Assessment of Choriocapillaris Flow Prior to Nascent Geographic Atrophy Development Using Optical Coherence Tomography Angiography

Abstract

Purpose: To assess the relationship between choriocapillaris (CC) loss and the development of nascent geographic atrophy (nGA) using optical coherence tomography angiography (OCTA) imaging.

Methods: In total, 105 from 62 participants with bilateral large drusen, without late age-related macular degeneration (AMD) or nGA at baseline, were included in this prospective, longitudinal, observational study. Participants underwent swept-source OCTA imaging at 6-month intervals. CC flow deficit percentage (FD%) and drusen volume measurements were determined for the visit prior to nGA development or the second-to-last visit if nGA did not develop. Global and local analyses, the latter based on analyses within superpixels (120 × 120-µm regions), were performed to examine the association between CC FD% and future nGA development.

Results: A total of 15 (14%) eyes from 12 (19%) participants developed nGA. There was no significant difference in global CC FD% at the visit prior to nGA development between eyes that developed nGA and those that did not (P = 0.399). In contrast, CC FD% was significantly higher in superpixels that subsequently developed nGA compared to those that did not (P < 0.001), and a model utilizing CC FD% was significantly better at predicting foci of future nGA development at the superpixel level than a model using drusen volume alone (P ≤ 0.040).

Conclusions: This study showed that significant impairments in CC blood flow could be detected locally prior to the development of nGA. These findings add to our understanding of the pathophysiologic changes that occur with atrophy development in AMD.

Atrophic changes in age-related macular degeneration (AMD) are characterized by the loss of photoreceptors, the retinal pigment epithelium (RPE), and choriocapillaris (CC).¹ Inflammatory, genetic, oxidative, and ischemic factors contribute to the multifactorial etiology of AMD, but their relative importance and timing of impact remain unknown.^2,3 

There is contradicting evidence regarding the timing of CC, RPE, and photoreceptor loss with atrophy development in AMD. Several histopathologic studies have suggested that CC loss precedes RPE loss.⁴ For instance, two previous studies found that eyes with early stages of AMD show increased CC dropout below sub-RPE deposits compared to areas without such deposits and that submacular CC dropout could be seen before any evidence of overlying RPE loss.^5,6 A previous electron microscopy study also reported that CC breakdown occurred prior to RPE degeneration in eyes with AMD.⁷ Light microscopy studies suggest choroidal and Bruch's membrane (BM) thickness remain similar in eyes with advanced AMD and aging eyes without disease, yet CC density is significantly reduced in the eyes with AMD.⁸ However, there are also data supporting the opposite notion, suggesting that RPE atrophy precedes CC loss. Utilizing morphometric analysis, Mcleod and colleagues⁹ found full RPE denudement with only partial CC loss in cross-sections through regions with geographic atrophy (GA). Multiple studies focused on histologic evaluation of the GA lesion edge have also observed preserved CC within areas of RPE atrophy.^6,10 Since RPE cells provide metabolic support to the photoreceptors, it has thus been postulated that photoreceptor loss seen in GA is secondary to RPE degeneration.² However, histopathologic studies demonstrated photoreceptor loss in areas with intact RPE, suggesting that damage to the metabolically demanding photoreceptors could be driven directly by ischemia from CC loss.^11,12 Overall, histopathologic assessments have yielded valuable insights into GA pathophysiology. However, their findings are limited by the static nature of histology, which is unable to capture dynamic changes in the CC, photoreceptors, and RPE layers over time. 

The advent of optical coherence tomography (OCT) and OCT angiography (OCTA) has facilitated the in vivo, longitudinal assessment of the RPE and CC in eyes with GA.¹³ A mounting body of literature supports the notion that CC loss precedes RPE degeneration in atrophic AMD. For instance, several studies have reported that eyes that develop atrophic AMD have a significantly larger extent of global CC flow deficits (FDs) compared to eyes that do not.^14–16 Several studies have also reported that an increasing extent of CC FDs around atrophic lesions is associated with a faster rate of lesion growth.^17–19 One of these studies, however, observed that the global extent of CC FDs was more closely associated with GA growth than its extent at the perimeter of existing GA lesions.²⁰ Therefore, it remains to be determined whether CC changes occur in a more localized or global manner prior to GA development. 

OCT imaging has enabled the identification of anatomic changes signifying cell death and survival in the outer retina. Our group has previously described these anatomic changes as nascent GA (nGA), a feature that portends GA formation and is defined by the presence of subsidence of the outer plexiform layer (OPL) and inner nuclear layer (INL) and/or a hyporeflective wedge-shaped band within Henle's fiber layer.²¹ We recently showed that eyes with nGA have a 78-fold increased rate of developing GA compared to eyes without nGA.²² Evaluating eyes prior to the development of nGA, therefore, allows us to understand if CC changes occur prior to the development of GA. 

In this study, we analyzed the CC prior to nGA development as defined on OCT to determine if (1) CC loss precedes nGA development at a global (whole eye) level, (2) CC loss precedes nGA development at a local level, and (3) whether CC loss can predict foci of future nGA development. This analysis was undertaken to provide further insights into the pathophysiology of atrophy development in AMD and the role CC dysfunction plays in this process. We hypothesize that CC loss will precede nGA development at both a whole eye and local level and that measurements of CC loss could thus be used to predict nGA development. 

Methods

This study included participants enrolled in a prospective, longitudinal, observational study of the early stages of AMD at the Centre for Eye Research Australia. This study received institutional review board approval and was conducted according to the International Conference on Harmonization Guidelines for Good Clinical Practice and the tenets of the Declaration of Helsinki. All participants provided written consent prior to enrollment in this study. 

Participants and Procedures

This study included participants who were 50 years or older with bilateral drusen >125 µm; this maximum drusen size criterion fits the criteria for intermediate AMD (iAMD) as defined by the Beckman classification scheme.²³ All participants were required to have a best-corrected visual acuity (BCVA) of 20/40 or better in both eyes at baseline. Participants with late AMD or a history of any systemic, neurologic, or ocular disease affecting the retina at baseline or at any point during the follow-up period were excluded. At baseline, late AMD was defined by the presence of either (1) exudative neovascular AMD, based on the presence of a lesion on fluorescein and/or indocyanine green angiography (which were performed on indication), or a retinal hemorrhage associated with fluid on OCT imaging or (2) atrophic AMD, based on the presence of nGA, or more progressed atrophic lesions on OCT imaging.²⁴ The presence of nGA was defined on OCTA structural B-scans based on the presence of either (1) subsidence of the OPL and INL and/or (2) a hyporeflective wedge-shaped band within Henle's fiber layer.^22,25 

All participants underwent BCVA measurements, retinal imaging, and dilated fundus examination at baseline and approximately 6-month intervals thereafter. To assess changes in the CC blood flow prior to the development of nGA, this substudy included eyes of participants with OCTA imaging performed at the visit of nGA development (n visit) and ≥4 months prior to nGA development (n-1 visit). This substudy also included eyes that did not develop nGA, analyzing the last two visits without late AMD that were ≥4 months apart from each other (considered the n-1 and n visits). OCTA analysis was performed on the first chronologic (n-1) visit included. 

Image Acquisition, Grading, and Annotations

At each visit, OCTA volume scans were obtained using the PLEX Elite 9000 device (Carl Zeiss Meditec, Dublin, CA, USA) over the central 6 × 6-mm region, and consisted of 500 B-scans, each with 500 A-scans and two repeated acquisitions to calculate decorrelation and for image averaging. The PLEX Elite 9000 is a swept source OCT (SS-OCT) instrument operating at a central wavelength of approximately 1060 nm, an A-scan rate of 100 kHz, and an A-scan depth of 3.0 mm. Due to increased RPE penetration, swept source imaging is preferred for CC visualization and quantification.^26,27 OCTA scans were required to have a signal strength of ≥7 and be free from significant artifacts (such as blink or motion artifacts) for inclusion. 

All PLEX Elite 9000 OCTA volume scans were graded for the definite presence of nGA by two experienced graders (RHG and ZW) together. Any disagreements between the two graders were resolved immediately by open adjudication. For eyes graded as having nGA, the lesion boundaries were outlined on each OCTA B-scan by one of the experienced graders (ZW). The lesion boundaries were defined by subsidence of the OPL and INL or the presence of a hyporeflective wedge-shaped band in Henle's fiber layer (Fig. 1E). The boundaries outlined on each B-scan thereby generated a two-dimensional en face projection of the nGA lesion (Fig. 1B). These lesion boundaries were outlined utilizing a custom software (Cross-Modality Annotation Software [XMAS]; Ophthalmic Neuroscience Unit, Centre for Eye Research Australia).²⁸ The en face nGA lesion tracings from the n-visit were spatially registered to the en face images from the n –1 visit utilizing a customized MATLAB (MathWorks, Natick, MA, USA) software designed for longitudinal image registration. Image registration was performed using fiducial points manually placed at corresponding bifurcations in the retinal vasculature in the en face full retinal slab images of the two visits (Figs. 1A, 1F). 

Figure 1.

 

Tracing of nGA and image coregistration. (A) OCTA en face full retinal projection (“N-visit”) showing fiducial points used for image registration at vessel bifurcation (red dots). (B) Annotated en face full retinal projection (“N-visit”); yellow overlay denotes projected nGA tracing as outlined on a B-scan by B-scan basis. (C) Annotated en face uncompensated CC (“N-visit”). (D) Annotated en face compensated CC (“N-visit”). (E) OCTA structural B-scan corresponding to line (I.) in A; the red arrows indicate the lesion boundaries based on OPL and INL subsidence or a hyporeflective wedge-shaped band in Henle's fiber layer. The B-scan shown covers the full extent of green dashed line. (F) En face full retinal projection (“N –1 visit”) showing fiducial points used for image registration at vessel bifurcation (green dots). (G) Annotated en face full retinal projection (“N –1 visit”); yellow overlay denotes co-registered nGA area traced at time of nGA development (“N-visit”; B). (H) Annotated en face uncompensated CC (“N –1 visit”). (I) Annotated en face compensated CC (“N –1 visit”). (J) OCTA B-scan corresponding to line (II.) in F. The B-scan shown covers the full extent of green dashed line.

Image Processing and Analysis

OCTA volume scans obtained at the first visit were processed to analyze how CC FDs differed between eyes and regions that subsequently developed nGA. 

OCTA volume scans were processed through the Choriocapillaris Slab Imaging v0.2 algorithm on the Advanced Retinal Imaging (ARI) Network Hub (Carl Zeiss Meditec) to obtain CC FD measurements. This algorithm uses a compensation scheme based on the method outlined by Zhang et al.,²⁹ which normalizes the OCTA signal using the corresponding OCT signal and thereby weakens the coupling between the OCTA signal and the intensity of the backscattered or reflected OCT beam (Fig. 2C). The algorithm then identifies the BM, and the CC blood flow was examined within a 20-µm en face slab that started 5 µm below the BM. The segmentation generated by the algorithm was reviewed to ensure proper BM delineation. Scans with erroneous BM segmentation were excluded from the study as the algorithm did not allow for manual adjustments in the segmentation. A total of 26 OCTA scans were excluded from analysis due to inadequate segmentation or reduced image quality secondary to significant blink or motion artifact. 

Figure 2.

 

Image processing steps for N –1 visit data. (A) En face choriocapillaris (5–25 µm posterior to BM) structural slab. (B) En face uncompensated CC flow slab. (C) En face compensated CC flow slab combining data from preceding structural and flow slabs. Note the reduction in apparent CC FDs after compensation of drusen shadowing. Note that compensation can reduce but does not eliminate FD generated by drusen shadowing. (D) En face compensated CC flow slab after FD binarization and 5-mm central circle selection, with the white pixels representing areas above the selected FD threshold. Global and local analyses were carried out within the central 5-mm circle only. (E) En face compensated CC FD% map. (F) En face RPE elevation map used to calculate drusen volume.

The en face CC flow slab was then binarized to identify regions with blood flow falling below a threshold that would be considered to represent FDs (Fig. 2D). Since binarization threshold selection can significantly affect CC FD measurements,^30–32 we used an independent training data set to estimate a global binarization threshold for compensated OCTA images. We have previously outlined a theoretical rationale for using global thresholding strategies to detect CC FD in normalized or compensated OCTA data; see Appendix I of Moult et al.³³ The independent training data set was composed of one eye from 14 individuals with iAMD seen at the New England Eye Center, who met all the eligibility criteria in this study and underwent OCTA imaging on the same SS-OCT instrument, using the same protocol. For each training eye, 10 CC FDs were randomly selected from an en face CC flow slab and manually traced by a single reader (EMM). The OCTA values of the pixels within the traced CC FDs were then used to derive the CC FD binarization level. Specifically, the CC FD binarization level was chosen as two standard deviations above the mean OCTA value within the traced CC FDs. Utilizing this independently determined binarization level, compensated CC flow slab images were binarized to generate CC FD maps. The binarized CC FD map was then spatially averaged using a Gaussian kernel (σ = 125 µm) to generate a smoothed image of CC flow deficit percentage (CC FD%; Fig. 2E). CC FDs having an area smaller than 5 pixels (∼172 µm²) were removed to emphasize pathologic CC FDs and reduce noise.³² RPE elevation maps were also generated using the Advanced RPE Analysis algorithm v0.7 on the ARI Network Hub to provide corresponding drusen volume measurements at each location (Fig. 2F). 

For global analyses at the eye level, we calculated the mean CC FD% and total drusen volume within the central 5-mm region (to avoid the potential impact of CC FDs associated with peripapillary atrophy) (Figs. 3A, 3E). For local analyses, we calculated the mean CC FD% and drusen volume within local regions termed superpixels that were generated as cells of a fixed size, each measuring 120 × 120 µm or 10 × 10 A-scans (Figs. 3B, 3F). A superpixel was categorized as having developed nGA if ≥50% of its comprising A-scans were annotated as having developed nGA (Fig. 3K). Local drusen volume was calculated by multiplying the mean RPE elevation values within a superpixel by the superpixel area (14,400 µm²). Similar to the global analysis, local analyses were performed only within the central 5-mm region. 

Figure 3.

 

Local analyses for N –1 visit data. Top row: Local CC FD% analysis. (A) CC FD% map. (B) CC FD% map with overlying 50 × 50 grid for superpixel transformation. (C) Superpixels with averaged CC FD%. (D) Magnified view of a 10 × 10 superpixel area showing averaged CC FD% per superpixel. Middle row: Local RPE elevation analysis. (E) RPE elevation map. (F) RPE elevation map with overlying 50 × 50 grid for superpixel transformation. (G) Superpixels with average RPE elevation. (H) Magnified view of 10 × 10 superpixel area showing averaged RPE elevation per superpixel. RPE elevation values were cube-root transformed to calculate drusen volume. Bottom row: nGA tracing. (I) nGA tracing location (red outline) as registered from the visit where nGA developed. (J) nGA tracing with overlying 50 × 50 grid for superpixel transformation. (K) Magnified view of 10 × 10 superpixel area showing area of nGA tracing (red outline), with only superpixels that comprised ≥50% of A-scans with nGA label considered as having developed nGA.

Statistical Analysis

To account for intraindividual and within-eye correlations, linear mixed models were utilized to examine the differences between the CC FD% between eyes and superpixels that subsequently developed nGA compared to those that did not. All models adjusted for the time interval between the two study visits. For analyses at the eye level, linear mixed models with and without adjustments for cube-root transformed drusen volume and age at the first visit were used, with random intercepts included at the individual level. Statistical analyses at the local level adjusted for drusen volume and eccentricity from the foveal center at each superpixel and age at the first visit, with random intercepts included at the individual and eye levels. Eccentricity was included given prior evidence that FD% varies based on foveal proximity.³⁴ Drusen can lead to shadow artifacts that affect CC quantification.²⁶ Compensation algorithms reduce, but do not eliminate, the effect of drusen shadowing on the underlying choriocapillaris.³⁵ As such, adjustment for drusen volume was performed to further control for such potential drusen shadowing evading image compensation. More important, however, drusen have been independently associated with areas of CC blood flow impairment (as discussed further later in our Discussion section). Drusen compensation, therefore, aims to capture potential disease-related CC blood flow impairments that can be present in areas with drusen. 

To investigate whether CC FDs could predict nGA development at a local level, covariate-adjusted receiver operating characteristic (ROC) curves were developed. Three predictive models were evaluated, each including the eccentricity of the superpixel: (1) drusen volume only (as the reference model for comparisons), (2) CC FD% only, and (3) CC FD% and drusen volume combined. Note that drusen volume in these models was cube-root transformed. These models were developed using Cox proportional hazard models, using a leave-one-participant-out cross-validation procedure to generate the predicted probabilities for developing nGA at the superpixel level. The performances of these models were compared based on their full and partial area under the ROC curves (AUC), based on specificities ≥90% for the latter, with linear adjustment for the time interval between the two study visits as the covariate. A bootstrap resampling procedure (n = 1000 resamples at the participant level) was used to calculate the standard errors for hypothesis testing when comparing the performance of these three models. 

Results

A total of 105 eyes from 62 participants were included in this study, who were on average 72 ± 7 years old (range, 54 to 86 years old) and predominantly (85%) female. A total of 15 eyes from 12 participants developed nGA. A total of 18 nGA lesions were identified, and the median size of these lesions was 0.25 mm² (interquartile range [IQR] = 0.12 to 0.38 mm²). The median time interval between the visit assessed for the development of nGA and the prior visit was 6 months (IQR = 5 to 7 months). This time interval was not significantly different between the eyes that subsequently developed nGA and those that did not (P = 0.580). At the visit prior to assessment of nGA, the median drusen volume and FD% within the central 5-mm region of the eyes that developed nGA were 0.13 mm³ (0.07 to 0.26 mm³) and 2.3% (IQR = 1.0% to 3.3%), respectively, and 0.07 mm³ (0.03 to 0.11 mm³) and 1.6% (IQR = 1.0% to 3.0%) in eyes that did not develop nGA, respectively. 

Difference in CC FDs and nGA Development

At the eye level, there was no significant difference in CC FD% between eyes that subsequently developed nGA, both with and without adjusting for the potential confounders of drusen volume and age (P ≥ 0.399; Table 1). 

Table 1.

 

Associations With Choriocapillaris Flow Deficit Percentage in an Eye at the Visit Prior to Assessment of nGA Development

 

At the local level, superpixels that subsequently developed nGA had a significantly larger FD% compared to those that did not, both with and without adjusting for the confounders of average drusen volume within each superpixel and age (P < 0.001; Table 2). Note that at both the eye and local levels, the extent of drusen present and age were significant confounders of CC FD% measurements (P ≤ 0.007; Tables 1 and 2). 

Table 2.

 

Associations With Choriocapillaris Flow Deficit Percentage (%) in a 120 × 120-µm Region (or Superpixel) at the Visit Prior to Assessment of nGA Development

 

Supplementary analyses demonstrated that FD% at the second visit was significantly larger within superpixels where nGA was present compared to superpixels without nGA. Moreover, when assessing FD within nGA superpixels, FD% was significantly larger at the visit of nGA development than at the visit prior to nGA development (P < 0.001, Supplementary Table S1). Further analyses showed that FD% was significantly larger in the regions that subsequently developed nGA and neighboring foci one superpixel away from these regions (all P < 0.001, Supplementary Table S2). However, this significant association was not present in foci two and three superpixels away from the region of nGA development (P ≥ 0.066, Supplementary Table S2), further supporting the focal nature of these CC changes. 

Prediction of nGA Development Based on CC FDs

When considering both the AUC and partial AUC at specificities ≥90%, the performance of predicting nGA development at the superpixel level was significantly higher based on a model using CC FD% measurements alone (model 2), compared to a reference model using the drusen volume only (model 1; P ≤ 0.040), and with a model where both parameters were used (model 3; P = 0.029 for both). These findings are summarized in Table 3 and illustrated in Figure 4. 

Table 3.

 

Performance of Different Models for Predicting the Development of nGA in a 120 × 120-µm Region (or Superpixel)

 

Figure 4.

 

Performance of different models for predicting the development of nGA at the superpixel level (120 × 120 µm); FD% = percentage of choriocapillaris flow deficits. Note that drusen volume measurements were cube-root transformed.

Discussion

A key question to answer in the pathogenesis of AMD and its progression is whether blood supply reduction precedes photoreceptors and RPE death or if the opposite is true. To answer this, we evaluated whether there was a larger extent of CC FDs in eyes and superpixels that subsequently developed nGA, compared to those that did not develop nGA. We found that while there was no significant difference in the global extent of CC FDs between eyes that subsequently developed nGA and those that did not, superpixels that subsequently developed nGA had a significantly larger extent of CC FDs. Further, the use of the CC FD% parameter, or its addition to a measure of drusen volume, significantly improved the local prediction of nGA development compared to a measure of local drusen volume alone. Together, these findings provide further insight into mechanisms driving atrophy development in early AMD. 

Age, Drusen Volume, and CC FD

In this study, we observed that increasing age and drusen volume were significantly and independently associated with increasing CC FDs in both global and local analyses. Multiple studies have shown increased CC FDs in aging eyes of healthy participants and individuals with iAMD, supporting the association seen in our analysis and the need to control for age in our analyses.^34,36 

It has been well established that CC quantification is vulnerable to shadow artifacts caused by reduced signal penetration from overlying retinal vasculature, vitreous opacities, and sub-RPE accumulations.^26,37,38 Although this is especially true of spectral domain OCT (SD-OCT) instruments operating at 840-nm wavelengths, such artifacts are also present in SS-OCT instruments operating at 1050-nm wavelengths.^27,39 To mitigate the effect of drusen shadowing on CC OCTA quantification, we utilized established compensation algorithms that rely on structural signal data to account for shadowing.²⁹ After careful compensation, we still observed a strong positive association between CC FD% and drusen volume at a global and local level. Our findings are supported by findings from prior OCTA studies that reported a larger extent of CC FDs below drusen and regions that subsequently showed drusen growth or new incident drusen.^40,41 While it is challenging to establish a gold standard with which to validate OCTA compensation algorithms, some histologic evidence points to reduced CC vascular density below drusen.⁵ Thus, although compensation methods remain imperfect,³⁵ we believe that the association between CC FD% and drusen volume observed in our study is at least partially related to underlying histopathologic vascular changes and not due to shadowing artifact alone. 

Global CC FD and nGA Development

In this study, we found no association between global CC FD% and the subsequent development of nGA. These findings differ from recent studies,^14–16 which reported greater CC FDs in eyes that subsequently developed incomplete and complete RPE and outer retinal atrophy (iRORA and cRORA, respectively), as defined by the Classification of Atrophy Meeting Reports.^13,42 While nGA is defined by subsidence of the OPL and INL and/or a hyporeflective wedge-shaped band in Henle's fiber layer, iRORA and cRORA encompass a wider range of anatomic changes related to photoreceptors, including thinning of the outer nuclear layer, and disruption of the external limiting membrane or the ellipsoid zone.⁴² The definition of iRORA and cRORA requires such photoreceptor degeneration to be seen in association with the presence of choroidal signal hypertransmission and RPE attenuation or disruption, with cRORA requiring these latter two features to be ≥250 µm in extent. A broader definition of photoreceptor degeneration allows for a wider range of lesions qualifying as iRORA and cRORA when compared to nGA. Our recent study demonstrated that while the presence and development of iRORA was associated with an increased risk of GA development, this association was significantly weaker than the association between nGA and future GA development.⁴³ Furthermore, the risk of GA development associated with iRORA appears to be accounted for by the development of nGA in a multivariable model including both features. 

Thus far, most studies evaluating the association between baseline CC FD and future iRORA or cRORA development have relied solely on SD-OCT imaging.^14,15 Previous studies^14–16 that evaluated the association between CC FDs and future iRORA or cRORA development have also either not observed a significant association between the presence of late AMD in the fellow eye and/or drusen volume and late AMD development or did not adjust for these factors, even though they are well-established risk factors for disease progression.^44,45 The above differences in the use of SD-OCT rather than SS-OCT, characteristics of the cohort, and/or analyses performed may thus account for observed differences in global findings. 

Local CC FD and nGA Development

The current study locally analyzed the CC prior to nGA development. We observed that CC FD% was significantly greater within superpixels that subsequently developed nGA compared to those that did not. Histology studies show that CC dysfunction occurs with aging.⁸ Studies show that CC FD% varies based on foveal proximity, and thus topographic position should be considered in local CC analyses.³⁴ However, our findings held true even after controlling for key confounders like superpixel eccentricity, drusen volume, and age. Our data are somewhat supported by previous OCTA studies that observed an association between global CC FDs and progression to iRORA and cRORA.^14–16 However, our findings suggest that the observed CC loss is likely more localized in nature. It is possible that the local CC changes observed in this study are indicative of early focal CC dysfunction in areas susceptible to atrophy formation and that such CC changes exceed those that occur with aging. In all, our local analysis adds to a growing body of literature that supports the notion that CC loss occurs very early in atrophy formation. Future studies are now needed to examine the temporal changes in such localized CC FDs prior to nGA development to examine this notion further and understand the timeframe of CC loss, which could be achieved through evaluating longitudinal cohorts seen over multiple follow-up visits. 

CC FD as a Predictor of nGA Development

Given the association between local CC FD% and subsequent nGA development seen at the cohort level, we examined whether CC FD% measures could be used to more accurately predict nGA development compared to, or in addition to, a measure of local drusen volume. We chose drusen volume as a comparative measure because it is a robust and reproducible variable associated with AMD progression.^44,45 Our analyses revealed that, at the local level, the CC FD% parameter performs significantly better when compared to drusen volume for nGA prediction and that the use of both parameters further improves nGA development prediction. While drusen volume remains a more practical measure to obtain due to its incorporation into widely available OCT devices, our observations provide crucial insights into the role of CC impairments in the pathogenesis of atrophic AMD. 

CC FD in AMD Pathophysiology

The CC is required for bringing nutrients and oxygen to the outer retina. The current vascular model for atrophy development in AMD suggests CC loss compromises metabolic exchange in the outer retina, leading to subsequent sub-RPE material accumulation, downstream drusen formation, photoreceptor degeneration, and atrophy development.^7,46,47 In accordance with this theory, data are accumulating that show new vessel formation from choroidal neovascularization in late AMD is protective against GA growth, further supporting vascular compromise as the driver for atrophy development.⁴⁸ Similarly, recent studies focusing on proangiogenic stimulation in AMD animal models have found that activation of Tie2, a regulator of vascular maturation and homeostasis, promotes CC regeneration and reduces hypoxia in neovascular AMD models.⁴⁹ Li and colleagues⁵⁰ showed that leukemia inhibitory factor, a mitogen for choroidal endothelial cells, reduces CC FDs and protects against retinal atrophy in a GA mouse model. Such findings support CC loss as a potential driver of GA formation in AMD and open new avenues for therapeutic development. Our results show that localized CC FDs precede atrophy detection and support the notion that early interventions preventing or slowing CC loss could also prevent or delay GA onset. 

Strengths and Limitations

The strengths of this study include use of SS-OCTA for improved image penetration below the RPE and CC visualization, as well as adjustment for superpixel eccentricity and RPE volume in the local analyses. Limitations include the limited number of eyes examined (n = 105) and of eyes that developed nGA (n = 15). For instruments using unnormalized OCTA algorithms, including the PLEX Elite 9000, a compensation step is necessary to reduce the dependency of the OCTA signal on the intensity of the backscattered/reflected OCT signal, which can be variably attenuated by factors including drusen and opacities.⁵¹ The compensation and segmentation algorithms selected were specifically designed for the output of the PLEX Elite 9000 device and are updated regularly by the device manufacturer. The selected compensation method utilizes volumetric structural data to compensate for signal attenuation from overlying pathologic changes, boosting the OCTA signal in areas of reduced OCT signal. Such a compensation approach can therefore lead to the unwanted disappearance and underdetection of CC FDs but also minimizes the likelihood of falsely identifying CC FDs in areas of low OCT signal.³⁵ Although our eligibility criteria permitted the inclusion of scans with a signal strength ≥7, note that 97% and 85% of included scans had a signal strength quality score of ≥8 and ≥9, respectively. Robust signal strength further improves the quality of our compensation outcomes. As stated earlier, compensation methodologies reduce, but do not eliminate, shadowing from overlying pathologic features, including calcified drusen and hyperreflective foci. This is evident in Figure 1, where there is greater CC loss at the n–1 visit when compared to the n-visit. As it is highly improbable that CC flow recovered over time, these discrepancies may be caused by changes in RPE morphology that evaded compensation. For instance, the collapse of a pigment epithelial detachment (PED) or druse will lead to reduced shadowing at the follow-up visit and the false appearance of regained CC flow. Although newer compensation strategies are under way, the current investigation is limited to available methods and did not account for such pitfalls. However, our statistical analyses controlled for RPE elevation to further mitigate the effect of RPE changes evading compensation. Lastly, several methodologies exist for the computation and analysis of CC FD. Studies have shown variation in binarization techniques, and threshold selection can greatly affect CC FD quantification.³⁰ The current study utilized a fixed thresholding/binarization method and derived the selected threshold from a normative database. This thresholding strategy has been shown to yield repeatable CC flow deficits across acquisitions, even in cases where the A-scan density changes.^29,52 The current study focused on relative changes in CC impairment. As such, the ability to produce consistent CC FD measures across participants and time intervals was prioritized. Nevertheless, since the threshold value was not derived from a widely available normative database, the ability to compare absolute CC FD measures between this study and others remains limited. Future work aimed at standardizing threshold and binarization methodologies in the field of OCTA research is needed to enable the comparison of CC FD measures across studies. 

Conclusions

This study demonstrates a significantly larger extent of CC FD% exist within superpixels that subsequently developed nGA compared to those that do not. Our findings add to a body of evidence suggesting vascular impairment in the CC is a possible driver of atrophy formation in AMD. 

Acknowledgments

Supported by National Health & Medical Research Council of Australia (Grants GNT1194667 [RHG] and #2008382 [ZW]), National Institutes of Health (Grant R01-EY011289-35 [JF]), the Retina Research Foundation (JF), the Beckman-Argyros Award in Vision Research (JF), the Champalimaud Vision Award (JF), Massachusetts Lions Club Grant (NW), Research to Prevent Blindness Challenge Grant (NW), and grants from the Macular Disease Foundation Australia (ZW and RHG). CERA receives operational infrastructure support from the Victorian Government. The sponsor or funding organizations had no role in the design or conduct of this research. 

Disclosure: E.C. Greig, None; E.M. Moult, None; I.N. Despotovic, None; L.A.B. Hodgson, None; V. Pramil, None; J.G. Fujimoto, Topcon (F), Optovue (F); N.K. Waheed, Nidek (C), Topcon (C), Ocudyne (C), Stealth (C), Novartis (C), Iolyx (C), Complement Therapeutics (C); R.H. Guymer, Bayer (C, F), Roche/Genentech (C, F), Apellis (C, F), Novartis (C, F); Z. Wu, None