Optical Coherence Tomography Assessment of Macular Thickness in Alzheimer’s Dementia with Different Neuropsychological Severities

This retrospective case-control study aimed to investigate associations between disease severity of Alzheimer’s dementia (AD) and macular thickness. Data of patients with AD who were under medication (n = 192) between 2013 and 2020, as well as an age- and sex-matched control group (n = 200) with normal cognitive function, were included. AD patients were divided into subgroups according to scores of the Mini-Mental State Examination (MMSE) and Clinical Dementia Rating (CDR). Macular thickness was analyzed via the Early Treatment Diabetic Retinopathy Study (ETDRS) grid map. AD patients had significant reductions in full macula layers, including inner circle, outer inferior area, and outer nasal area of the macula. Similar retinal thinning was noted in ganglion cells and inner plexiform layers. Advanced AD patients (MMSE score < 18 or CDR ≥ 1) showed more advanced reduction of macular thickness than the AD group (CDR = 0.5 or MMSE ≥ 18), indicating that severe cognitive impairment was associated with thinner macular thickness. Advanced AD is associated with significant macula thinning in full retina and inner plexiform layers, especially at the inner circle of the macula. Macular thickness may be a useful biomarker of AD disease severity. Retinal imaging may be a non-invasive, low-cost surrogate for AD.


Introduction
Alzheimer's dementia (AD) is a progressive neurodegenerative disorder that is the most common cause of dementia [1]. Currently, at least 44 million people are estimated to live with dementia worldwide, which is predicted to more than triple by 2050 as the population ages [2]. A definitive diagnosis of AD is based on amyloid β and tau pathology determined by invasive procedures [3]. cognitive examination was obtained from the department of Neurology. Patients with d abetic macular edema (DME), proliferative diabetic retinopathy (PDR), glaucoma, opti neuropathy, and ocular disease that may lead to significant changes in retinal thicknes were excluded.
The included patients had records of OCT exams at ophthalmic visits, as well as OC images taken close to the date of AD diagnosis within the time interval of 1-2 years. OC images were generated by the Spectralis OCT (Heidelberg Engineering, Heidelberg, Ger many). Initial quality control was performed by experienced ophthalmologists and tech nicians during the clinical OCT examination, to filter out images with low resolution o improper format as well as eyes with age-related macular disease (AMD), macular hole (MH), VMT (vitreomacular traction), and epiretinal membrane (ERM), even chorioretina atrophy or prominent posterior staphyloma. Precise matching of patients' age (by birt year) and sex was done to generate the comparison group (control group) of non-AD pa tients. Patients receiving OCT exams between 2013 and 2020, filtered by the exclusion cr teria cited above, were included. Of these patients, under initial matching with sex an age and exclusion of outliers, 175 patients had AD under medication and 188 patients ha no cognitive impairment. Figure 1 shows the study sample inclusion and exclusion algo rithm of the study.

Measurements
The data augmentation procedure was performed by exporting the macular thickness map into different retinal layers, including full layers of macula, nerve fiber layer, ganglion cell layer, and inner plexiform layer, in the total four categories of thickness map. The entire thickness map was exported into the Early Treatment Diabetic Retinopathy Study (ETDRS) grid map, which was divided into nine sectors (one 1-millimeter-diameter circle at the center, four inner-quarter annuluses between the 1-and 3-millimeter-diameter circles, and four outer-quarter annuluses between the 3-and 6-millimeter diameter circles), as previously described [8]. Figure 2 showed the ETDRS grid map adapted from previous studies [9,10], and abbreviations and numbers were employed to represent the nine regions of macular thickness map in the present study.

Measurements
The data augmentation procedure was performed by exporting the macular thickness map into different retinal layers, including full layers of macula, nerve fiber layer, ganglion cell layer, and inner plexiform layer, in the total four categories of thickness map. The entire thickness map was exported into the Early Treatment Diabetic Retinopathy Study (ETDRS) grid map, which was divided into nine sectors (one 1-millimeter-diameter circle at the center, four inner-quarter annuluses between the 1-and 3-millimeter-diameter circles, and four outer-quarter annuluses between the 3-and 6-millimeter diameter circles), as previously described [8]. Figure 2 showed the ETDRS grid map adapted from previous studies [9,10], and abbreviations and numbers were employed to represent the nine regions of macular thickness map in the present study.

Subgroup Analysis
Patients with AD were then divided into subgroups according to MMSE scores and CDR scores ( Figure 1). MMSE is an exam of cognitive function, including orientation, concentration, attention, verbal memory, naming, and visuospatial skills. The exam is based on a total score of 30 points, and the better cognitive function is related to a higher score. As for CDR, it is rated according to six domains of cognitive function, including memory, orientation, judgment, problem solving, community affairs, home and hobbies, and personal care. The result is integrated into a 5-point scale as follows: 0, no impairment; 0.5, questionable impairment; 1, mild impairment; 2, moderate impairment; and 3, severe impairment. Given that no standardized MMSE cutoff values for dementia staging were available, we followed previous studies and classified AD patients into mild cognitive impairment (MMSE score ≥ 18) and moderate or severe impaired cognitive function (MMSE < 18) [11,12]. Because staging was based on CDR scores, patients were classified into one group with more severe cognitive impairment (CDR ≥ 1) and one group with mild cognitive impairment (CDR < 1). For those AD patients without MMSE or CDR records in the database, another group was classified as the group with missing data. The characteristics of age and sex among the three subgroups based on MMSE scores or CDR scores are presented in Appendix Table A1. Specifically, 87 patients who had MMSE scores ≥ 18, 37 patients with MMSE scores < 18, and 51 patients with missing MMSE data were recorded. In the CDRbased group, 58 patients with CDR = 0.5, 48 patients with CDR ≥ 1, and 69 patients with

Subgroup Analysis
Patients with AD were then divided into subgroups according to MMSE scores and CDR scores ( Figure 1). MMSE is an exam of cognitive function, including orientation, concentration, attention, verbal memory, naming, and visuospatial skills. The exam is based on a total score of 30 points, and the better cognitive function is related to a higher score. As for CDR, it is rated according to six domains of cognitive function, including memory, orientation, judgment, problem solving, community affairs, home and hobbies, and personal care. The result is integrated into a 5-point scale as follows: 0, no impairment; 0.5, questionable impairment; 1, mild impairment; 2, moderate impairment; and 3, severe impairment. Given that no standardized MMSE cutoff values for dementia staging were available, we followed previous studies and classified AD patients into mild cognitive impairment (MMSE score ≥ 18) and moderate or severe impaired cognitive function (MMSE < 18) [11,12]. Because staging was based on CDR scores, patients were classified into one group with more severe cognitive impairment (CDR ≥ 1) and one group with mild cognitive impairment (CDR < 1). For those AD patients without MMSE or CDR records in the database, another group was classified as the group with missing data. The characteristics of age and sex among the three subgroups based on MMSE scores or CDR scores are presented in Appendix A, Table A1. Specifically, 87 patients who had MMSE scores ≥ 18, 37 patients with MMSE scores < 18, and 51 patients with missing MMSE data were recorded. In the CDR-based group, 58 patients with CDR = 0.5, 48 patients with CDR ≥ 1, and 69 patients with missing data were reported. Multivariable linear regression models were used later to analyze the associations between macular thickness and MMSE as well as CDR.

Statistical Analysis
Frequencies were used to present the proportions of demographic characteristics. Chi-square tests were used to compare differences between categorical variables. We first did the hypothesis testing for normality for the measures of ETDRS grid macular thickness using the Anderson-Darling test [13], which is one of the common approaches for the normality test, and the results supported the assumptions of normal distribution. We then used independent t-tests or ANOVA tests to compare continuous variables between groups. Multivariable linear regression models were used to compare the differences in ETDRS grid macular thickness between groups (AD versus controls, and different severity level of dementia based on MMSE and CDR) while controlling for age and sex. A two-tailed p < 0.05 was considered significant in all statistical tests. We further used the Bonferroni approach to calculate adjusted p-values among multiple site comparisons to handle multiple testing concerns [14]. All statistical operations were performed using SAS version 9.4 (SAS Analytics, Cary, NC, USA). Table 1 shows the baseline demographic characteristics of age (based on birth year) and sex between matched AD patients (N = 175) and the comparison group (N = 188). No statistically significant differences were noted between these two groups. For AD patients, the mean age was 77.144 years with 65.14% older than 75 years, and 61.14% were female. For the non-AD comparison group, the mean age was 76.76 years with 62.23% older than 75 years, and 61.70% were female. MMSE and CDR scores among AD patients are also presented in Table 1.

Association between Macular Thickness and Alzheimer's Dementia
As shown in Table 2, retinal thickness was thinner in patients with AD in comparison to the control group with normal cognitive function. In the full macula layers, significant differences were observed between the groups in the inner superior The results were similar in the inner plexiform layer (IPL) and ganglion cell layer (GCL), which showed significantly decreased macular thickness in AD, especially in the inner circle part and outer inferior area in the ETDRS grid. However, when the data were defined solely on the NFL layer, no significant differences were noted in macular thickness between the AD group and the control group. Thus, when we converted the statistical results into plots of the ETDRS grid map, the area marked with a star represented a significant reduction of macular thickness in different retinal layers in AD ( Figure 3).

Associations between Macular Thickness and Severity of Dementia via Subgroup Analysis Stratified by MMSE and CDR Scores
In the initial results of MMSE subgroup analysis, significant differences were found in macular thickness in between-group comparisons, but the results were not consistent (Appendix B). After multivariable linear regression, when compared to the control group, group 2, with MMSE scores < 18, showed no significant reduction in any layer of macula thickness, and group 1, with MMSE ≥ 18, showed significantly thinner full macular thickness in the inner superior region (−6.00 μm (95%CI: −9.83, −2.17), Bonferronip-value 0.021) ( Table 3). In the missing data group, the macular thickness was thinner than that in the control group in full layers of the retina, NFL, GCL, or IPL, and in almost every region of the macula.
Regarding CDR scores, the initial results showed no significant differences in between-group comparisons (Appendix B). After multivariable linear regression according When any one of the single layers, including NFL, GCL, and IPL, was statistically integrated to another one, or when the three single layers were all statistically combined, the results were similar, presenting significantly thinner thickness in the inner part of the macula in AD than in the control group. In addition, according to the Bonferroni approach to calculate adjusted p-values, the significant differences were noted in the inner area of macula in full thickness, IPL layer, and in GCL + IPL layer.

Associations between Macular Thickness and Severity of Dementia via Subgroup Analysis Stratified by MMSE and CDR Scores
In the initial results of MMSE subgroup analysis, significant differences were found in macular thickness in between-group comparisons, but the results were not consistent (Appendix B). After multivariable linear regression, when compared to the control group, group 2, with MMSE scores < 18, showed no significant reduction in any layer of macula thickness, and group 1, with MMSE ≥ 18, showed significantly thinner full macular thickness in the inner superior region (−6.00 µm (95% CI: −9.83, −2.17), Bonferronip-value 0.021) ( Table 3). In the missing data group, the macular thickness was thinner than that in the control group in full layers of the retina, NFL, GCL, or IPL, and in almost every region of the macula.  Regarding CDR scores, the initial results showed no significant differences in betweengroup comparisons (Appendix B). After multivariable linear regression according to the Bonferroni approach to calculate adjusted p-values, in group 2, with CDR scores ≥ 1, significant reduction was observed in the macular thickness compared to that of the control group. The full layer of the retina revealed the most prominent results, especially in the inner superior region (−8.13 µm (95% CI: −12.81, −3.46), Bonferroni p-value = 0.006), inner temporal region (−7.34 µm (95% CI: −11.91, −2.78), Bonferroni p-value 0.015), and inner inferior region (−8.58 µm (95% CI: −13.48, −3.68), Bonferroni p-value = 0.006) ( Table 4), and the results were similar in the inner temporal region of the IPL and ganglion cell complex. Interestingly, no significantly thinner macular thickness was found in the NFL layer. In group 1, with CDR < 1, macula thickness was generally decreased but no significant difference was noted via the Bonferroni approach. Macular thickness in the missing data group was thinner when compared to that of the control group, but the greater amount of macular thickness reduction was noted in group 2, with CDR score ≥ 1, rather than in the missing data group. Therefore, these results indicated that the advanced stage of dementia was associated with thinner macular thickness. The association between ETDRS macular thickness of the full retina layer and different groups of neuropsychological severity was presented in Figure 4. Table 4. ETDRS grid macular thickness stratified by CDR scores. CDR: clinical dementia rating; NFL: nerve fiber layers; GCL: ganglion cell layer; IPL: inner plexiform layer; GCC: ganglion cell complex (NFL + GCL + IPL); N: numbers of patients; SD: standard deviation; 95% CI: 95% confidence interval; p-value < 0.05 is considered as statistically significant (*), and the bold numbers mean statistically significance with p-value < 0.05.   Note: Age and sex were controlled and adjusted in the multiple variable regression models.

Discussion
The present study showed that retinal thickness was thinner in patients with AD in comparison to that of people without AD. Significant differences were found in the thickness of full macular layers and IPL thickness, but not in NFL thickness. In addition, major differences were found at the inner sectors of the macula in the ETDRS grid map.
The retina shares similarities with the brain, and the concept of the retina as an exceptionally convenient window by which to assess the neuronal and vascular changes in the central nervous system (CNS) started decades ago. Various retinal changes in AD have been investigated in the last three decades. In 1986, Hinton et al. discovered in postmortem histological research that AD patients possessed prevalent axonal degeneration, decline in a variety of retinal ganglion cells (RCGs), and decline in the thickness of the retinal nerve fiber layer (RNFL) compared to those of age-matched controls [15]. This finding was further supported by electrophysiological research revealing unusual patterns in the electroretinogram as feasible documentation of RGC degeneration in AD [16,17]. Despite several inconsistencies, most studies revealed alterations in visual function, including visual acuity, contrast sensitivity, and color perception, as well as structural change such as a decrease in RNFL and GCL thickness and the pattern of ERG amplitude, a reduction of optic nerve fibers, and an increase of glial reactivity and cell loss in AD retinas. Several animal studies on the analysis of the OCT images showed a statistically significant thinning of the nerve fiber layer in AD mouse retinas compared to wild type controls, though there were some inconsistent findings in different reports. The inconsistency might be due to the young age of the tested animals or may suggest an initial inflammatory response that may lead to cell death and the consequent reduction in thickness later on [18][19][20]. Through the advance of ocular imaging technology, the current OCT technology allows high-resolution in vivo cross-sectional images as well as quantifiable and reproducible calculations of the macula and the possibility of ocular biomarkers for systemic diseases, including AD.
Results of the present study were compatible with previous reports demonstrating thinning of the retina, although in variable sectors [21][22][23][24][25][26][27]. Considering the pathological change of the retina in AD shows the deposition of amyloid β plaques in the retinal ganglion complex (RGC), which is a combination of nerve fibers, ganglion cells, and inner plexiform layers, RGC is the major structure affected in AD patients. Due to centrifugal distribution, there is more RGC in the inner part of the retina, which helps to explain the result showing change of full layer thickness, IPL, and GCL + IPL in the inner circle of macula related to thinner thickness due to AD. In our study, AD patients tended to have thinner thickness of full layer thickness, IPL, and GC-IPL, rather than solely NFL, and the result was similar to several reports [28,29]. The proportion GC-IPL is approximately 50% in RGC, whose cell bodies are over 10 to 20 times the diameter of their axons [29], and we hypothesized that GC-IPL might be more vulnerable to the damage in Alzheimer's dementia (AD). In addition, though reduction of RNFL thickness in AD was reported [23], the evidence of dynamic changes of NFL and GC-IPL during AD progression was never reported [30]. A recently published convolutional neural network (CNN) to detect symptomatic Alzheimer's dementia (AD) using a combination of multimodal retinal images and patient data also showed that GC-IPL maps were the most useful single inputs for prediction [31].
It is highly expected that changes in retinal thickness of AD patients correlates with the severity of AD. Currently, the relationship between macular thickness and the severity of cognitive function impairment is still not well established. A systematic review supported the presence of RNFL thinning and mild cognitive impairment (MCI), particularly amnestic MCI, though less remarkable than the associations with AD [32]. The present study included only AD patients who had received treatment, but not those with MCI. We categorized the AD patients into mild AD and moderate-to-severe AD based on MMSE scores and CDR grading. After multivariable linear regression, the MMSE scores correlated positively with macular thickness, and the association between CDR grading and macular thickness was much more significant. The significantly changed areas were at the inner circle of the macula, compatible with general analysis of AD patients versus controls. Variable sensitivity and specificity of MMSE scores may be due to patients' age, sex, or education status [33], while for CDR scores, the sensitivity and specificity were reported to be as high as 87% and 99% [34], respectively. Thus, CDR scores may be a more objective and consistent parameter for evaluating the clinical cognitive condition. Since patients in the present study were missing certain MMSE or CDR data, the whole missing data group presented similar findings as those with AD. We speculated that patients in the missing data group might the more severe AD patients who could not complete the psychiatric and cognitive examinations. In other words, our results might support a possible association between AD and retinal thinning, which would be very helpful in monitoring the disease course or even treatment response. Integrated with brain imaging and biomarker and neuropsychological assessment, the eye could potentially improve AD risk assessment, detection, and monitoring [35].
This study has several limitations. First, it was a retrospective study, and OCT images taken around the time of AD diagnosis were retrieved, suggesting there may be a gap between the diagnosis or disease severity and the images taken. Some patients might not visit neurology department and ophthalmology department around the same time, and this was the possible reason leading to the gap in the retrospective study. Retrospective design also limits generalization of results to other populations and does not allow the ruling out of selection bias. Second, nearly 40% of cognitive examinations were missing from patients' records at the time of switching from paper to electronic medical records or were simply not done. We are planning to conduct a further prospective longitudinal study where the patients may receive OCT imaging in a timelier manner during the initial diagnosis of AD, which will allow associations between the trend of macular thickness and the change of MMSE or CDR scores within several years to be detected prospectively. However, it should be noted that the present study included large case numbers, and the other systemic or ocular diseases were excluded or matched before analysis.

Conclusions
In conclusion, significant retinal thinning was noted in the retinas of AD patients, especially in full retinal layers and inner plexiform layers at the inner circle of the macula. Results of this study support that macular thickness may be a useful biomarker of the disease severity of AD. Retinal images may be non-invasive, low-cost surrogates for AD.