Long-Term Natural History of Treatment-Naïve Geographic Atrophy in Age-Related Macular Degeneration

Abstract

Purpose: To evaluate the long-term fundus autofluorescence–based growth rate (GR) of treatment-naïve patients with geographic atrophy (GA) in age-related macular degeneration.

Methods: We conducted a prospective, single-center, observational study between February 2013 and September 2024 at the Department of Clinical Neuroscience, Karolinska Institutet, and St. Erik Eye Hospital, Stockholm/Solna, Sweden. Clinical examination and fundus autofluorescence were performed in patients with GA owing to dry age-related macular degeneration. The area and the absolute and square root transformed GR were analyzed every 6 months.

Results: We examined 432 eyes and enrolled 204 eyes (111patients). The median follow-up was 21 months (minimum–maximum, 5–123). Of 73 fovea-sparing, 22 eyes converted to foveal-involving over a median of 24 months. The mean growth for the total cohort was 1.597 mm²/y and 0.264 mm/y after square root transformation. Bilateral (1.621 mm²/y; 0.267 mm/y), multifocal (1.961 mm²/y; 0.322 mm/y), and fovea-sparing (1.987 mm²/y; 0.234 mm/y) lesions showed significantly faster growth when analyzed in isolation. In a mixed statistical model that controlled for bilaterality, only fovea status remained a significant influencer on the square root transformed GR (P < 0.001).

Conclusions: In this long-term GA cohort, an influence of lesion characteristics on GRs can be observed. Fovea sparing, multifocality, and bilaterality showed faster growth, depending on the statistical model. Patients presenting with one or more of these lesion characteristics hold a high potential for benefit of future treatments because a growth slow down may be more likely to be achieved. In fovea-sparing cases, functional preservation may be possible.

Translational Relevance: By analyzing the data of one of the most extensive geographic atrophy patient cohorts in the Nordics, this study establishes a dataset on the long-term treatment-naïve growth dynamics. It provides a reference for upcoming preclinical treatment developments and clinical trial end points.

Introduction

Worldwide, 32.4 million individuals are blind (visual acuity <3/60), and approximately 6.5% of these cases are due to age-related macular degeneration (AMD).¹ AMD affects more than 180 million people globally and this number is expected to increase to 288 million patients by 2040.^1,2 Independent of disease stage, the nonexudative, dry form of the disease accounts for approximately 85% of all AMD cases.³ Estimates assume that approximately 6 million patients are currently affected worldwide by the late state of dry AMD with geographic atrophy (GA).⁴ GA is defined as atrophy of the outer retina owing to degeneration of photoreceptors, retinal pigment epithelium cells, and the choriocapillaris.⁵ To date, two complement system inhibitors, pegcetacoplan (C3 inhibitor) and avacincaptad pegol (C5 inhibitor), are approved by the U.S. Food and Drug Administration.^6–8 Both can improve the structural course of the disease by slowing down GA growth. Neither of the two can restore vision. The European Medicines Agency judged the clinical benefit for the patients insufficient for European market approval of pegcetacoplan.⁹ Thereupon, avacincaptad pegol was withdrawn by the manufacturer from the application process.¹⁰ This development highlights the need for European long-term data on GA kinetics and new biomarkers. For this purpose, the natural history of GA in treatment-naïve patients with dry AMD was examined. The former and current gold standards for GA imaging, fundus autofluorescence (FAF), and optical coherence tomography (OCT) were used to ensure long-term data comparability and imaging precision. This is one of the most comprehensive GA cohorts in Scandinavia. 

Methods

Study Design

We conducted a prospective, single-center, observational study between February 2013 and September 2024 at the Department of Clinical Neuroscience, Karolinska Institutet, Stockholm, and at the St. Erik Eye Hospital, Solna, Sweden. The study was approved by the local ethics commission (project number: Dnr 2024-02838-02) and adheres to the tenets of the 1964 Declaration of Helsinki and its later amendments. 

Imaging

FAF images were analyzed in this study. Described in 1995,¹¹ FAF has been used as the gold standard for GA imaging until cross-sectional OCT became the current imaging gold standard for GA.¹² The establishment of the patient cohort of this study precedes the newer OCT-based terminology for GA, complete retinal pigment epithelial and outer retinal atrophy.^13,14 To ensure long-term comparability, the old nomenclature, GA, was kept for the analysis in this study. GA was defined as area of hypoautofluorescence with a possible hyperautofluorescent surrounding junctional zone, corresponding with the anatomically visible lesion in clinical fundus biomicroscopy.¹⁵ 

The FAF images were acquired with a confocal scanning laser ophthalmoscope, Heidelberg Retina Angiograph, on a Spectralis system (Heidelberg Engineering, Heidelberg, Germany). A blue light spectrum laser with 486-nm wavelength was used, which excites the lipofuscin molecules within the retinal pigment epithelium cells.^11,16 A 30° lens was used. If fixation stability of the patient allowed, the slower but better quality high-resolution mode was used with a summation of 100 single images to one output image (automated real time mean¹⁷ of 100). The scan area was centered on the GA lesion, at the same time making sure to also capture all satellite lesions if present. The same protocol was repeated for each patient for each follow-up visit. 

Participants

Patients with a clinical diagnosis of AMD and GA were identified by a senior AMD specialist (AK) at St. Erik Eye Hospital. Furthermore, leaflets were sent out to local ophthalmologists with the request for referral of patients with GA. Best-corrected visual acuity (BCVA) testing was performed in accordance with Early Treatment of Diabetic Retinopathy Study (ETDRS) letter charts at a distance of 4 m (LQ). Intraocular pressure measurement using rebound tonometry (iCare Finland Oy, Vantaa, Finland) was done (LQ). Slit lamp biomicroscopy (BQ 900, Haag-Streit, Köniz, Switzerland) (AK, DRM), FAF (details as described elsewhere in this article), and spectral-domain OCT (Spectralis, Heidelberg Engineering; high-resolution mode, automated real time mean of 12, volume scan centered on macula with 49 line scans) were performed at baseline to confirm the diagnosis and describe GA lesion characteristics (LQ, DRM). Patients were followed every 6 months with a clinical assessment and FAF imaging. Inclusion criteria to enter the study, apart from patient's consent, were GA lesion owing to AMD and at least one follow-up visit 3 or more months apart from baseline with BCVA measurements and retinal imaging. Exclusion criteria were a history of exudative AMD, presence of macular neovascularization, macular pathologies such as vitreomacular traction syndrome, macular hole (full-thickness macular hole, lamellar macular hole), macular fibrosis, and macular degeneration of other etiology. Specifically, any suspicion of macular neovascularization on the baseline OCT imaging led to exclusion of the eye. This included eyes with intra- or subretinal exudation, as well as eyes with signs of nonexudative, silent macular neovascularizations such as pigment epithelium detachment, double-layer sign of retinal pigment epithelium and Bruch's membrane, and subretinal hyper-reflective subretinal material. OCT angiography was performed, if necessary, because it soon became available in our hospital (in 2017). 

The clinical data were exported in a pseudonymized format by removing identifiable personal information and allocating unique but not back-traceable study identifiers. The encoding key was stored separately on a hard drive and was only accessible to study investigators. 

The corresponding imaging data were reviewed to confirm the diagnosis of AMD with GA and to exclude retinal comorbidities (AK). The same pseudonymization process was done with the imaging data (DRM). 

Owing to a FAF/OCT device change in February 2014, imaging data were only available from this timepoint onward. In case the lesion did not completely fit into a 30° FAF scan (e.g., owing to excentric image acquisition), 55° wide-angle images were analyzed if available; otherwise, the images were excluded. Wherever available, corresponding OCT scans were used to cross-check the extent of the GA lesion and the involvement of the fovea (fovea sparing vs. fovea involving). The fovea status was defined on a structure–function approach: The baseline OCT images were screened for atrophy of the retinal pigment epithelium and outer retina with respective dorsal hypertransmission on the foveolal scan and three scans above and below. With a 5.8 × 5.8 mm (20° × 20°), 49-line, high-resolution volume scan the single line scans were 122 mm apart from each other. By reviewing the foveolal line scan ±3 scans the central ∼720 mm (corresponding with the inner dashed circle of the ETDRS-like overlay on the Spectralis device) were covered. Eyes with a complete atrophy in this area were defined as fovea involving. In eyes that showed structurally a partially fovea-sparing condition on the OCT at baseline (for an example, see Fig. 1), the final classification into fovea sparing vs. fovea involving was made using visual acuity as functional parameter. This decision was supported by the findings of previous studies that had shown that a purely structural-based classification might be insufficient, especially because there is a lack of consensus for an OCT-based definition of fovea sparing.¹⁸ The extent in structural and functional damage is not necessarily similar.¹⁹ In literature, a BCVA cut-off of greater than 65 and 70 or more ETDRS letters, respectively, is most commonly applied to a fovea-sparing definition.^18,20–22 For this study, a BCVA of 65 or more ETDRS was defined as fovea sparing. Otherwise, the participant was enrolled in the fovea-involving group. Further subgrouping was done based on the lesion configuration (multifocal vs. unifocal) and laterality (unilateral affection of GA vs. bilateral GA). 

Figure 1.

 

Example of a partially fovea-sparing condition. (A) Near infrared en face image with a GA lesion (white) in the macular area temporal and inferior around the fovea. Green box represents the area of the total 49 horizontal OCT scans. Green line represents the location of the cross-sectional OCT scan number 30 that is depicted in (B). (B) Horizontal cross-sectional OCT scan through the fovea showing complete retinal pigment epithelial and outer retinal atrophy (cRORA) with dorsal hypertransmission located temporal and inferior perifoveally. The fovea's structure seems to be partially intact.

Image Analysis

Semiautomated measurement of the GA lesion area was done on the FAF images with the Heidelberg RegionFinder software (HSFRegFinder v2.6.5.0) that is built into the Spectralis system (Spectralis Viewing Module v7.0.4, Heidelberg Engineering). All RegionFinder analyses were done under supervision by one study investigator (DRM). Manual corrections were made where the automated image analysis algorithm was considered to have failed by the responsible study investigator (DRM), for example, misinterpreted retinal vessels for GA areas or incorrect lesion boundary recognition. In these cases, the algorithm was guided by manually set boundaries. 

The lesion area measurements in square millimeters were exported directly from RegionFinder as a comma-separated values file and imported and organized in Excel (v16.78 for Mac, Microsoft Corp., Redmond, WA) and matched to the clinical data by the study identifier. 

The baseline visit was defined as earliest available FAF image showing GA. The last follow-up visit was defined as latest FAF image until data lock. The follow-up intervals were calculated and defined in days to be most precise. For readability, the final results were given in months, assuming that 1 month mathematically consists of 30.42 days (365 days/12 months). The GA growth rate (GR) (in mm²/y) was calculated as the lesion area difference between the baseline and respective last visit divided by the individual maximum follow-up time. The square root–transformed GR (sqrtGR) was calculated as previously shown in literature to obtain values that are less dependent on the absolute baseline lesion area.^23–25 To visualize the progression over time, GR values were plotted over time for each follow-up for each included eye (Fig. 2). The sqrtGRs are displayed (in mm/year) compared with the respective previous follow-up for each eye (gray dots). To descriptively illustrate the GR trend a locally estimated scatterplot smoothing (blue line) and 95% confidence interval bands (gray) were fitted to the plot. To highlight the long-term GRs, an additional figure was plotted showing the complete individual sqrtGR for each included eye with a follow-up of 5 or more years (Supplementary Fig. S1). 

Figure 2.

 

Plot of the dynamic of the square root transformed GA GR (sqrtGR) [mm/year] (y axis) per the follow-up visit [years] (x axis) compared with the respective previous follow-up for each eye. Each dot represents a measurement point of an eye. The blue line represents the locally estimated scatterplot smoothing (LOESS) line. The 95% confidence interval is illustrated by one gray band above and one below the LOESS curve.

Statistical Analysis

Data were analyzed using R (v4.3.1, The R Foundation for Statistical Computing c/o Institute for Statistics and Mathematics, Vienna, Austria), JASP (v0.19.1 for Apple Silicon, Department of Psychological Methods, University of Amsterdam, Amsterdam, the Netherlands) and StatPlus Pro (v8.0.1.0 for MacOS, AnalystSoft, Walnut, CA). Descriptive statistics for all subgroups were computed. Normal distribution was tested with the Kolmogorov–Smirnov test. To evaluate differences between the unpaired independent, not-normally distributed subgroups a two-tailed Mann–Whitey U (z value) test was used. A P value of less than 0.05 was considered statistically significant. To evaluate the influence of the lesion characteristics on the GR in a combined model, an analysis of variance was performed. The independent categorical variables (fixed factors) were laterality (unilateral vs. bilateral), configuration (unifocal vs. multifocal), fovea status (fovea sparing vs. fovea involving). The numerical dependent variables were GR and sqrtGR from BL, respectively. To account for a possible bias of bilateral inclusion a statistical mixed model was performed for sqrtGR from BL with random factor participant ID. The participant ID was identical for the right eye and the left eye of same participant and was therefore used as random factor to correct for bilaterality. 

Results

For this study, 432 eyes with GA were examined (data lock on September 4, 2024) of which 204 eyes (n = 204) of 111 patients met the inclusion criteria. Table 1 summarizes the inclusion process. Patients with a GA lesion owing to dry AMD were included. All other confounding retinal conditions, such as exudative AMD, uncontrolled glaucoma, diabetic retinopathy, and inherited retinal diseases, were excluded from the analysis. Cataract that allowed for sufficient clinical biomicroscopy and imaging was not an exclusion criterion, nor was an uneventful cataract surgery without complications. Detailed demographic data are given in Table 2. The median age of the study participants was 79 years, well in line with AMD being an age-related condition. The eye and gender distribution turned out to be almost symmetrical. The mean spherical equivalent and intraocular pressure were well within the physiological range. The BCVA at baseline was significantly better for fovea-sparing participants, as expected. Table 3 provides a summary of characteristics of the total cohort. The maximum follow-up period was 123 months, with a median follow-up time of 21 months. For eyes with fovea sparing at baseline, 22 eyes (30%) converted to foveal involvement over a median follow-up period of 24 months (Table 3). The total cohort grew in mean by 3.59 mm² over the whole follow-up period, which represents a statistically significant growth (P < 0.001). Table 4 provides the GR analysis for all patients as well as for subgroups based on lesion characteristics. The mean overall GR was 1.597 mm²/year and the mean sqrtGR was 0.264 mm/y. It seemed that bilateral lesions grew faster than unilateral ones. Independently, multifocal lesions showed a faster growth than unifocal lesions. When evaluating the influence of the fovea status on the GR, significantly faster growth for fovea-sparing patients was noted independently on the other lesion characteristics (Mann–Whitney U all P values < 0.05). Table 5 provides an analysis of variance to evaluate the influence of the lesion characteristics on the GR in a combined model and to identify possible interactions. The model suggested that only laterality, remained as a significant factor (P = 0.021) influencing GR (Table 5). Using the same model for sqrtGR, laterality (P = 0.030) and fovea status (P = 0.001) showed significant influence (Table 5). The significance of lesion configuration (unifocal vs. multifocal) disappeared in the combined model. The mixed model controlling for bilaterality also showed that fovea status is the main influencer on sqrtGR (P < 0.001) and the lesion configuration was nonsignificant either for GR (P = 0.742) or sqrtGR (P = 0.719) (Table 6). Figure 2 shows the dynamic of the sqrtGR over the follow-up period. The sqrtGR showed a short decrease over the first 12 months, remained almost constant for the following 12 months (year 2) and was then followed by an almost linear, slow decrease over the following years. The same trend could be noted in the individual sqrtGR plots for all eyes with a long-term follow-up of 5 or more years (Supplementary Fig. S1). 

Table 1.

 

Participant Inclusion

 

Table 2.

 

Demographics of Included Participants

 

Table 3.

 

Results Summary for Total Cohort

 

Table 4.

 

Results Subgroups

 

Table 5.

 

Analysis of Variance With Dependent Variable Absolute and Square Root Transformed GA GR, Respectively, From Baseline to Respective Last Follow-up Visit

 

Table 6.

 

Generalized Linear Mixed Model Dependent Variable Absolute (GR) and Square Root–transformed GA GR (sqrtGR), Respectively, From Baseline to Respective Last Follow-up Visit

 

Discussion

AMD is still lacking sufficient treatment. Therefore, longitudinal natural history data of treatment-naïve GA growth kinetics are essential for the understanding of the pathophysiology of this disease and predicting individual future growth. A broad dataset on GA growth is needed to act as a reference against upcoming treatments. This study comprises the most extensive GA patient cohort in the Nordics with structured, high-quality data over a long follow-up period. 

Overall GRs observed in this long-term study (1.60 mm²/y and 0.26 mm/y, respectively) are in line with previous original studies where GR values range between 1.20 and 2.79 mm²/y and 0.21 and 0.39 mm/y, respectively (Table 7). In their review articles, Fleckenstein et al.⁵ and Wang et al.²⁶ reported similar values to this study with an overall absolute GR across studies of 1.78 mm²/y and 1.66 mm²/y, respectively. 

Table 7.

 

Overview of Original Literature on Geographic GR (Sorted by Publication Year) in Absolute (mm²/y) and Square Root-Transformed (mm/y)] Measures

 

GA GRs have been shown to vary considerably.²⁷ The long-term data from this study confirm that the growth dynamic is not purely linear. It suggests a slow down of growth over the total observation period. In detail, the GR showed a quick, hyperbolic-shaped decline during the first year, remained almost constant the second year before entering a state of almost linear, slow decline in the following years (Fig. 2 and Supplementary Fig. S1). This finding is congruent with a slow down with increasing distance from the fovea which was found by Vogl et al.,²⁷ Moult et al.,²⁸ and Mauschitz et al.²⁹ To describe the growth dynamic for long-term follow-ups, Fleckenstein et al.³⁰ suggested a nonlinear model. Sunness et al.³¹ and Arslan et al.³² suggested a sigmoid curve as a dynamic model to describe the tailing-off effect of GA growth. Despite that, linear statistical models are often found to be most accurate to represent GA growth dynamics.³² One reason might be a bias introduced by short observation periods in real-world data, where the toe and head end of the sigmoid curve might be underrepresented.^30,32 Owing to a lack of disease awareness and effective treatments, patients present or are referred late in the course of the disease. Then, GA growth already is in its linear state. On the other end, patients might have dropped out before the tailing-off could be measured.³² The variation of GA growth depending on the age of the lesion is an important factor when evaluating GA growth dynamics and when assessing potential treatment efficacy. This factor leads to a limitation of the present study as the cohort included participants with variable individual follow-up times owing to continued recruitment and loss to follow-up (including death). Figure 2 discloses the sqrtGR for each included eye at its respective last visit. Yet, the actual age of the GA lesion remains mostly unknown as the conversion date from intermediate AMD without GA to late-stage AMD with GA was not ascertainable for the majority of participants. Because this factor affects other published studies in a similar manner, GA GRs during their respective observation periods can be evaluated and are comparable between studies (Table 7). 

Previous studies have shown that not only lesion age causes variability in the growth pattern, but also biological processes and multifactorial lesion characteristics.^4,30 In this study, a significant overall influence of the foveal involvement, laterality, and lesion configuration on the GR could be noted when assessed separately (Mann–Whitney U test) (Table 4). In concordance with the faster growth of fovea-sparing lesions in this study, Fleckenstein et al.⁵ observed higher GRs for this phenotype. Similarly, they found a significantly faster growth for multifocal and bilateral lesions. In the present study, when evaluated independently from each other, all lesion characteristics were statistically significant (Table 4). In contrast, Reiter et al.³³ did not find a significant influence on GA growth for lesion configuration when analyzing it in a mixed model. When controlling for bilateral inclusion, we also found that fovea status was the only influencer on sqrtGR (Table 6). The differences in the significance presented in Tables 4–6 suggest that there are interactions between the lesion characteristics. This is of translational relevance as recorded GRs can act as surrogate marker for future individual growth.³⁴ Longer recorded GRs bear the potential for refined individual growth predictions, which are important for preclinical and clinical research. 

Conclusions

In this GA cohort, we demonstrate an influence of lesions characteristics on GRs with follow-up times of up to more than 10 years. Evaluated individually, fovea-sparing, multifocal lesion configuration and bilateral lesions showed significant faster GA growth. In combined statistical models, fovea sparing remained as a significant risk factor for faster growth that was corrected for baseline lesion size by square root transformation. Patients presenting one or more of these lesion characteristics hold a high potential for benefit of future treatments because a slow-down of lesion growth may be more likely to be achieved. In fovea-sparing cases, functional preservation may be possible. 

Acknowledgments

The authors thank the staff of St. Eriks Eye Hospital that helped to schedule and organize the patient visits. Their contribution made the study workflow as smooth as possible for all involved parties. 

DRM is supported by Stiftung OPOS, St. Gallen, Switzerland; Alfred Vogt-Stiftung, Zurich, Switzerland. PAW is supported by St. Erik Eye Hospital philanthropic donations; Vetenskapsrådet 2022-00799; ARMEC Lindebergs Stiftelse. 

Disclosure: D.R. Muth, Annexon Bio (C), Astellas (C), Custom Surgical (C), Roche (C); L. Quérat, None; A.P. Venkataraman, None; A. Dominguez-Vicent, None; G. Petrovski, Abbvie (C), Alcon (C), Allergan (C), Bulbitech (C), Profundus (C), RetinaRISK (C), Roche (C), Santen (C); P.A. Williams, None; F. Locri, None; S.A. Zweifel, Alcon (C), Allergan (C), Bayer (C), Endogena (C), Novartis (C), Roche (C), Zeiss (C); A. Kvanta, Alder Therapeutics (C), Apellis (C), Novartis (C), Roche (C)