Study design and subjects
This was a cross-sectional study carried out between January 2018 and March 2019 in the Oftalmika Eye Hospital in Bydgoszcz, Poland. The research was conducted in accordance with the principles of the Helsinki Declaration and the study protocol was approved by the local bioethics committee. Written informed consent was obtained from all participants. Each participant enrolled in the study was examined by a psychologist and cognitive function was assessed using the Mini-Mental State Examination (MMSE) screening test. An interview with each subject was conducted by a physician to obtain demographic information, medical and neurologic history, and a risk factor profile. All patients included in the study underwent a detailed ophthalmological examination which included: best-corrected visual acuity (BCVA) assessment, tonometry (Icare TAO1i, USA), slit-lamp biomicroscopy to assess iridocorneal angle, and dilated fundus examination. Thickness of the peripapillary RNFL (pRNFL) and retinal macular region were measured using SD-OCT. Retinal vessel density was assessed in the same region using OCTA. Examinations were carried out over one day by a single ophthalmologist.
The group of AD patients were referred from the Psychoneurology of the Elderly Center in Bydgoszcz. AD was diagnosed by a psychiatrist physician according to the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) and the NIA–AA criteria [21]. To confirm the presence of fibrillar brain amyloid, PET imaging with florbetapir F 18 radioligand was performed. Images were constructed using the standard uptake value ratio (SUVr) based on 90–110 min of acquired data. Global values were computed based on the volume weighted average of frontal (superior, middle, and inferior frontal gyrus), parietal (posterior cingulate, superior parietal gyrus, postcentral gyrus, and inferolateral remainder of parietal lobe), and temporal (parahippocampal gyrus, hippocampus, medial temporal lobe, superior, middle, and inferior temporal gyrus) regions. All SUVr images were visually read by an experienced nuclear physician. If SUVr was less than 1.5, AD subjects were classified as amyloid-positive [22]. Patients with mild to moderate dementia (MMSE score, 10–23) qualified for entry into the study. Additional inclusion criteria were a normal intraocular pressure (IOP; < 21 mmHg) and the absence of ocular fundus changes suggestive of glaucoma. Due to low reliability of static perimetry test in patients with AD, the examination was not performed in this group of patients.
Patients with POAG and cognitively normal according to neuropsychological assessment were enrolled in the study based on the presence of features of glaucoma optic neuropathy, accompanied by a decrease in pRNFL thickness corresponding to loss of visual field based on standard automated perimetry (SITA Standard 24-2, Humphrey Field Analyser II, Carl Zeiss Meditec). Glaucomatous visual filed damage was defined as a glaucoma hemifield test (GHT) outside normal limits and a pattern standard deviation (PSD) outside the 95% normal limit, confirmed by at least two consecutive reliable tests (fixation losses ≤ 33%, false-positives, and false-negatives ≤ 20%).
Participants in the healthy control group had an IOP of less than 21 mmHg, normal optical nerve head (ONH) images without asymmetry, pRNFL thickness within normal limits, normal results in visual field examination, defined as a PSD within the 95% confidence interval (CI), and a GHT result within normal limits. Control subjects were ascertained to be cognitively normal according to neuropsychological assessment.
General exclusion criteria included: age below 50 and above 85, BCVA ≤ 0.6, refractive defect above ± 3.0 Dsph, IOP > 23 mmHg, ocular trauma, vascular or non-vascular retinopathies, non-glaucomatous optic neuropathies, macular disease with a history of eye surgery, except uncomplicated cataract phacoemulsification for all groups and uncomplicated anti-glaucoma surgery only for the POAG group when at least 3 months have passed since surgery. People with neurodegenerative diseases other than AD, a history of alcohol abuse or carbon monoxide poisoning, hypothyroidism, or other serious chronic medical conditions were also excluded from the study.
OCTA and SD-OCT acquisitions
The study used the RTVue XR Avanti (Optovue Inc., Fremont, CA, USA) SD-OCT device with AngioVue software (version 2017.1.0.151), which provides non-invasive visualization of the retinal vascular network using the split-spectrum amplitude-decorrelation angiography (SSADA) algorithm. The system is implemented on an existing commercially available SD-OCT platform that provides both retinal thickness and vessel density measurements. By simultaneously acquiring the OCT and OCTA volume of the AngioVue scan and using automatic segmentation, both the vessel density and thickness can be obtained from the same scan with accurate registration of analyzed areas. The OCTA device has the ability to perform 70,000 A-scans per second and allows measurements with an axial resolution of 5 µm using a light source with a wavelength of 840 ± 10 nm and a bandwidth of 45 nm. To correct motion artifacts, OCTA combines orthogonal fast-scan directions (horizontal and vertical) and is equipped with DualTrac Motion Correction Technology [23]. The software is equipped with a three-dimensional Projection Artifact Removal algorithm to reduce projection artifacts in deeper layers from the OCTA volume on a per voxel basis using information from the OCT and OCTA volumes to differentiate the OCTA signal from projection artifacts in situ [24].
The protocol for macular scanning consisted of B-scans covering a 6x6-mm area repeated horizontally and vertically. Each B-scan contained 400 A-scans with the center located at the fixation point. Scanned images of the 4.5x4.5-mm areas centered on the ONH also consisted of two sets of B-scans repeated horizontally and vertically, each consisting of 400 A-scans. For further analysis, only good technical measurements with a scan quality (SQ) index of 6 or higher on a 10-point scale with which a commercial device was equipped, qualified. Measurements with motion artifacts on en face images (irregular patterns of vessels or a blurred boundary of the ONH) were discarded, as well as those with poor segmentation of individual vascular plexuses.
Vessel Density Analysis
The study was conducted on all patients between 10:00 and 16:00 following pupil dilation. Data analyses, including automatic segmentation of the superficial vascular plexus (SVP) and deep vascular plexus (DVP) of the macula and the peripapillary radial peripapillary capillary (pRPC) layer in the ONH area, followed by automatic measurements of vessel density, were performed on commercially available software. Vessel density was calculated as the percentage of area occupied by flowing blood vessels in the selected region. For ONH scans, vessel density was analyzed in the peripapillary area, which extends outwards from the ONH border with an elliptical area between 2–4 mm. The RPC layer was defined as extending from the inner limiting membrane (ILM) to the posterior border of the RNFL. In the macula, analyses of vessel density and retinal thickness were performed on the entire surface of 6x6-mm en face images, inner circle of the Early Treatment of Diabetic Retinopathy Study chart (i.e., foveal area, 1-mm diameter circle), parafoveal area (rings between 1 mm and 3 mm from the center of the fovea), perifoveal area (rings between 3 mm and 6 mm from the center of the fovea), and its sectors. The SVP comprised the area between the ILM and the outer boundary of the inner plexiform layer (IPL), while the DVP comprised the area between the outer boundary of the IPL and the outer boundary of the outer plexiform layer (OPL).
Retinal Layer Thickness Analysis
Thickness of the retinal layer was evaluated using the same 6x6-mm and 4.5x4.5-mm OCTA acquisitions used for the vessel density analysis. Mean retinal thickness in the peripapillary and foveal, parafoveal, and perifoveal areas were output by the software. Furthermore, the software automatically segmented the pRNFL as well as the IRL and ORL of the macula. The IRL includes the RNFL, GCL, and IPL, whereas the ORL includes layers starting from the inner nuclear layer (INL) up to the outer portion of the hyper-reflective line corresponding to the retinal pigment epithelium (RPE) (Fig. 1).
Statistical Analysis
Summary statistics for normally distributed continuous variables are presented as mean ± one standard deviation (SD) and median with interquartile range (IQR) for non-normally distributed variables. Categorical variables are presented as frequencies. Differences between continuous, normally distributed variables were analyzed using Student’s t-test or analysis of variance (ANOVA) with Bonferroni adjustment for multiple tests. Differences among non-normally distributed data were assessed using the Wilcoxon or Kruskal-Wallis test. When multiple patient groups were compared, multiple testing corrections were applied. Differences between categorical variables were assessed using the chi-square test or Fisher’s exact test for independence. To compare distributions of vessel density and retinal layer thickness in HC, AD, and POAG groups, the linear mixed effect model (LMM) was used, which takes into account the correlation between repeated observations for the same individual (inter-eye correlation).
A direct comparison of vessels density and retinal layers thickness, which different dynamic ranges, was obtained by normalizing values as percentage losses. For patients from the AD and POAG groups, the percentage loss of retinal structure compared to the HC was obtained for the considered variables, i.e., vessel density in the SVP and DVP and thickness of the IRL and ORL calculated using the LMM model [25]. Values of percentage losses were adjusted for inter-eye correlation, age, gender, and SQ, where applicable.
The LMM model was applied to study the association between percentage losses of vessel density and thickness of retinal layers, with thickness as the dependent variable and vessel density, age, gender, and SQ as independent variables. Results are reported as the coefficient of determination (R2), which evaluates the amount of variance in the dependent variable explained by the model. To investigate associations between the percentage losses of considered variables and the presence of AD and POAG, generalized estimating equations (GEE) for correlated multinomial responses were applied [26]. Results are expressed as odds ratios with 95% CI per 1% loss of function.
Results are considered statistically significant when the p-value is less than 0.05. Statistical analyses were performed in the R software (version 3.6.2) using the gls function in the lme4 package, r2glmm package, and multgee package.