Geographic Atrophy on Fundus Autofluorescence and Color Fundus Photographs: Association with Deep Visual Sensitivity Losses

Abstract

Purpose

To determine the functional characteristics of color fundus photograph (CFP)- and fundus autofluorescence (FAF)-defined geographic atrophy (GA) lesions by evaluating the prevalence of repeatable deep visual sensitivity defects.

Design

Reader study.

Participants

One hundred seventy-one pairs of CFP and FAF images from 60 eyes of 53 individuals.

Methods

High-density, targeted microperimetry testing (with Goldmann Size III stimuli) was performed twice per visit in a 3.5° (approximately 1000 μm) diameter region of interest with retinal pigment epithelium and outer retinal atrophy on OCT imaging. Twelve readers from 6 established reading centers assessed CFP and FAF images within these regions sampled on microperimetry for GA, and performed annotations where GA was deemed to be present. Geographic atrophy on CFP was defined as a well-demarcated, roughly round or oval region of hypopigmentation, separately with and without requiring increased visibility of the underlying choroidal vessels (referred to as CFP-defined GA₁ and GA₂, respectively). GA on FAF was defined as a region of definite decreased autofluorescence.

Main Outcome Measures

Prevalence of a repeatable ≤10 dB defect for CFP- and FAF-defined GA ≥175 μm, and the minimum lesion size showing a ≥90% prevalence of a repeatable ≤10 dB defect (deemed characteristic of regions with a truly nonresponding test location on microperimetry).

Results

Color fundus photograph-defined GA₁ and GA₂ and FAF-defined GA ≥175 μm were graded as present in 13%, 31%, and 41% of images, respectively, and 77%, 67%, and 62% lesions, respectively, had a repeatable ≤10 dB defect. Only CFP-defined GA₁ ≥625 μm, CFP-defined GA₂ ≥650 μm, and FAF-defined ≥675 μm had a ≥90% prevalence of a repeatable ≤10 dB defect.

Conclusions

Color fundus photograph and FAF-defined GA lesions ≥175 μm do not show the same functional characteristics as regions with a truly nonresponding test location, and only much larger lesions (approximately ≥650 μm) showed such similar functional characteristics. These findings provide crucial insights when considering CFP- and FAF-defined atrophic endpoints for clinical studies.

Financial Disclosure(s)

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Geographic atrophy (GA) is a late-stage manifestation of age-related macular degeneration (AMD) that is characterized by the progressive loss of the retinal pigment epithelium (RPE), with accompanying overlying photoreceptor degeneration and underlying choriocapillaris loss.1, 2, 3, 4, 5 Geographic atrophy has historically been defined on color fundus photographs (CFPs) or clinical examination as a “sharply delineated roughly round or oval area of hypopigmentation or depigmentation or apparent absence of the RPE in which choroidal vessels are more visible than in surrounding areas that must be at least 175 μm in diameter.”⁶ More recently, fundus autofluorescence (FAF) imaging⁷ has become increasingly used to detect the presence of GA and to quantify its extent. On FAF imaging, GA is defined as a region of definite decreased autofluorescence, representing RPE loss.⁸ The onset of CFP- and FAF-defined GA has been used as a clinical endpoint in early intervention trials9, 10, 11, 12 and observational studies.13, 14, 15, 16, 17

It has been assumed that regions of GA are nonseeing, given that overlying photoreceptor degeneration is expected to have occurred in such regions with RPE loss.^18,19 However, we recently found that only 82% and 60% of regions with CFP- and FAF-defined GA, respectively, returned a repeatable deep visual sensitivity defect (with a threshold of ≤10 dB)²⁰ on high-density, targeted microperimetry testing.21, 22, 23 In contrast, we observed from a recent computer simulation study of microperimetry test results of eyes of individuals with bilateral large drusen that >90% of regions simulated to have a truly nonresponding test location would show a repeatable deep visual sensitivity defect.²⁴ Furthermore, our recent Delphi study reported consensus that the onset of end-stage atrophic AMD is an appropriate clinical endpoint to evaluate early interventions and that such an endpoint should be defined by anatomical changes that are associated with marked, but not necessarily complete, functional loss.²⁵ The highest level of agreement was observed for the criterion that ≥90% of such end-stage atrophic AMD lesions should have a repeatable deep visual sensitivity defect.²⁵ These recent findings thus suggest that incident CFP- and FAF-defined GA may not have the same functional characteristics as a region with a truly nonresponding location, nor at a level of functional loss expected for an end-stage atrophic AMD clinical endpoint.

This study, involving 6 reading centers, thus sought to independently determine the functional characteristics of CFP- and FAF-defined GA lesions (as currently defined and used in early intervention trials and observational studies) by evaluating the prevalence of repeatable deep visual sensitivity defects. It also sought to identify the minimum diameter of such lesions that would show the same functional characteristics as a region with a truly nonresponding location.²⁴ This information can be used to inform future clinical trials and studies about the minimum dimension needed for such GA lesions that would show functional characteristics expected of a clinical endpoint representing end-stage atrophic AMD.²⁵ This study also sought to determine interreader agreement for assessing such CFP- and FAF-defined GA lesions, to understand their robustness as a clinical endpoint.

Methods

This study was a multicenter reader study that sought to understand the functional characteristics of CFP- and FAF-defined GA lesions and the minimum size of such lesions that are associated with repeatable deep visual sensitivity defects, and to evaluate interreader agreement. This study was undertaken using a previously collected clinical dataset, as described further here.

Grading Dataset

The grading dataset consisted of 171 pairs of CFP and FAF images obtained from 60 eyes of 53 participants who were enrolled in an observational study of early atrophic AMD changes.²² That study was undertaken at the Centre for Eye Research Australia and was conducted in adherence with the tenets of the Declaration of Helsinki and International Conference on Harmonization Guidelines for Good Clinical Practice. Institutional review board approval was obtained for this study from the Royal Victorian Eye and Ear Hospital, and all participants in this study provided informed consent.

Participants

Participants in the aforementioned observational study that was used to provide the grading dataset in this study were required to have ≥1 eye with ≥1 RPE and outer retinal atrophy lesion on an OCT volume scan,26, 27, 28 without evidence of neovascular AMD. These participants were seen at ≥1 visits, as described previously in detail,²² and underwent multimodal retinal imaging and microperimetry at each visit.

Retinal Imaging

Nonstereoscopic CFPs centered on the macula with a 45° diameter field-of-view were obtained using the Canon CR6-45NM device (Canon). Fundus autofluorescence images, also centered on the macula and covering a 30° × 30° region with 100 frames averaged, were acquired using the Heidelberg Spectralis HRA + OCT (Heidelberg Engineering) device. Note that FAF images with this device were obtained using a 488 nm excitation light source, and it captured emitted fluorescence signals between 500 and 700 nm. OCT volume scans also centered on the macula, sampling the central 20° × 20° region and consisting of 97 B-scans (each with 1024 A-scans and 16 frames averaged), were also acquired using the same device.

Microperimetry Testing

Participants underwent high-density, targeted microperimetry testing using the Macular Integrity Assessment device (CenterVue) on the same day as multimodal imaging. This testing was performed when sampling an atrophic lesion of interest, using an isotropic stimulus pattern consisting of 37 test locations that covered a 3.5° (or approximately 1000 μm) diameter region (Fig 1).21, 22, 23 The atrophic lesion of interest was required to be smaller than the stimulus pattern (approximately ≤750 μm in diameter) so that the stimulus pattern also sampled nonatrophic regions. If >1 atrophic lesion was present in an eye, only lesion(s) isolated from other lesions were considered eligible for testing (to avoid sampling multiple lesions with the stimulus pattern).

Figure 1.

Example of the (A) CFP and (B) FAF images that were evaluated in this study. The region evaluated for the grading task in this study (approximately 1000 μm in diameter) is outlined by a green circle on each image. This region corresponded to the location where high-density, targeted microperimetry testing was performed, as shown in (C). Note that the CFP and FAF images were assessed separately by each reader, who was also masked to the findings from microperimetry testing. CFP = color fundus photograph; FAF = fundus autofluorescence.

At the first visit, ≥3 high-density, targeted microperimetry tests were performed on each eye. At subsequent visits, ≥2 tests were performed on each eye. The “follow-up” function was used for all tests after the first test during the first visit. Note that the first test ever performed was excluded from the analyses in this study to minimize systematic underestimation of visual sensitivity loss and potential learning effects as observed previously for such a first ever test.^29,30 Furthermore, only study visits with 2 reliable tests—defined based on tests with ≤25% false-positive errors—were included in this study to determine the presence of a repeatable deep (≤10 dB) visual sensitivity defect.

Grading Task

The grading task was completed by the following 6 established reading centers: the Cologne Imaging Reading Center and Laboratory, Doheny Image Reading and Research Laboratory, Duke Reading Center, GRADE Reading Center, Utah Retinal Reading Center, and Wisconsin Reading Center. Two experienced readers from each reading center were selected by the medical director to participate in this reader study. The grading task first involved determining if GA was present in the region sampled on microperimetry (see further in this section), and then secondly, if present, the GA regions were annotated. Geographic atrophy was defined on CFP as a sharply delineated roughly round or oval area of hypopigmentation or depigmentation, and separately with and without requiring this area to have choroidal vessels that are definitely more visible than surrounding areas (termed CFP-defined GA₁ and GA₂, respectively, in this study).^6,31, 32, 33 On FAF, GA was defined as a region of definite decreased autofluorescence.⁸ Geographic atrophy on both imaging modalities needed to be ≥175 μm in diameter and were annotated with custom-written software (Cross-Modality Annotation Software).^34,35

The aforementioned grading dataset was randomly partitioned into 2 sets, such that each reader evaluated one set of CFPs and a different set of FAF images. Each CFP and FAF image was thus assessed by 6 readers, one from each reading center (rather than both readers at each reading center, to ensure that this grading task could be feasibly completed). Assessments of these images were limited to only the approximately 1000-μm diameter region sampled by microperimetry, which was determined through manual coregistration of both the CFP and FAF images with the near-infrared reflectance image from the microperimeter. To coregister the CFP and FAF images with the near-infrared reflectance image from the microperimeter, fiducial points at retinal bifurcations were identified on each image, and then radial transformations were applied, as described in detail previously.^23,35 Figure 1 shows an example of the CFP and FAF images evaluated and shows the specific region that was assessed (based on where high-density targeted microperimetry testing was performed).

Data and Statistical Analyses

The prevalence of CFP- and FAF-defined GA lesions based on different cutoffs (≥175, ≥250, and ≥500 μm) and the prevalence of a repeatable ≤10 dB visual sensitivity defect on high-density targeted microperimetry were then determined. These estimates were derived using a mixed-effects logistic regression model, including random effects at the level of the reading center, individual, and eye. These random effects were included to account for correlations between assessments from the same reading center and for correlations between 2 eyes for an individual and multiple visits per eye. Interreader agreement for the presence of CFP and FAF-defined GA lesions based on the different cutoffs above were also evaluated with Gwet first-order agreement coefficient (AC₁)—a chance-corrected measure of agreement (with values ranging from –1 to 1).³⁶ The AC₁ remains robust across variations in the prevalence of the trait being evaluated, unlike the more widely evaluated κ coefficient.³⁷ Note that the level of interreader agreement for the AC₁ values were described as follows: poor agreement (<0.00), slight agreement (0.00–0.20), fair agreement (0.21–0.40), moderate agreement (0.41–0.60), substantial agreement (0.61–0.80), and near-perfect agreement (>0.80).³⁸

In addition, our recent study using computer simulation of microperimetry test results of eyes of individuals with bilateral large drusen showed that 91% of regions that were simulated to have a truly nonresponding test location would have a repeatable ≤10 dB defect.²⁴ Note that truly nonresponding test locations in this recent study was derived based on responses to stimuli presented at the optic nerve head of healthy individuals.²⁴ Our recent Delphi study also reported that the highest level of agreement within an expert group was achieved for the specific criterion that ≥90% of end-stage atrophic AMD should show a repeatable ≤10 dB defect. We thus sought to identify the minimum CFP and FAF-defined GA lesion size (in 25 μm intervals, between ≥175 and ≥750 μm) that would have a ≥90% prevalence of such a repeatable ≤10 dB defect on high-density targeted microperimetry. Lesions of such a size would thus represent those exhibiting functional characteristics comparable with a region with a truly nonresponding test location, which would be expected of end-stage atrophic AMD lesions. We also derived the prevalence and interreader agreement for such lesions.

Finally, we determined the between-modality and between-definition interreader agreement for the CFP and FAF-defined GA lesions (i.e., agreement between CFP-defined GA₁ and GA₂ with FAF-defined GA and between CFP-defined GA₁ and GA₂). For this assessment, we evaluated the agreement between GA as determined on one modality or definition by a reader from one site with GA as determined on another modality definition by a reader from all the other sites. All analyses in this study were performed using Stata/BE (software version 18; StataCorp).

Results

This study included 171 pairs of CFPs and FAFs from 60 eyes of 53 individuals. These individuals were on average 75 ± 8 years old (range, 53–86 years old), and there were 39 (74%) individuals who were female. The image pairs were acquired over a median of 2 visits (interquartile range = 1–5 visits).

Prevalence of Repeatable Deep Visual Sensitivity Defects in Different GA Lesions

Color fundus photograph-defined GA₁ and GA₂, and FAF-defined GA, ≥175 μm in diameter were present in 13% (95% confidence interval [CI] = 7–19%), 31% (95% CI = 21–40%), and 41% (95% CI = 37–46%) of the images, respectively (Table 1). There was substantial interreader agreement for CFP-defined GA₁ (Gwet AC₁ = 0.74 [95% CI = 0.68–0.80]), but only moderate agreement for CFP-defined GA₂ and FAF-defined GA (Gwet AC₁ = 0.41 [95% CI = 0.33–0.49] and 0.52 [95% CI = 0.44–0.59], respectively; Table 1). Table 1 shows how only 77% (95% CI = 64–90%), 67% (95% CI = 62–71%) and 62% (95% CI = 58–66%) of CFP-defined GA₁ and GA₂ and FAF-defined GA, respectively, that were ≥175 μm in diameter had a repeatable ≤10 dB defect. These were all lower than the approximately 90% expected prevalence of such a defect for lesions with a truly nonresponding test location.²⁴

Table 1.

Findings from the Grading by 12 Readers across 6 Reading Centers for GA Lesions on CFPs and FAF for Different-Sized Cutoffs

	Prevalence (%) of		Gwet AC1
	Lesion	≤10 dB Defect∗
CFP-defined GA1 (requiring vessel visible at base)
≥175 μm	13 (7–19)	77 (64–90)	0.74 (0.68–0.80)
≥250 μm	13 (7–19)	77 (64–90)	0.74 (0.68–0.80)
≥500 μm	8 (4–12)	82 (68–96)	0.85 (0.80–0.89)
Minimum diameter‡: ≥625 μm	4 (2–6)	92 (79–100)	0.93 (0.90–0.96)
CFP-defined GA2 (without requiring vessel visible)
≥175 μm	31 (21–40)	67 (62–71)	0.41 (0.33–0.49)
≥250 μm	30 (20–40)	66 (62–71)	0.42 (0.34–0.51)
≥500 μm	14 (9–19)	77 (70–84)	0.78 (0.72–0.84)
Minimum diameter‡: ≥650 μm	7 (4–9)	90 (77–100)	0.91 (0.88–0.95)
FAF-defined GA
≥175 μm	41 (37–46)	62 (58–66)	0.52 (0.44–0.59)
≥250 μm	41 (37–45)	62 (59–66)	0.52 (0.45–0.60)
≥500 μm	15 (13–17)	79 (68–90)	0.82 (0.76–0.87)
Minimum diameter‡: ≥675 μm	7 (4–9)	95 (87–100)	0.95 (0.92–0.97)

AC₁ = first-order agreement coefficient; CFP = color fundus photograph; FAF = fundus autofluorescence; GA = geographic atrophy.

^∗A test with ≥1 location that is ≤10 dB on 2 microperimetry tests.

^‡Minimum lesion diameter (in 25 μm increments) where ≥90% of lesions had a repeatable ≤10 dB defect on microperimetry. The 95% confidence intervals are presented in parentheses for all the values shown.

Instead, only CFP-defined GA₁ and GA₂ and FAF-defined GA lesions that were ≥625 μm, ≥650 μm, and ≥675 μm, respectively, had a ≥90% prevalence of a repeatable ≤10 dB defect. These lesions were present in 4% (95% CI = 2–6%), 7% (95% CI = 4–9%), and 7% (95% CI = 4–9%) of eyes, respectively, and there was near-perfect interreader agreement for assessing such lesions (Gwet AC₁ = 0.93, 0.91, and 0.95, respectively; Table 1).

Between-Modality and Between-Definition Agreement for GA Lesions

Table 2 shows how there was only slight interreader, between-modality, and between-definition agreement between CFP-defined GA₁ and GA₂ with FAF-defined GA (Gwet AC₁ = 0.16 [95% CI = 0.13–0.19] and 0.03 [95% CI = 0.00–0.06], respectively), and moderate agreement between CFP-defined GA₁ and GA₂ (Gwet AC₁ = 0.49 [95% CI = 0.47–0.52]). However, when considering GA lesions that had a ≥90% prevalence of a repeatable ≤10 dB defect described above, there was near-perfect interreader agreement for all pairwise comparisons between CFP-defined GA₁ and GA₂ and FAF-defined GA (Gwet AC₁ = 0.91–0.92; Table 2).

Table 2.

Gwet First Order Coefficient (AC₁) for Between-Modality or Between-Definition and Interreader Agreement for GA

Size Cutoff	CFP-Defined GA1 vs. FAF-Defined GA	CFP-Defined GA2 vs. FAF-Defined GA	CFP-Defined GA1 vs. CFP-Defined GA2
≥175 μm	0.16 (0.13–0.19)	0.03 (0.00–0.06)	0.49 (0.47–0.52)
≥250 μm	0.18 (0.15–0.21)	0.06 (0.03–0.08)	0.50 (0.48–0.53)
≥500 μm	0.77 (0.75–0.79)	0.71 (0.69–0.73)	0.79 (0.78–0.81)
Minimum diameter for lesion with repeatable defect∗	0.91 (0.90–0.92)	0.91 (0.91–0.92)	0.92 (0.92–0.93)

AC₁ = first-order agreement coefficient; CFP = color fundus photograph; FAF = fundus autofluorescence; GA = geographic atrophy; CFP-Defined GA₁ = color fundus photograph-defined geographic atrophy when requiring choroidal vessels to be visible than surrounding areas; CFP-Defined GA₂ = color fundus photograph-defined geographic atrophy without requiring choroidal vessels to be visible than surrounding areas.

^∗Minimum lesion diameter (in 25 μm increments) where ≥90% of lesions had a repeatable ≤10 dB defect on high-density, targeted microperimetry testing.

Examples of Findings in This Study

Three examples of regions showing a repeatable ≤10 dB defect on high-density, targeted microperimetry testing are shown in Figure 2. In case 1, all readers identified CFP-defined GA₁ and GA₂ and FAF-defined GA that was ≥500 μm. In case 2, all readers found FAF-defined GA that was ≥175 μm, but only ≤1 reader identified CFP-defined GA₁ and GA₂ that was ≥175 μm. In case 3, ≥4 readers observed CFP-defined GA₁ and GA₂ that was ≥175 μm, but none of the readers identified FAF-defined GA that was ≥175 μm.

Figure 2.

Three examples of lesions showing a repeatable ≤10 dB defect on high-density, targeted microperimetry testing, to illustrate cases where GA was or was not detected based on different imaging modalities or different definitions. The number of readers who identified GA on CFP and FAF is shown. Note that the CFP and FAF images were assessed separately by each reader, who was also masked to the findings from microperimetry testing. CFP = color fundus photograph; FAF = fundus autofluorescence; GA = geographic atrophy.

Two examples of regions without a repeatable ≤10 dB defect are shown in Figure 3. In case 4, there were ≥4 readers who observed FAF-defined GA and CFP-defined GA₂ that was ≥175 μm, and only 1 reader found CFP-defined GA₁ that was ≥175 μm. In case 5, all readers identified FAF-defined GA that was ≥175 μm, but ≤2 readers observed CFP-defined GA₁ and GA₂ that was ≥175 μm.

Figure 3.

Two examples of lesions without a repeatable ≤10 dB defect on high-density, targeted microperimetry testing, to illustrate cases where GA was or was not detected based on different imaging modalities or different definitions. The number of readers who identified GA on CFP and FAF is shown. Note that the CFP and FAF images were assessed separately by each reader, who was also masked to the findings from microperimetry testing. CFP = color fundus photograph; FAF = fundus autofluorescence; GA = geographic atrophy.

Discussion

This study revealed that CFP and FAF-defined GA ≥175 μm in diameter do not show the same functional characteristics as a region with a truly nonresponding location, nor what one might expect of end-stage atrophic AMD lesions. It also showed that CFP-defined GA that required increased visibility of the underlying choroidal vessels (CFP-defined GA₁) showed a greater level of functional loss and better interreader agreement than lesions such increased visibility of the underlying choroidal vessels was not required (CFP-defined GA₂) and FAF-defined GA. These findings provide important insights into the functional characteristics of these lesions, especially when considering potential clinical endpoints in future early intervention trials.

This study builds upon recently reported preliminary findings based on grading performed by the site that conducted the clinical study used to derive the grading dataset evaluated in this study.²⁰ This recent study found that only 82% and 60% of regions with CFP and FAF-defined GA ≥175 μm, respectively, showed a repeatable deep visual sensitivity defect on high-density, targeted microperimetry testing in the dataset evaluated in this study. These findings were based on assessments performed by readers at the center that led the prospective clinical study from which the grading dataset in this study was derived, and only CFP-defined GA₁ was evaluated. In this current study, we similarly found that 77% and 62% of regions with CFP-defined GA₁ and FAF-defined GA ≥175 μm, respectively, had a repeatable deep visual sensitivity defect, based on assessments performed by 12 independent, masked readers from 6 reading centers. We also found that 67% of regions with CFP-defined GA₂ ≥175 μm had such a repeatable ≤10 dB defect, which was lower than observed for CFP-defined GA₁. The prevalence of CFP-defined GA₁ was lower than CFP-defined GA₂ and FAF-defined GA ≥175 μm in diameter (13% vs 31% and 41%, respectively), but there was higher interreader agreement for these lesions (Gwet AC₁ = 0.74 vs 0.41 and 0.52, respectively). However, there was only slight between-modality, interreader agreement between CFP-defined GA₁ and GA₂ with FAF-defined GA ≥175 μm (Gwet AC₁ = 0.16 and 0.03, respectively), which were much lower than the within-modality and within-definition, interreader agreement for these lesions.

Our recent computer simulation study showed that >90% of regions with a truly nonresponding test location would have a repeatable ≤10 dB defect,²⁴ and our recent Delphi study also reported that the highest level of agreement was observed for the specific criteria that ≥90% of end-stage atrophic lesions should have a repeatable ≤10 dB defect.²⁵ However, the prevalence of such a repeatable ≤10 dB defect was only 62%-77% when considering all regions with CFP and FAF-defined GA lesions meeting the criteria of being ≥175 μm. It is only when considering all regions with GA lesions meeting the criteria of being ≥625 to ≥675 μm that a ≥90% prevalence for a repeatable ≤10 dB defect was seen, and the prevalence of these lesions was more consistent between the different modalities and definitions (4%, 7%, and 7% for CFP-defined GA₁ and GA₂ and FAF-defined GA, respectively). There was near-perfect within-modality and within-definition, interreader agreement for the presence of such lesions with a ≥90% prevalence for a repeatable ≤10 dB defect (Gwet AC₁ = 0.91–0.95), and also near-perfect between-modality agreement (Gwet AC₁ = 0.91).

The aforementioned findings that a ≥90% prevalence of a repeatable ≤10 dB defect (expected for regions with a truly nonresponding location and expected of end-stage atrophic AMD lesions) was not seen when considering all regions with CFP and FAF-defined GA meeting the criteria of being ≥175 μm in diameter is perhaps not unexpected given previous findings from eyes with more extensive GA.^35,39 Specifically, a study by Pfau et al³⁹ reported that microperimetry test locations that fell approximately 187 μm inside the FAF-defined GA margin had higher visual sensitivities than expected for an absolute scotoma even after accounting for false-positive response rates. Higher visual sensitivities were also seen in regions with residual Henle fiber layer and outer nuclear layer than those without. Similarly, a previous study by Wu et al³⁵ reported that for microperimetry test locations that were entirely within 0 to 250 and >250 μm inside the FAF-defined GA margin, only 64% and 88% of locations showed nonresponse to a 10 dB stimuli. It is thus unsurprising that FAF-defined GA lesions ≤350-500 μm in diameter (which would have ≥1 test location within ≤187 μm³⁹ and ≤250 μm³⁵ of GA margin) would show such residual sensitivity based on these previous findings. These observations are also supported by previous histopathological findings that surviving photoreceptor cell bodies within gliotic tissue can be found in atrophic regions.1, 2, 3 However, these observations could also be explained by potential sampling errors with clinical microperimetry, occurring with fixation shifts not adequately compensated for with fundus tracking at 25 frames per second for the device used in these studies.

The findings in this study that there was a higher prevalence of FAF-defined GA than CFP-defined GA₁ but less severe functional loss for the former potentially indicate that FAF-defined GA may be detected earlier than CFP-defined GA. This notion is supported by findings by Holmen et al,⁴⁰ who reported that only approximately half of those with incident FAF-defined GA had CFP-defined GA₂ (both based on a lesion that was ≥430 μm in diameter) detected at the same visit in the Age-Related Eye Disease Study 2 FAF Ancillary Study. Domalpally et al⁴¹ also reported that among all images taken over a 5-year follow-up period of those included in the Age-Related Eye Disease Study 2 FAF Ancillary Study, GA ≥430 μm was detected in 25% and 20% of FAF images and CFPs, respectively. However, 23% of visits with FAF-defined GA did not have CFP-defined GA₂, while 7% of visits with CFP-defined GA₂ did not have FAF-defined GA. The authors thus suggested that areas of “hypoautofluorescence emerge earlier than GA on colors,” but “hypoautofluorescence is not specific for GA… [as drusen] and depigmentation also may present with changes in FAF images.”⁴¹ This may in part explain why CFP-defined GA₂ was also more prevalent and showed less severe functional loss than CFP-defined GA₁ in this study, as the former could more likely represent areas of only partial depigmentation without complete RPE loss.

Nonetheless, the findings of this study provide important insights into the functional characteristics of the clinical endpoint of incident CFP and FAF-defined GA as previously used in early intervention trials9, 10, 11, 12 and observational studies.13, 14, 15, 16, 17 It showed that such GA lesions ≥175 μm do not show the same functional characteristics as a region with a truly nonresponding location,²⁴ nor at a level of functional loss expected for an end-stage atrophic AMD clinical endpoint based on our recent Delphi study.²⁵ Indeed, there was only minimal intermodality agreement between such smaller GA lesions, suggesting that they may not represent definitive atrophic changes seen consistently across multiple modalities. Instead, only CFP and FAF-defined GA lesions approximately ≥650 μm met the expected level of functional loss for end-stage atrophic AMD from our recent Delphi study,²⁵ and we also observed near-perfect intermodality agreement for such larger and more functionally impaired lesions (Gwet AC₁ = 0.91). However, future longitudinal studies could seek to identify robust earlier surrogates, such as those detected on OCT26, 27, 28^,42, 43, 44, 45 that are most closely associated with the subsequent onset of such larger and more functionally impaired end-stage atrophic AMD lesions defined on CFP and FAF imaging. Such earlier surrogates, and also such end-stage atrophic AMD lesions, may be especially useful as endpoints in future earlier-phase trials to assess the efficacy and safety of preventative interventions in the early stages of AMD.

Limitations of this study include the size of the grading dataset evaluated (171 CFP and FAF image pairs from only 60 eyes of 53 individuals), collected from a single site. Future studies including a larger and more diverse dataset are needed to understand if the findings of this study are generalizable. However, a strength was the evaluation of this dataset by 12 readers from 6 established reading centers. Another limitation of this study was that CFPs and FAF images were evaluated independently, and as such, the effectiveness of a multimodal imaging approach (which could also additionally include OCT imaging) was therefore not assessed. The concurrent evaluation of both CFPs and FAFs as part of a multimodal imaging approach has been performed in some previous studies.46, 47, 48 The use of near-infrared reflectance (NIR) imaging, or both NIR and OCT imaging, to supplement assessment of FAF-defined GA, has also been performed in other studies.^49,50 However, these approaches have not historically been as widely performed as assessing CFPs or FAF images alone for GA in early intervention trials9, 10, 11, 12 and observational studies13, 14, 15, 16, 17 to date, but it is becoming increasingly more common. Finally, sampling errors that could occur with clinical microperimetry—which can occur due to fundus tracking at only 25 frames per second with the device used—is a limitation when characterizing visual sensitivity loss in this study. Future studies using smaller stimuli or integrating adaptive optics during microperimetry testing^51,52 may provide further insights into the functional characteristics of atrophic AMD.

In conclusion, this study demonstrated that CFP- and FAF-defined GA ≥175 μm do not show the same functional characteristics as regions with a truly nonresponding location or as expected of end-stage atrophic AMD lesions. However, only lesions that are approximately ≥650 μm show such a comparable level of functional loss. These findings provide important insights for the consideration of CFP- and FAF-defined atrophic endpoints for clinical studies.

Manuscript no. XOPS-D-25-00632.