**Purpose**:
To determine the impact of topographic locations on the progression rate of geographic atrophy (GA).

**Methods**:
We searched in five literature databases up to May 3, 2019, for studies that evaluated the growth rates of GA lesions at different retinal regions. We performed random-effects meta-analyses to determine and compare the GA effective radius growth rates in four location groups defined by two separate classification schemes: (1) macular center point involved (CPI) or spared (CPS) in classification 1, and (2) foveal zone involved (FZI) or spared (FZS) in classification 2. We then estimated the GA growth rate in eight topographic zones and used the data to model the GA expansion.

**Results**:
We included 11 studies with 3254 unique eyes. In studies that used classification 1, the effective radius growth rate was 30.1% higher in the CPS group (0.203 ± 0.013 mm/year) than in the CPI group (0.156 ± 0.011 mm/year) (*P* < 0.001). This trend became significantly more prominent in classification 2, where the growth rate was 61.7% higher in the FZS group (0.215 ± 0.012 mm/year) than in the FZI group (0.133 ± 0.009 mm/year) (*P* < 0.001). The estimated GA effective radius growth rates in eight retinal zones fit a Gaussian function, and the modeling of GA expansion gave rise to various GA configurations comparable to clinical observations.

**Conclusions**:
This study indicates that the GA progression rate varies significantly across different retinal locations. Our analysis may shed light on the natural history and underlying mechanism of GA progression.

^{1}GA has been reported to affect roughly 6 million people globally, and its prevalence increases dramatically with age.

^{2}

^{,}

^{3}During the early stage of GA, the lesion typically starts in the parafovea and then enlarges to form a ring surrounding the fovea.

^{4}During this stage of foveal sparing,

^{5}

^{–}

^{8}patients usually have decent central foveal function with preserved visual acuity.

^{4}

^{,}

^{9}As GA continues to progress, it will reach the fovea and result in a dramatic loss in central visual acuity and eventually legal blindness.

^{1}

^{,}

^{4}

^{,}

^{5}

^{,}

^{9}

^{4}

^{,}

^{10}

^{–}

^{15}Thus, it can potentially serve as a biomarker to predict the GA growth rate in patients and allow stratification for confounding in clinical trials. However, the terminologies used to describe GA locations were inconsistent in the literature, and there are at least five different terminologies: foveal or extrafoveal,

^{13}

^{,}

^{15}

^{,}

^{16}center involved or not involved,

^{10}central or noncentral,

^{12}subfoveal or non-subfoveal,

^{11}with or without RPE atrophy under the foveal center.

^{14}After closely examining the definitions of the terminologies, we have found two different classification schemes to describe the baseline location of GA based on whether GA lesions involve (1) the center point of the macula (classification 1) or (2) the foveal zone (classification 2). However, even within one classification, the GA area growth rate of each group still varies widely. For example, the growth rate in GA involving the center point of the macula ranges from 1.06 to 1.89 mm

^{2}/year.

^{11}

^{,}

^{12}Also, it is currently unknown which classification scheme would result in a more clinically significant difference in the GA growth rate between groups.

^{4}

^{,}

^{9}

^{,}

^{17}However, there is a significant disagreement on the degree to which the topographic zone will impact the GA growth rate. The reported average growth rate of GA area varies dramatically in each topographic zone (e.g., 0.04–1.14 mm

^{2}/year in the most central zone of the macula) and the definition of the topographic zones also varies among different studies.

^{4}

^{,}

^{9}

^{,}

^{17}Moreover, it is currently unknown whether the progression rate of GA remains constant within one topographic region or changes continuously as a function of the distance to the foveal center. The mechanism underlying the differential GA progression rates between in the fovea and extrafovea is also unclear.

^{18}

^{12}

^{,}

^{19}and patients with advanced AMD in one eye are sometimes advised to take the supplements.

^{20}

*n*is the mean follow-up time (years),

*A*is the mean baseline area, and

*G*is reported mean annual GA area growth rate. The SE of the GA effective radius growth rate was calculated by \(\sqrt {\frac{{0.0795AG_1^2{n^2} + 0.0795A_1^2{{( {\sqrt {A + Gn} - \sqrt A } )}^2}}}{{A{n^2}( {A + Gn} )}}} \), where

*n*is mean follow-up time (years);

*A*and

*A*

_{1}are the mean and SE of baseline GA area, respectively; and

*G*and

*G*

_{1}are the mean and SE of the reported annual GA area growth rate, respectively. This equation was derived from error propagations of the function \(mean\;GA\;radius\;growth\;rate = \;\frac{1}{{\sqrt \pi \times n}} \times \;( {\sqrt {A + n \times G} - \sqrt A } )\). Other necessary extrapolation methods are detailed in Table 1. Disparities between the reviewers were resolved through discussion and subsequent consensus.

^{2}.

^{21}In addition, several previous studies showed that using the square-root transformed GA area (equivalent to effective radius) reduces the dependence of the GA growth rate on the baseline.

^{22}

^{,}

^{23}We also reported the conventional outcome measure (GA area growth rate expressed in mm

^{2}/year) in the present paper.

^{24}for R 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria). We chose the random-effects approach to allow for unexplained heterogeneity across studies.

^{25}To compare the mean difference in the GA growth rate (derived from the above random-effects meta-analysis) between classification 1 and classification 2, we performed an unpaired

*t*-test. To confirm the observation that GA effective radius enlarges linearly as a function of time in each GA group, we plotted the average GA effective radius as a function of time after enrollment for each group. However, the baseline GA sizes usually vary widely among different studies,

^{21}suggesting that the different patient populations were at different time points of the disease course when they were enrolled in the studies. To correct for the differences in the entry time into the clinical studies, we added a horizontal translation factor (in years) to each raw data subset.

^{21}

^{,}

^{26}

^{–}

^{31}The translation factor essentially converted the horizontal axis from time after enrollment to inferred duration of GA, where the inferred duration of GA = time after enrollment + translation factor. To find the optimum translation factors, we first estimated a wide range for the translation factor of each study. We then adjusted one of the translation factors by 1 month at a time within the estimated range until the

*r*

^{2}was maximized for the cumulative trend line with a predetermined slope of GA effective radius growth rate calculated from the above random-effects meta-analysis.

^{4}we used the reported average radius of the fovea (750 µm) as the cut-off radius for this zone.

^{32}For each extrapolated topographic zone, we then estimated the GA effective radius growth rate by calculating the weighted mean (weighted by the number of eyes in each study) of the GA effective radius growth rates in the corresponding zones from the included studies. This estimation assumes that the GA effective radius growth rate is relatively unchanged within each zone. For example, to estimate the GA effective radius growth rate in the first zone (0–500 µm from the foveal center), we calculated the weighted mean of the GA effective radius growth rates in the residual foveal island zone in Lindner et al.,

^{4}in the 0 to 600 µm zone in Mauschitz et al.,

^{17}in the 0 to 500 µm zone in Sayegh et al.,

^{9}and in the 0 to 1800 µm zone in Sunness et al.

^{5}(Supplementary Table S4). Based on the relationship between the GA effective radius growth rate and the distance to the foveal center (retinal eccentricity), we modeled the expansion course of GA lesions over 30 years using MATLAB software (MathWorks; Natick, MA). For the modeling, we used the GA effective radius growth rate (mm/year) at each location as the estimated length (mm) that the GA border would travel in 1 year at the same location.

^{33}which is one of the most widely used risk of bias assessment tools for meta-analysis of observational studies.

^{34}Inconsistencies were discussed until agreement was reached. Heterogeneity was assessed by calculating the

*I*

^{2}statistic in each random-effects meta-analysis. Also, for random-effect meta-analyses with data from at least three studies, a sensitivity analysis was performed by removing one study at a time to assess whether a single study influenced the outcomes of the meta-analyses. To investigate the impact of potential confounding factors on estimated GA growth rates and statistical comparisons, we performed subgroup analysis stratified by imaging methods and study types if there were at least two studies in the subgroup.

^{4}

^{,}

^{5}

^{,}

^{9}

^{–}

^{17}Of note, one of the included articles (Holz et al., 2018

^{11}) reported data from two studies, and two of the included articles (Schmitz-Valckenberg et al., 2016

^{13}and Mauschitz et al., 2012

^{17}) were about the same study but reported data in different analyses (Tables 1 and 2).

^{2}) and CPS (4.677 ± 0.102 mm

^{2}) groups are similar, and the baseline GA sizes in the FZI (8.348 ± 0.686 mm

^{2}) and FZS (6.940 ± 0.712 mm

^{2}) groups are comparable.

^{4}

^{,}

^{5}

^{,}

^{9}

^{,}

^{17}Of note, one study (the geographic atrophy progression study)

^{13}

^{,}

^{17}reported data for GA groups stratified by classification 1 and data regarding topographic distributions of GA lesions in at least two topographic zones. Thus, this study was included in two separate analyses. The excluded articles and reasons for exclusion are listed in Supplementary Table S2. All 11 included articles were deemed to have a low risk of bias as assessed by the Newcastle-Ottawa Scale (scores between 6 and 8; Supplementary Table S3).

^{2}/year) in the four GA location groups (CPI, CPS, FZI, and FZS groups) are shown in Supplementary Figure S1. The forest plots comparing the GA area growth rate between the pair in classification 1 (CPI versus CPS) and the pair in classification 2 (FZI vs. FZS) are shown in Supplementary Figure S2. From the random-effects meta-analyses shown in the forest plots, the GA area growth rate was 33.1% higher in the CPS group (1.995 ± 0.261 mm

^{2}/year) than in the CPI group (1.499 ± 0.180 mm

^{2}/year) (

*P*< 0.001) (Supplementary Fig. S3). Similarly, the GA area growth rate was found to be 58.8% higher in the FZS group (2.196 ± 0.130 mm

^{2}/year) than in the CPI group (1.383 ± 0.094 mm

^{2}/year) (

*P*< 0.001) (Supplementary Fig. S3).

^{21}

^{–}

^{23}Thus, to better account for different baseline lesion sizes, we determined the GA effective radius growth rate (mm/year) in all four GA location groups using random-effects meta-analyses (Supplementary Fig. S4). We then conducted random-effects meta-analyses to compare the effective radius growth rate between the CPI and CPS groups and between the FZI and FZS groups (Fig. 2). As demonstrated in Figures 2 and 3A, the GA effective radius growth rate was 30.1% higher in the CPS group (0.203 ± 0.013 mm/year) than in the CPI group (0.156 ± 0.011 mm/year) (

*P*< 0.001); the GA effective radius growth rate was 61.7% higher in the FZS group (0.215 ± 0.012 mm/year) than in the FZI group (0.133 ± 0.009 mm/year) (

*P*< 0.001). Interestingly, the mean difference in the GA effective radius growth rate between the pair in classification 2 (0.082 ± 0.014 mm/year between the FZS and FZI groups) was 70.8% higher than that in classification 1 (0.048 ± 0.005 mm/year between the CPS and CPI groups) (

*P*= 0.01) (Fig. 3B). This result suggests that the involvement of the foveal zone (classification 2) is a more clinically significant prognostic factor to predict the GA growth rate compared to the involvement of the center point of the macula (classification 1).

*I*

^{2}ranged from 0% to 32%, as shown in Figure 2 and Supplementary Figure S2. The sensitivity analysis shows that after removing one study at each time, the statistical significances in Figure 2 and Supplementary Figure S2 did not change significantly, suggesting that our results were not driven by any one of the included studies. In the stratified analysis, the GA effective radius growth rate was consistent between prospective interventional studies and prospective observational studies (0.179 ± 0.010 vs. 0.193 ± 0.014 mm/year;

*P*= 0.56) (Supplementary Fig. S5). After the removal of one retrospective observational study, the GA growth rate in the FZS group was still significantly higher than the growth rate in the FZI group (

*P*< 0.001), and the difference in the GA growth rate between the two groups was unaffected (Supplementary Fig. S6). Similarly, there was no significant difference in the GA growth rate reported by studies using fundus autofluorescence (0.186 ± 0.010 mm/year) versus studies using color fundus photography (0.178 ± 0.013 mm/year) (

*P*= 0.65) (Supplementary Fig. S7). Also, the choice of imaging method did not affect the estimated difference in the GA growth rate between the CPS and CPI groups (Supplementary Fig. S8).

*r*

^{2}(between 0.98 and 0.99) over nearly 10 years (Fig. 4B), suggesting that the GA effective radius increases linearly over the elapsed time in each GA group but with a distinct growth rate.

^{35}we modeled three GA lesions with different onset times (5 years apart) starting in the parafovea (Fig. 6C). Interestingly, the three GA lesions would grow into a ring-shaped lesion with fovea sparing and eventually cover the entire fovea (Fig. 6C and Supplementary Video S3), which is similar to the clinical observations of GA expansions starting outside the foveal island.

^{4}

^{,}

^{36}

*P*< 0.001). Interestingly, the clinical significance of the two classifications may not be equal. We found that the mean difference in the GA growth rate between the two groups in classification 2 (0.082 ± 0.014 mm/year) was 70.8% higher than that in classification 1 (0.048 ± 0.005 mm/year) (

*P*= 0.01) (Fig. 3B). The current data suggest that, although classification 2 (three studies with 386 eyes) was less popular than classification 1 (five studies with 2608 eyes), the GA involvement of the foveal zone (i.e., classification 2) may be a stronger biomarker to predict the GA growth rate and allow stratification for confounding in clinical trials.

^{13}

^{,}

^{15}

^{,}

^{16}center involved or not involved,

^{10}central or noncentral,

^{12}subfoveal or non-subfoveal,

^{11}and with or without RPE atrophy under the foveal center.

^{14}To resolve the conflicts and avoid potential confusions, we propose the use of the terms “center point involved” or “center point spared” to describe GA involvement of the center point of the macula (i.e., classification 1), and we propose the use of the terms “foveal zone spared” and “foveal zone involved” to describe GA involvement of the foveal zone (i.e., classification 2), which can be defined as the central circular zone that is 750 µm in radius.

^{32}

^{5}

^{,}

^{37}but the underlying mechanism for the various configurations remains unknown. Our study suggests that the occurrence of different shapes of GA lesions may not be random. Rather, the different patterns that we observe may simply be due to the fact that the GA growth rate changes at different retinal locations, leading GA lesions to evolve into the various shapes at different time points. This hypothesis is also consistent with a previous observation that more than 50% of GA lesions changed from one configuration to another over a few years.

^{5}By applying our Gaussian topographic profile of the GA effective radius growth rate (Supplementary Fig. S9B), we were able to model the course of GA expansions over the elapsed time (Fig. 6). Importantly, our model gave rise to multiple GA shapes (circular/oval, kidney-shaped, horseshoe, and ring at different stages of GA; see Figs. 6B and 6C) that are consistent with the configurations reported in the literature. Also, we predicted that a GA lesion starting in the foveal center would remain symmetric and circular (Fig. 6A), and GA lesions starting in the parafovea would cover the majority of the macula over a long period of time while still sparing part of the fovea until late in the course of the disease (second to the bottom images in both Figs. 6B and 6C, referred to by others as foveal sparing phenomenon). Both predictions corresponded well with previous observations.

^{5}

^{–}

^{7}

^{,}

^{38}

^{39}

^{–}

^{43}Some other groups suggested that the relatively increased choroidal blood supply might be protective to the fovea.

^{17}

^{,}

^{44}

^{,}

^{45}This may be supported by the findings that eyes with decreased choriocapillaris density are correlated with increased size and/or number of drusens,

^{46}and may also be supported by a recent study including an analysis of 33 eyes that found a positive correlation between the GA growth rate and the choriocapillaris flow impairment around the GA lesions.

^{47}The third hypothesis is based on the observation of the relatively high risk of developing AMD in eyes with decreased values of macular pigment.

^{48}In this hypothesis, the high density of macular pigment in the fovea may exhibit a protective effect and hinder GA progression. Although our findings might not exclude any of the hypotheses, the observed continuous variation of the GA growth rate across different retinal eccentricities points toward a local explanation (e.g., local anatomic variations) for the differential growth rates, rather than a systemic reason that would affect a topographic region homogeneously.

^{49}We found no significant interstudy heterogeneity among the included studies as assessed by the

*I*

^{2}statistical test in the random-effects meta-analyses shown in Figure 2 and Supplementary Figure S2. The sensitivity analysis showed that no single study affected the statistical significance in the random-effects meta-analyses. The stratified analysis demonstrated that different imaging methods or study types did not significantly affect the main conclusions, which supports the validity of our present meta-analysis results.

^{49}To further explore the impact of potential confounding factors, we performed subgroup analyses stratified by study designs and imaging methods. The results suggested that neither the study design nor the choice of imaging method affected the main conclusions significantly (Supplementary Figs. S5–S8). However, due to the limited number of studies in the literature that have investigated the association between the GA growth rate and topographic locations, we were unable to investigate the impact of other potential confounding factors (e.g., patient demographics, other characteristics of GA lesions).

^{17}) assessed the GA growth rate. Due to the limited number of studies, we were unable to perform a subgroup analysis to explore confounding effects. Thus, further longitudinal studies with a large number of patients affected by GA are needed to investigate the GA growth rate (mm/year) in linear axes in different retinal locations. Despite the limited data in the current literature, the modeling of GA expansion (Fig. 6) based on the derived topographic profile of the GA growth rate (Fig. 5) was consistent with the clinical observations of GA progression and resulted in various shapes of GA lesions described in the previous literature.

^{4}

^{,}

^{36}Third, in the literature, only three studies (386 eyes) used classification 2 to describe GA locations, as compared to the studies using classification 1 (five studies with 2608 eyes). To further compare the two classifications, a large longitudinal cohort study is needed to compare the effective radius growth rate of GA lesions in the spared and involved groups based on each classification.

^{4}did not specify an exact definition of the residual foveal island, so we used the reported average radius of the fovea (750 µm) as the cut-off radius for this zone.

^{32}As a comparison, we excluded this study and repeated all related analyses. We did not find any significant change in our results, and the topographic profile of the GA effective radius growth rate still fit a similar Gaussian function (Supplementary Fig. S9). Finally, because several of the studies we included were observational studies, survival bias is a potential concern.

**L.L. Shen**, None;

**M. Sun**, None;

**S. Khetpal**, None;

**H.K. Grossetta Nardini**, None;

**L.V. Del Priore**, Astellas Institute for Regenerative Medicine(C)

*Arch Ophthalmol*. 1975; 93: 19–25. [CrossRef] [PubMed]

*Lancet Glob Health*. 2014; 2: e106–e116. [CrossRef] [PubMed]

*Ophthalmology*. 2018; 125: 369–390. [CrossRef] [PubMed]

*Ophthalmology*. 2015; 122: 1356–1365. [CrossRef] [PubMed]

*Ophthalmology*. 1999; 106: 1768–1779. [CrossRef] [PubMed]

*J Vis Impair Blind*. 2008; 102: 600–610. [CrossRef] [PubMed]

*Invest Ophthalmol Vis Sci*. 2009; 50: 3915–3921. [CrossRef] [PubMed]

*Ophthalmology*. 2013; 120: 829–836. [CrossRef] [PubMed]

*Am J Ophthalmol*. 2017; 179: 118–128. [CrossRef] [PubMed]

*Ophthalmology*. 2013; 120: 2666–2671. [CrossRef] [PubMed]

*JAMA Ophthalmol*. 2018; 136: 666–677. [CrossRef] [PubMed]

*Ophthalmology*. 2018; 125: 1913–1928. [CrossRef] [PubMed]

*Ophthalmology*. 2016; 123: 361–368. [CrossRef] [PubMed]

*Ophthalmology*. 2018; 125: 1556–1567. [CrossRef] [PubMed]

*Transl Vis Sci Technol*. 2018; 7: 40. [CrossRef] [PubMed]

*Invest Ophthalmol Vis Sci*. 2016; 57: 2283–2289. [CrossRef] [PubMed]

*Invest Ophthalmol Vis Sci*. 2012; 53: 4932–4939. [CrossRef] [PubMed]

*JAMA*. 2000; 283: 2008–2012. [CrossRef] [PubMed]

*Arch Ophthalmol*2009; 127: 1168–1174. [CrossRef] [PubMed]

*Curr Opin Ophthalmol*. 2010; 21: 184. [CrossRef] [PubMed]

*Ophthalmol Retina*. 2018; 2: 914–921. [CrossRef] [PubMed]

*Ophthalmology*. 2011; 118: 679–686. [CrossRef] [PubMed]

*JAMA Ophthalmol*. 2013; 131: 110–111. [CrossRef] [PubMed]

*J Stat Softw*. 2010; 36: 1–48. [CrossRef]

*BMJ*. 2019; 364: k4597. [CrossRef] [PubMed]

*JAMA Ophthalmol*. 2013; 131: 335–340. [CrossRef] [PubMed]

*Ophthalmology*. 2009; 116: 1901–1907. [CrossRef] [PubMed]

*Am J Ophthalmol*. 2007; 143: 83–89.e82. [CrossRef] [PubMed]

*Retina*. 2019; 39: 1829–1839. [CrossRef] [PubMed]

*Ophthalmol*

*Retina*. 2019; 3: 305–315. [CrossRef] [PubMed]

*Ophthalmology*. 2019; 126: 1288–1296. [CrossRef] [PubMed]

*Stereoscopic Atlas of Macular Diseases: Diagnosis and Treatment*. St. Louis, MO: Mosby-YearBook; 1987: 464–469.

*The Newcastle-Ottawa Scale (NOS) for Assessing the Quality of Nonrandomised Studies in Meta-Analyses*. Ottawa, Ontario: Ottawa Health Research Institute, University of Ottawa; 2011. Available at: http://www.ohri.ca/programs/clinical_epidemiology/oxford.htm. Accessed June 11, 2017.

*World J Meta-Anal*. 2017; 5: 80–84. [CrossRef]

*Arch Ophthalmol*. 2002; 120: 579–584. [CrossRef] [PubMed]

*Eye*. 1988; 2: 552–577. [CrossRef] [PubMed]

*Invest Ophthalmol Vis Sci*. 2001; 42: 1051–1056. [PubMed]

*Invest Ophthalmol Vis Sci*. 2017; 58: 1001–1007. [CrossRef] [PubMed]

*Doc Ophthalmol*. 2004; 109: 1–8. [CrossRef] [PubMed]

*Invest Ophthalmol Vis Sci*. 2000; 41: 267–273. [PubMed]

*Prog Retin Eye Res*. 2011; 30: 115–128. [CrossRef] [PubMed]

*Invest Ophthalmol Vis Sci*. 1996; 37: 1236–1249. [PubMed]

*Ophthalmology*. 2007; 114: 1728–1735. [CrossRef] [PubMed]

*Eye*. 1987; 1: 120–125. [CrossRef] [PubMed]

*Am J Ophthalmol*. 2010; 150: 40–47.e2. [CrossRef] [PubMed]

*Invest Ophthalmol Vis Sci*. 2011; 52: 1606–1612. [CrossRef] [PubMed]

*PLoS ONE*. 2019; 14: e0212563. [CrossRef] [PubMed]

*Invest Ophthalmol Vis Sci*. 2001; 42: 439–446. [PubMed]

*Cochrane Handbook for Systematic Reviews of Interventions*. Version 5.1.0 [updated March 2011]. London, UK: The Cochrane Collaboration; 2011.