Sequence and Detectability of Changes in Macular Ganglion Cell Layer Thickness and Perfusion Density in Early Glaucoma

Abstract

Purpose: To investigate whether macular perfusion density (PD) changes measured over time by optical coherence tomography angiography (OCTA) are detectable before progressive macular ganglion cell layer (GCL) thinning in early glaucoma.

Methods: This prospective longitudinal cohort study involved patients with early open-angle glaucoma and healthy subjects imaged by OCT and OCTA every 4 months. GCL thickness and macular PD were evaluated in 16 tiles in the macula. We estimated baseline percentage losses of GCL thickness or macular PD in glaucoma patients with age-corrected normative values derived from the healthy subjects. Additionally, the threshold slope separating glaucoma patients from healthy subjects with 90% specificity was used to determine the number of patients with steeper slopes than the threshold slope.

Results: Eighty patients with glaucoma and 42 healthy subjects were included. In eight tiles (50%), patients with a significant macular PD slope had a significantly greater baseline percentage loss of GCL thickness relative to macular PD compared to patients without a significant macular PD slope. Furthermore, in 15 tiles (94%), a greater baseline percentage loss of GCL thickness relative to PD was significantly correlated with faster PD slopes. In contrast, only one tile (6%) showed these significant trends for GCL slopes. The number of patients with faster GCL slopes than threshold slopes was significantly larger than patients with faster PD slopes in 12 tiles (75%).

Conclusions: A decrease in GCL thickness precedes a measurable decrease in macular PD. Early glaucomatous progression is more frequently detectable with changes in GCL thickness compared to macular PD.

Optical coherence tomography angiography (OCTA) is becoming a widely used diagnostic tool in many posterior pole diseases. It identifies perfusion in blood vessels by detecting reflectance changes in backscattered light during OCT imaging. These changes are thought to be due to movement of blood cells, allowing noninvasive visualization and quantification of retinal microcirculation.¹ 

It has been demonstrated that perfusion density (PD) in the superficial vascular plexus of the macula detected by OCTA decreases with increasing severity of glaucoma² and that this decrease can be observed in the early stages of the disease.³ However, studies on diagnostic accuracy based on imaging at one time point are equivocal in concluding whether macular PD measurements are inferior,^4,5 superior,^6,7 or comparable^8,9 to macular ganglion cell layer (GCL) thickness obtained by OCT. These findings make it unclear whether macular GCL thinning precedes macular PD loss or vice versa. Only a limited number of longitudinal studies have investigated whether GCL thickness or macular PD is more sensitive in detecting glaucomatous progression.^10–12 Generally, these studies have suggested that changes in macular PD were more evident than GCL thinning¹⁰ and more closely related to visual field change.¹¹ However, these studies included patients with relatively advanced damage, and, to date, there is no evidence from longitudinal studies whether GCL thinning or macular PD loss is more valuable in detecting changes in early glaucoma. This knowledge may also provide clues as to whether macular PD changes lead to GCL thinning. 

Normal age-related changes account for a large proportion of retinal nerve fiber layer (RNFL) and GCL thickness changes observed in glaucoma patients.^13,14 Similarly, it is possible that age-related changes in PD in healthy subjects, demonstrated by some investigators^12,15 but not others,¹⁰ could account for a proportion of the changes observed in progressive glaucoma. Currently, no published longitudinal studies on changes in PD observed in glaucoma patients factored in normal age-related changes, which are important in determining the value of PD in assessing glaucomatous progression. 

This study had two objectives. The first was to determine whether macular PD changes could be detected earlier than GCL thinning in early glaucoma, and the second was to evaluate the performance of GCL thickness and macular PD measurements, after accounting for age-related changes, in detecting changes in early glaucoma. 

Methods

The present study reports results from an ongoing prospective study on the value of macular imaging and perimetry for detecting the progression in patients with early glaucoma. Patients with early open-angle glaucoma and healthy control subjects are tested every 4 months with both conventional and macular standard automated perimetry, in addition to OCT and OCTA imaging of the optic nerve head and macula. The study received approval from the Nova Scotia Health Research Ethics Board of the Nova Scotia Health Authority and adhered to the tenets of the Declaration of Helsinki. All involved participants provided written informed consent. 

Participants

Glaucoma patients were recruited from the Eye Care Centre at Nova Scotia Health, and healthy subjects were recruited from the non–hospital-based community via local media advertisements. The inclusion criteria for glaucoma patients were (1) clinical diagnosis of open-angle glaucoma, including pseudoexfoliation glaucoma; (2) visual acuity ≥ 6/12; (3) clinically determined glaucomatous changes of the optic nerve head; and (4) glaucomatous visual field loss, defined as outside normal limits by the Glaucoma Hemifield Test (24-2 test pattern of the Humphrey Field Analyzer; Carl Zeiss Meditec, Dublin, CA, USA). Exclusion criteria were (1) non-glaucomatous ocular disease, (2) visual field mean deviation (MD) worse than −6 decibels (dB) with the 24-2 test pattern, (3) systemic disease or treatment affecting the visual field, and (4) refraction exceeding 6-diopter (D) equivalent sphere or 3-D astigmatism. 

The inclusion criteria for healthy subjects were (1) visual acuity ≥ 6/12, (2) normal eye examination with intraocular pressure (IOP) ≤ 21 mmHg, and (3) normal visual field defined as within normal limits by the negative Glaucoma Hemifield Test. Exclusion criteria were (1) chronic ocular disease, (2) systemic disease or treatment affecting the visual field, and (3) refraction exceeding 6-D equivalent sphere or 3-D astigmatism. One randomly selected eye was included in the analysis if both eyes were eligible. We included participants with at least 2 years of follow-up. 

Perimetry

Perimetry was performed, immediately prior to imaging, with the 24-2 and 10-2 test patterns of the Humphrey Field Analyzer with the Swedish Interactive Thresholding Algorithm and a Goldmann size III stimulus with the appropriate near correction. For this study, only baseline values were reported. 

OCT and OCTA Imaging

The OCT images were acquired with a raster scan pattern covering 30° horizontally and 25° vertically with 61 horizontal B-scans, each containing 768 A-scans, averaged nine times (SPECTRALIS OCT2; Heidelberg Engineering, Heidelberg, Germany). Images were captured with real-time eye tracking to stabilize the images during acquisition. The instrument software (Heidelberg Eye Explorer; Heidelberg Engineering) segments the GCL and calculates its thickness within an area of 24° × 24°, consisting of 8 × 8 tiles, each subtending 3° × 3°. The OCTA images were obtained with a raster scan pattern centered on the fovea, subtending 15° × 15° and containing 384 horizontal B-scans. Each B-scan contained 768 A-scans, with an average of five scans per B-scan. Because of the difference in the OCT and OCTA imaging fields, only tiles in OCT imaging (assigned numbers 1 through 16) (Fig. 1) that completely overlapped with the OCTA were used. Only well-centered images with quality score ≥ 25 dB were included in the analysis. Poor-quality images, with residual motion artifacts that were determined by a single evaluator (RT) or with quality score of <25 dB, were reacquired and included in the analysis if they were satisfactory and met quality score criteria. Image segmentation in these images was also checked. All data were converted to right-eye format. 

Figure 1.

 

Schematic diagram showing the analyzed tiles in this study. The blue outer square indicates the area measured by OCT, which outputs GCL thickness for each of the 8 × 8 tiles. The red square indicates the area measured by OCTA. The black inner square indicates the 16 (4 × 4) tiles, fully overlapped by OCT and OCTA, which were analyzed in this study. Each tile represents an area of 3° × 3°. The tile numbers in the analysis are indicated.

Perfusion Density

Two-dimensional OCTA images of the superficial vascular plexus were segmented and calculated by the instrument software for export. OCTA images were binarized and denoised to enhance vessel detection according to previously published methods.^5,16,17 Binarized images were created using an adaptive thresholding algorithm to distinguish areas with flow from areas without flow. Objects smaller than 6 pixels were considered noise and removed. Small holes, defined as clusters of flow less than 10 pixels within the flow-free region, or vice versa, were filled with flat disk-shaped structuring elements with a radius of 1 pixel. The images were divided into 16 tiles, in the same manner as the OCT images. Macular PD was defined as the percentage of pixels with detected flow divided by the total number of pixels in the image in each of the 16 tiles. Image processing to derive macular PD measurements was accomplished with Python 3.9.13 with the opencv-python 4.7.0.72 library. 

Calculation of Differences in Baseline Percentage Loss Between GCL Thickness and Macular PD

We used the OCT and OCTA data obtained at the first visit of healthy subjects to compute relative baseline percentage losses of GCL thickness and macular PD in glaucoma patients. Univariate regression analysis, with age as an independent variable, was performed for GCL thickness and macular PD in each of the 16 tiles using these data of healthy subjects. The coefficients of age and intercepts of the regression analysis were used to create the regression equation with age for each of the 16 tiles for GCL thickness and macular PD. Then, the age of each glaucoma patient was substituted into the equation, and the expected normal value in each of the 16 tiles was computed. The baseline percentage loss for each glaucoma patient in each of the 16 tiles was then calculated according to the following formula:  

\begin{eqnarray*} && Baseline{\rm{\ }}Percentage{\rm{\ }}Loss{\rm{\ }}\\ && = {\rm{\ }}[ ( {Estimated{\rm{\ }}Normal{\rm{\ }}Value - Actual{\rm{\ }}Value} ) \\ && {\rm{\ }} \div {\rm{\ }}Estimated{\rm{\ }}Normal{\rm{\ }}Value ]{\rm{\ }} \times {\rm{\ }}100\end{eqnarray*}

 

The difference in baseline percentage loss was derived by subtracting the baseline percentage loss of GCL thickness from the baseline percentage loss of macular PD using the following formula:  

\begin{eqnarray*} && \textit{Difference}{\rm{\ }}in{\rm{\ }}Baseline{\rm{\ }}Percentage{\rm{\ }}Loss{\rm{\ }} \\ && = \left( {Baseline{\rm{\ }}Percentage{\rm{\ }}Loss{\rm{\ }}of{\rm{\ }}PD} \right) \\ && {\rm{\ }} - {\rm{\ }}\left( {Baseline{\rm{\ }}Percentage{\rm{\ }}Loss{\rm{\ }}of{\rm{\ }}GCL} \right)\end{eqnarray*}

Therefore, if the difference in baseline percentage loss was positive, then macular PD loss was greater. If it was negative, then GCL thickness loss was greater. 

Statistical Analysis

Characteristics of glaucoma patients and healthy subjects were compared with a Mann–Whitney U test and χ² test. The regression analysis using the least-squares method was performed to determine the GCL thickness and macular PD slopes. A negative slope with P < 0.05 was considered statistically significant. The number of significant GCL thicknesses and macular PD slopes were compared using the χ² test. The difference in baseline percentage loss was compared among glaucoma patients with and without significant GCL thickness or macular PD slopes with a Mann–Whitney U test. The correlation between the difference in the baseline percentage loss of GCL thickness and macular PD and the GCL thickness or macular PD slopes was assessed with Spearman's rank correlation analysis. 

For each GCL thickness and macular PD slope value, we determined how well glaucoma patients and healthy subjects could be differentiated (i.e., sensitivity and specificity, using receiver operating characteristics analysis) for each of the 16 tiles. The threshold slope that could differentiate glaucoma patients from healthy subjects with a specificity of 90% was determined. Glaucoma patients with a steeper slope than the threshold slope were defined as patients with glaucomatous progression of GCL thickness or macular PD. The numbers of patients with glaucomatous progression of GCL thickness and macular PD were compared by the χ² test. Effect sizes were calculated using Cohen's d to compare GCL thickness and macular PD between glaucoma patients and healthy subjects, with the healthy subjects set as the reference to ensure positive values. 

A linear mixed-effect (LME) model was used to analyze factors affecting GCL thickness or macular PD slopes. The follow-up measurements of global GCL thickness or macular PD of all 16 tiles were used as dependent variables. Diagnosis of glaucoma, baseline global GCL thickness or macular PD, baseline age, IOP, image quality, and follow-up time (slope in healthy subjects) were used as fixed effects. To determine the effects on the slopes, interaction terms of follow-up time with diagnosis of glaucoma, baseline measurement, and baseline age were included. Random effects per participant included intercept and IOP. For the LME model, baseline age, baseline GCL thickness, and baseline macular PD were centered on the mean baseline in healthy subjects (baseline age, 66.3 years; baseline GCL thickness, 43.2 µm; baseline macular PD, 31.8%). 

Data analyses were performed with Python 3.9.13 and R 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria) with the lme4 1.1-31¹⁸ and lmerTest 3.1.3¹⁹ packages for LME analysis. Statistical significance was assumed when P < 0.05. 

Results

Eighty eyes of 80 glaucoma patients and 42 eyes of 42 healthy subjects were included in the study. All participants, except one healthy Asian subject, were white Caucasians. Eight (10%) glaucoma patients had evidence of pseudoexfoliation. Glaucoma patients were non-significantly older than healthy subjects (medians, 70.6 and 65.7 years, respectively); however, as expected, they had greater visual field, GCL thickness, and macular PD loss (Table 1). A moderate effect size of 0.5 was observed for macular PD, whereas a much larger effect size of 1.4 was found for GCL thickness, highlighting a greater difference between the glaucoma patients and healthy subjects in GCL thickness. The follow-up period of glaucoma patients ranged from 2.5 to 4.3 years (with a range of six to 14 examinations), whereas that of the healthy subjects ranged from 2.1 to 4.4 years (with a range of six to 13 examinations). Healthy subjects had a slightly longer follow-up compared to glaucoma patients (medians, 4.1 and 3.9 years, respectively) (Table 1), but there were no significant differences in the number of examinations (10 each). Of a total of 1250 images analyzed, we removed 76 images (6%) that were thought to be impacted by motion artifacts. 

Table 1.

 

Participant Baseline and Follow-Up Characteristics

 

Significant GCL thickness slopes were observed in 12 to 27 of the glaucoma patients (15%–34%) in each tile, compared to four to 13 patients with significant macular PD slopes (5%–16%). In each tile, the proportion of glaucoma patients with significant GCL thickness slopes was greater than those with significant macular PD slopes. This difference in proportions was significant in 14 of the 16 tiles (88%; see Supplementary Table S1). In contrast, significant GCL thickness slopes were observed in three to 11 healthy subjects (7%–26%), compared to zero to nine subjects for significant macular PD slopes (0%–21%). Only one tile (6%) showed a difference in proportion that was statistically significant (see Supplementary Table S1). 

Regression coefficients of age (range, –0.27 to 0.06 µm/year for GCL thickness and –0.15% to 0.00%/year for macular PD) and intercept (range, 29.78–65.11 µm for GCL thickness and 24.47%–44.05% for macular PD) from which age-corrected expected values of GCL thickness and macular PD of glaucoma patients were calculated are shown in Supplementary Table S2. The median baseline percentage loss of GCL thickness (range, 5.2%–36.6%) and macular PD (range, –6.7% to 12.0%) using these age-corrected expected values in each tile is shown in Supplementary Figure S1. The differences in baseline percentage loss of glaucoma patients calculated by subtracting GCL thickness loss from macular PD loss are shown in Figure 2. The median values of the differences (range, –23.3% to –5.7%) were negative in all 16 tiles, meaning that baseline percentage loss of GCL thickness was higher than that of macular PD. 

Figure 2.

 

Differences in baseline percentage loss of GCL thickness and macular PD in the glaucoma patient. Differences were calculated by subtracting the baseline percentage loss of GCL thickness from the baseline percentage loss of macular PD. Numbers represent medians of the difference; quartiles are shown in parentheses.

There was only one tile (6%) in which the differences in baseline percentage loss among glaucoma patients with and without significant GCL thickness slopes during follow-up reached statistical significance (Fig. 3). In contrast, there were eight tiles (50%) in which the differences in baseline percentage loss among glaucoma patients with and without significant macular PD slopes during follow-up were statistically significant (Fig. 3). Additionally, in glaucoma patients, each of the 16 tiles showed a correspondingly lower correlation between GCL thickness slope and difference in baseline percentage loss (Fig. 4) compared to macular PD slope and difference in baseline percentage loss (Fig. 5). These findings indicate that glaucoma patients with significant macular PD slopes tended to have a higher baseline percentage loss of GCL thickness relative to macular PD. 

Figure 3.

 

Differences in baseline percentage loss among glaucoma patients with (top left) and without (top right) significant GCL thickness slopes and with (bottom left) and without (bottom right) significant macular PD slopes during the follow-up. Numbers represent median (interquartile range) differences in baseline percentage loss. *Tiles in which differences in the baseline percentage loss among glaucoma patients with and without significant slopes were statistically significant.

Figure 4.

 

Scatterplots of GCL thickness slopes and differences in baseline percentage loss in glaucoma patients. The dots indicate differences in baseline percentage loss calculated by subtracting GCL thickness loss from macular PD loss (horizontal axis) and GCL thickness slope during the follow-up period (vertical axis) for each glaucoma patient. The darker the point, the greater the overlap of glaucoma patients. There was a significant correlation between GCL thickness slope and the difference in baseline percentage in only Tile 12. Spearman's correlation coefficient is indicated by r.

Figure 5.

 

Scatterplots of macular PD slopes and differences in baseline percentage loss in glaucoma patients. The dots indicate difference in baseline percentage loss calculated by subtracting the GCL thickness loss from the macular PD loss (horizontal axis) and macular PD slope during the follow-up period (vertical axis) for each glaucoma patient. The darker the point, the greater the overlap of glaucoma patients. There were significant correlations between macular PD slope and difference in baseline percentage loss in all tiles except Tile 13.

The number of glaucoma patients with steeper GCL thickness and macular PD slopes than the respective threshold slopes that distinguish glaucoma patients from healthy subjects with 90% specificity are shown in Figure 6. There were significantly more glaucoma patients with steeper GCL thickness slopes than steeper macular PD slopes in 12 tiles (75%) (Fig. 6). 

Figure 6.

 

Number and percentage of glaucoma patients with steeper GCL thickness (left) and steeper macular PD (right) slopes than the respective values that distinguish glaucoma patients from healthy subjects with 90% specificity. In each tile, more glaucoma patients had steeper GCL thickness slopes compared to macular PD slopes. *Tiles with statistically significant differences.

The results of the LME model analysis for global GCL thickness and macular PD are shown in Table 2. The GCL thickness slope in healthy subjects was –0.22 µm/year (P < 0.01), whereas the macular PD slope in healthy subjects (0.13%/year) was not significant. The differences in slopes between glaucoma patients and healthy subjects were –0.17 µm/year (P < 0.01) for GCL thickness and –0.22%/year (P = 0.04) for macular PD. There were no significant effects of baseline GCL thickness (0.00 µm/year/µm), and baseline age (0.00 µm/year/year) on GCL thickness slope, whereas there were slight effects of baseline macular PD (–0.04%/year/%; P = 0.01) and baseline age (–0.02%/year/year; P < 0.01) on macular PD slope. 

Table 2.

 

Fixed-Effect Coefficients From LME Models for GCL Thickness and Macular PD

 

Discussion

A substantial amount of research has been conducted to evaluate the role of OCTA in glaucoma. Earlier studies suggested that it could be a valuable tool to detect glaucoma in its earliest stages, as decreases in PD may indicate a functional deficit by signaling a decrease in metabolic demand.^3,9 Even if the origin of glaucomatous damage is at the level of the optic nerve head,^20,21 it is conceivable that changes in PD more distally in the superficial vascular plexus could occur before optic nerve structural changes or changes in GCL thickness are detected. 

However, because most clinical studies on OCTA in glaucoma are cross-sectional, they do not provide strong evidence on whether PD changes precede structural change.²² The results of the present study showed that the difference in baseline percentage loss between GCL thickness and macular PD, imaged with the SPECTRALIS, tended to be greater in glaucoma patients with a subsequent significant macular PD slope than in glaucoma patients without a significant macular PD slope. Furthermore, macular PD slopes during the follow-up were significantly correlated with the difference in baseline percentage loss between GCL thickness and macular PD in most tiles of the imaged area. In marked contrast, this correlation was not observed between GCL thickness slopes and the differences in baseline percentage loss between GCL thickness and macular PD. These results indicate that more significant loss of GCL thickness at baseline, relative to macular PD, more likely leads to a subsequent decrease in macular PD, but not vice versa. Therefore, this prospective study supports the hypothesis that GCL thinning occurs prior to detectable changes in macular PD and corroborates our previous cross-sectional observations.¹⁶ 

We found that GCL thickness slopes distinguished glaucoma patients and healthy subjects better than macular PD slopes, suggesting that GCL thickness may be a better index than macular PD for detecting early glaucomatous progression. Consistent with our findings, Hou and colleagues¹⁵ reported that the number of glaucoma suspect eyes (glaucomatous optic neuropathy without visual field damage) that showed a significant decrease in ganglion cell complex thickness was 1.7 times greater than eyes showing a significant change in macular PD, suggesting that changes in GCL thickness likely occur prior to detectable changes in macular PD. On the other hand, two longitudinal studies on glaucoma patients with median baseline MDs of –4.2 dB¹² and –6.9 dB,¹⁰ respectively, showed a faster macular PD slope than ganglion cell complex slope. Furthermore, in patients with severe glaucoma with baseline MDs of around –15 dB, macular PD slope but not ganglion cell complex or RNFL thickness slope could distinguish patients with and without visual field progression.¹¹ These results align with our findings and suggest the possibility that changes in GCL thickness are more frequently observed in early glaucoma, whereas changes in macular PD are more frequently observed in severe glaucoma. 

The estimates of GCL thickness slopes in glaucoma patients from the LME analysis in this study are comparable to those of a previous report²³; however, the macular PD slope was shallower compared to previously reported values.^10,12,15 In healthy subjects, we found a significant change in GCL thickness but not in macular PD slopes. This finding may suggest that there are negligible effects of age on macular PD or that the low reproducibility of PD²⁴ may mask age-related changes. Another important factor could be image quality, which has a larger impact on macular PD compared to GCL thickness.^16,17 

Our study is limited in that we evaluated a relatively small area of the macula (12° × 12°). Although similar to other longitudinal studies using macular PD,^10,12,15 it still only covers about one-third of the 68 points in the 10-2 visual field test pattern.^25,26 Structural changes reported in the peripapillary retina or over the larger macular area^27,28 may occur prior to changes in the central macula; thus, our observations on longitudinal changes in macular PD relative to GCL thickness may not be universally applicable. Additionally, there are several devices that image retinal perfusion, each with different methods of image acquisition and image and data analysis; hence, our conclusions may not be applicable to all OCTA devices. We did not include an analysis of the peripapillary PD because segmentation errors in this region are common and can lead to erroneous results.²⁹ Furthermore, unlike the macula where the vascular beds are well defined and laminated, in the optic nerve head the considerable topographical gradients and large blood vessels can cause signal attenuation and increased noise in PD measurements.²⁹ This study is based on a relatively small sample size, and there was a 5-year difference in the median age between the glaucoma patients and the healthy subjects. Although this difference was not statistically significant, there is a possibility that the calculation of estimated normal values of GCL thickness and macular PD based on the data of the healthy subjects may not have been optimal. Using age-matched subjects may have allowed us to more accurately consider the effects of age. Finally, we did not use a grading scale for determination of motion artifacts in the OCTA imaging scale³⁰; therefore, it is possible that there is potential bias due to the lack of a unified criteria for the selection of images. 

In conclusion, the results of this study suggest that a decrease in GCL thickness may precede a decrease in macular PD measured by the SPECTRALIS; however, it is possible that factors such as image quality,^16,17,31 reproducibility,^24,32 and potential sensitivity of current OCTA technology may limit the detection of subtle serial changes. From a practical standpoint, our study indicates that progression in patients with early glaucomatous is more likely to be detected with changes in GCL thickness compared to macular PD. 

Acknowledgments

Supported by a grant from the Canadian Institutes of Health Research (PJT159564 to BCC); by the Glaucoma Research Society of Canada (CAS); and by a Mathers Foundation postdoctoral research award (RT). 

Disclosure: R. Tomita, None; C.A. Smith, None; O.M. Dyachok, None; G.P. Sharpe, None; P.E. Rafuse, None; L.M. Shuba, None; M.T. Nicolela, Heidelberg Engineering (F); B.C. Chauhan, Heidelberg Engineering (F)