Evaluating the Effects of C3 Inhibition on Geographic Atrophy Progression from Deep-Learning OCT Quantification: A Split-Person Study

The FILLY phase 2 trial (Clinical Trials identifier: NCT02503332) reported the effect of pegcetacoplan on GA as evaluated using fundus autofluorescence [13]. This trial included 246 study subjects, with a single eye per person randomised into one of three treatment arms (intravitreal pegcetacoplan monthly [PM] or every other month [PEOM] and a pooled sham group including monthly and PEOM sham injections). This study involves human participants who were enrolled as patients at 46 sites in the USA (New England Institutional Review Board, University of Miami, Mayo Clinic, Institutional Review Board of the Cleveland Clinic Foundation, Duke University Health System Institutional Review Board, and Research Compliance Office Stanford University), Australia (Bellberry Ltd), and New Zealand (Northern A Health and Disability Ethics Committee, Health and Disability Ethics Committees, and Ministry of Health). The study was performed in accordance with the tenets of the Declaration of Helsinki, International Conference on Harmonisation Good Clinical Practice guidelines and all applicable regulations. Institutional review board or ethics committee approval was obtained at each site. All patients provided written informed consent.

The present analysis only considered patients with spectral domain (SD)-OCT imaging acquired using the Spectralis ophthalmic imaging platform (Heidelberg Engineering, Heidelberg, Germany) with an acquisition density of at least 25 B-scans per volume at both baseline and 12 months post-baseline. In addition, patients with either unilateral GA or choroidal neovascularisation (CNV), an exudative complication of AMD characterised by growth of aberrant vessels, were excluded from the analysis (Electronic Supplementary Material [ESM] Fig. S1), leaving a total of 144 patients (1 study eye and 1 fellow eye per patient).

The primary outcome of the analysis was a comparison between study eyes and their untreated counterparts in terms of mean change of RPE and outer retinal atrophy (RORA) from baseline to 12 months (in mm²). Secondary outcomes included change from baseline to 12 months for the constituent OCT features of RORA as per the CAM (Classification of Atrophy Meetings) definitions (defined as homogeneous choroidal hypertransmission and absence of RPE measuring at least 250 μm with overlying outer retinal thinning and photoreceptor loss), PRD in isolation (defined as PRD without overlapping RPE-loss or hypertransmission) and macula completely absent of any GA changes (intact macula). A non-OCT-based secondary outcome of interest was the area of GA quantified by manual segmentation of fundus autofluorescence (FAF) images [11]. Finally, Pearson correlation coefficients were used to assess the symmetry of baseline areas of GA as measured by both OCT and FAF, as well as concordance in growth rate by treatment group (RORA only).

The model used to automatically quantify GA features from OCT images has been previously published [14]. The area included in this analysis was restricted to the macula, as defined by the Early Treatment Diabetic Retinopathy Study (ETDRS) region, a circle measuring 6 mm in diameter centred around the fovea [15]. This allowed for accurate and near-instantaneous quantification of area from available OCT volumes.

All data analyses were carried out with R (https://​www.​r-project.​org/​) [16]. As a rule, the characteristic of interest could not be approximated using the normal curve, and non-parametric tests (Wilcoxon rank test) were used for comparisons. Statistical tests were two-sided with alpha = 0.05. The Pearson correlation coefficient was used to compare change in treated and their paired untreated eyes.

Of the 246 patients enrolled in the FILLY study, 144 had bilateral GA and therefore were taken forward for analysis (ESM Fig. S1). In the present analysis, these patients with bilateral GA were on average (SD) 79.4 (7.54) years of age, predominantly female (63%), and Caucasian (96%). This was also observed when sub-stratified by treatment group (Table 1). These demographic parameters of patients with bilateral GA did not observably differ from patients with unilateral GA or CNV (Table 1).

The GA lesions were characterised using both FAF and OCT imaging at baseline (Table 1c; Fig. 1). The total areas of GA lesions between study and fellow eyes were compared to query individual-level differences (i.e. inter-eye lesion symmetry) (Fig. 1; ESM Table S1); a strong positive concordance in baseline GA size was demonstrated when measured by either FAF (Pearson correlation coefficient = 0.67, P < 0.001) or OCT (RORA taken to be GA; Pearson correlation coefficient = 0.71, P < 0.001) (Fig. 1a). Indeed, the absolute difference between the study and fellow eyes was right-skewed with a median difference of 2.45 mm² (mean 3.30 mm², SD 2.87) when using FAF to query GA (Fig. 1b;ESM Table S1). A lower median difference at the individual level was observed when using OCT to quantify GA: 1.92 mm² (mean 2.48 mm², SD 2.13; Pearson correlation coefficient = 0.71, P < 0.001). Similarly, each of the constituent features of GA at the OCT level was also symmetrical at the patient level, including PRD (median 2.01 mm²; Pearson correlation coefficient = 0.73, P < 0.001); PRD in isolation (median 1.37 mm²; Pearson correlation coefficient = 0.72, P < 0.001); and RPE-loss (median 1.90 mm²; Pearson correlation coefficient = 0.71, P < 0.001).

The growth of GA at 12 months following baseline was compared for study eyes and untreated fellow eyes across each of the treatment groups (Table 2; Fig. 2; ESM Fig. S2). Nearly all OCT features of the GA growth rate were demonstrably slower in the study eyes of the PM treatment group when compared to corresponding untreated fellow eyes, including: RORA (study eye 0.792 mm² vs. fellow eye 1.13 mm², P = 0.003), PRD (0.739 mm² vs. 1.23 mm², P = 0.015), RPE-loss (0.789 mm² vs. 1.17 mm², P = 0.007) and intact macula (− 0.735 mm² vs. − 1.29 mm², P = 0.011); the exception was hypertransmission (HRT; 1.18 mm² vs. 1.04 mm², P = 0.186) (Table 2). In the PEOM treatment group, mean growth of GA was nominally lower in the study eyes than in the fellow eyes for any of the OCT features (RORA: study eye 1.14 mm² vs. fellow eye 1.44 mm², P = 0.353; PRD: 1.10 mm² vs. 2.08 mm², P = 0.104; RPE-loss: study eye 1.16 mm² vs. fellow eye 1.58 mm², P = 0.286; intact macula: − 1.22 mm² vs. − 2.17 mm², P = 0.121). When comparing study and fellow eyes in the sham group, average growth rates for OCT correlates of GA were not consistently lower in study eyes and a statistically significant difference was not observed. The annual growth rate of GA was also considered with FAF imaging (Table 2;Fig. 2). Here, a statistically significant difference was not detected between study and fellow eyes for any of the treatment groups.

An alternate way of visualising treatment effect on GA growth is represented in Fig. 3, which plots the distribution of patient-level RORA growth at 12 months around a line of no effect. In the case of no treatment effect, one would expect differences in growth between the study and fellow eyes to be normally distributed with the line as its origin and approximately half the values on either side. A treatment effect would be observed as asymmetrical growth, with slower growth observed in the treated eyes than in the untreated fellow eyes. In the sham group, approximately half of its participants fell on each side of the line of no effect. The correlation of enlargement rates between eyes was high (R = 0.64), suggesting high concordance of GA growth. Individual-level GA growth was higher in the fellow eye than in the study eye for 55% (32/58) versus 45% (26/58) of patients, respectively. The average (SD) distance to the line of no effect was 0.1 (1.2) mm². Conversely, the mean (SD) difference in GA area between the fellow and study eye in the PEOM group was 0.42 (1.2) mm², with 69% (36/52) of patients having faster GA growth in their fellow eye. Overall, concordance between fellow and study eye remained high, with a correlation coefficient of R = 0.68. In the PM group, however, a significant correlation of GA growth was no longer observed between fellow and study eyes (R = 0.21; P = 0.23). In the PM group, 75% (42/56) of patients had a higher GA growth rate in the fellow eye at 12 months, and the mean (SD) difference between eyes was 0.45 (1.2) mm². Assuming that GA growth is symmetrical at the individual level, these data suggest that a mean 0.45 mm² and 0.42 mm² of RORA over 1 year was prevented by monthly and every other monthly pegcetacoplan therapy, respectively, with the caveat of preserved inter-eye concordance and lack of statistical significance for GA area growth comparisons in the PEOM group.