the aforementioned studies and caused the observed higher prevalence compared with our study.

The thinning of mRNFL and mGCL found in the present for those with T2DM and MCI is comparable with previous findings in samples without diabetes (28). The thinning of the retinal layers may reflect a potential concurrent retinal and cerebral neurodegeneration. A recent, large MRI study including 2,131 participants reported no association between mRNFL thinning and smaller brain volume. However, mRNFL thinning was correlated with volume reduction in the occipital–parietal cortex in people with multiple sclerosis (29,30). In this study, most of the differences in retinal parameters were found in the inner region. This could indicate that MCI is associated with lesions in specific retinal areas. This is strengthened by MRI studies that have shown an association between regional changes in the mRNFL and regional cerebral atrophy in individuals with high genetic risk of Alzheimer disease (31); however, it should be noted that results are inconsistent, and others have not found a positive association (32). Therefore, it remains uncertain whether retinal thinning could be a proxy measurement for cerebral degeneration.

We did not find any differences in retinal fractal dimension. Previous findings regarding fractal dimension and MCI have been inconsistent. For example, a population-based, cross sectional study of 1,202 participants found that cognitive impairment was more common in those with lower fractal dimensions, but in a subgroup analysis stratified for diabetes, this association was no longer significant (33). Furthermore, a recent cross-sectional study of 1,431 participants, of whom 22% had diabetes, found no significant association between fractal dimension and MCI (34). Differences in results may be due to different variables considered relevant for the regression model used. Thus, whereas Cheung et al. (33) adjusted for age, sex, low income, low education, hypertension, smoking, hyperlipidemia, and chronic kidney disease, O’Neill et al. (34) corrected for age, sex, alcohol consumption, smoking status, educational attainment, physical activity, history of cardiovascular disease, hypertension triglycerides, diabetes, medication, mean arterial blood pressure, BMI, and HDL.

A previous cross-sectional study of 137 individuals with newly diagnosed T2DM found a higher venular tortuosity in persons diagnosed with cognitive dysfunction by a telephone interview, but no alterations were observed in vessel width, fractal dimension, or arterial tortuosity (9). Our study does not support an altered venular tortuosity in individuals with T2DM and MCI, but differences in results may be due to a dissimilarity in the analysis software used to assess vessel tortuosity. Differences in results may also relate to differences in the accuracy of the diagnosis of cognitive impairment between studies (phone interview versus detailed NTB). In general, retinal vessel width, fractal dimension, and retinal vessel tortuosity has been associated with other disorders, including cardiovascular disease (35) and cerebrovascular disease (36), which may indicate that this marker, on its own, may be too unspecific.

In this study, the retinal venular oxygen saturation in the nasal quadrant appeared to be lower in participants with MCI when compared with those without MCI, although this was not statistically significantly different after statistical adjustments. A small study including 42 people with MCI and healthy control participants found increased mean venular and arterial saturation in those with MCI; however, diabetes was an exclusion criterion and the study did not include a quadrant analysis, thus making comparison between studies difficult (37).

Strengths of the present study include its prospective nature and the in-depth neuropsychological evaluation, which would be expected to have provided an accurate assessment of MCI, but we did not include subjective cognitive complaints. Limitations include the relatively few participants and the study’s cross-sectional design, which does not allow causal relationships to be examined. Furthermore, gradings of retinal markers were not obtained by investigators masked to the cognitive status of participants. Last, caution should be taken when interpreting the data presented, which has the potential for spurious associations because of multiple testing. It should be noted that only a few associations were statistically significant, but these might help guide future research.

In conclusion, we found MCI in almost one in three people with T2DM, as well as changes in retinal structural and metabolic markers. Longitudinal studies are warranted to further investigate the temporal associations of structural, metabolic, and functional retinal markers in people with T2DM and MCI. In this regard, the Retinal and Cognitive Dysfunction in Type 2 Diabetes: Unravelling the Common Pathways and Identification of Patients at Risk of Dementia (RECOGNISED) study (https://www.recognised.eu/), funded by the European Union’s Horizon 2020, which includes cross-sectional and longitudinal prospective clinical studies as well as experimental basic science studies, aims to elucidate biomarkers, including retinal biomarkers, of risk of cognitive impairment as well as pathogenic mechanisms of disease in people with T2DM.

Acknowledgments. Participants in this study were also recruited for the RECOGNISED (Retinal and Cognitive Dysfunction in Type 2 Diabetes: Unravelling the Common Pathways and Identification of Patients at Risk for Dementia) study (ClinicalTrials.gov identifier NTC04281186), if deemed eligible. The authors acknowledge the RECOGNISED consortium, funded by the European Union’s Horizon 2020 research and innovation program under grant agreement 847749.

Funding. This work was supported by Odense University Hospital PhD Foundation (grant 4339) and Overlægerådets forskningsfond (grant A4651).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. F.N.P. recruited participants and analyzed the data except OCT-A images, which were analyzed by D.Y. and C.Y.C. F.N.P. and L.S. performed the statistical analysis. F.N.P. wrote the manuscript with input from N.L., G.J.B., L.E., R.S., T.P., F.P., and J.G. F.N.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.