December 2014
Volume 55, Issue 12
Free
Clinical and Epidemiologic Research  |   December 2014
Reappraisal of Geographic Atrophy Patterns Seen on Fundus Autofluorescence Using a Latent Class Analysis Approach
Author Affiliations & Notes
  • Marc Biarnés
    Institut de la Màcula i de la Retina, Barcelona, Spain
    Universitat Pompeu Fabra, Barcelona, Spain
  • Carlos G. Forero
    Health Services Research, Institut Hospital del Mar d'Investigacions Mèdiques, Barcelona, Spain
    Centro de Investigación Biomédica en Red de Epidemiología y Salud Pública, Barcelona, Spain
  • Luis Arias
    Hospital Universitari de Bellvitge, L'Hospitalet de Llobregat, Barcelona, Spain
  • Jordi Alonso
    Universitat Pompeu Fabra, Barcelona, Spain
  • Jordi Monés
    Institut de la Màcula i de la Retina, Barcelona, Spain
    Barcelona Macula Foundation, Barcelona, Spain
  • Correspondence: Marc Biarnés, Institut de la Màcula i de la Retina (Hospital Quirón Teknon), C/ Vilana, 12, Office 90, 08022 Barcelona, Spain; marcb@institutmacularetina.com
  • Carlos G. Forero, Health Services Research Unit, IMIM (Institut Hospital del Mar d'Investigacions Mèdiques), C/ Doctor Aiguader, 88, Edifici PRBB, 08003 Barcelona, Spain; cgarcia@imim.es
Investigative Ophthalmology & Visual Science December 2014, Vol.55, 8302-8308. doi:10.1167/iovs.13-13542
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Marc Biarnés, Carlos G. Forero, Luis Arias, Jordi Alonso, Jordi Monés; Reappraisal of Geographic Atrophy Patterns Seen on Fundus Autofluorescence Using a Latent Class Analysis Approach. Invest. Ophthalmol. Vis. Sci. 2014;55(12):8302-8308. doi: 10.1167/iovs.13-13542.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose.: To reappraise fundus autofluorescence (FAF) patterns in patients with geographic atrophy (GA) with the aim of simplifying the existing classification and to evaluate their stability with time using a data-driven approach, latent class analysis (LCA).

Methods.: One hundred seventy-one patients in the prospective, natural history study on GA (GAIN, NCT01694095) with a minimum follow-up of 12 months were screened. Five experienced observers independently evaluated FAF patterns according to the original classification, and LCA was used to determine the new, emerging categories (classes). A set of prespecified FAF features was then used to characterize each resulting class.

Results.: Seventy-five eyes of 59 subjects with a median follow-up of 19 months were included. The optimal LCA model resulted in five classes, which showed an association with GA size, among others. The classes did not change in a given individual during the study period, but the time frame may have been too short to evaluate hypothetical transitions.

Conclusions.: The original description of FAF patterns, which is based exclusively on distribution of hyperautofluorescence around GA, ultimately classifies patients according to area of atrophy. These results suggest that FAF patterns are not true phenotypes and that they rather represent different stages of the disease. This may have implications regarding the role of lipofuscin on disease pathogenesis.

Spanish Abstract

Introduction
Geographic atrophy (GA) is the advanced form of dry age-related macular degeneration (AMD) and is characterized by large areas of retinal pigment epithelium (RPE) atrophy, which grow a median of 1.3 to 2.6 mm2/y.1 Atrophy of the RPE is thought to precede photoreceptor and choriocapillaris loss.2 The poor understanding of its pathogenesis hinders the development of rational therapies. 
Lipofuscin accumulation is an important feature of aging and AMD.3 Lipofuscin is a byproduct formed by the incomplete phagocytosis of photoreceptor outer segments that builds up within the RPE. According to results from basic research, an important bisretinoid of lipofuscin, N-retinyledene-N-retinylethanolamine (A2E), has deleterious effects on cell physiology,4,5 although its role in GA pathogenesis is a matter of current debate.68 Given lipofuscin's naturally occurring autofluorescent properties, fundus autofluorescence (FAF) imaging can be used to evaluate noninvasively its topographic distribution in vivo.9 
In 2001, Holz et al.10 reported that areas of incident atrophy secondary to GA took place solely in regions that had previously shown elevated levels of autofluorescence on FAF. Later, the Fundus Autofluorescence in Age-related Macular Degeneration (FAM) Study described up to 10 different categories (which were defined as patterns or phenotypes) of GA according to the distribution of high autofluorescence surrounding the areas of atrophy.11 These original GA patterns form the “Specific” classification (Table 1). This was simplified into two additional classifications with progressively fewer categories, the “General” and “Simplified” classifications (these descriptive terms are used in this article to refer to each classification used in the FAM Study). Those patterns were associated with statistically significant and clinically relevant differences in the median rate of GA growth, ranging from 0.38 to 3.02 mm2/y, and they represent the most important risk factor for GA progression described to date. The association between increased FAF and incident atrophy in those studies supported the hypothesis of the causal effect of lipofuscin on GA enlargement. Actually, one line of experimental therapy, visual cycle modulation, aims at reducing the rate of accumulation of toxic metabolites such as lipofuscin within the RPE to halt GA progression.12 
Table 1
 
Geographic Atrophy Patterns as Originally Described With Fundus Autofluorescence in the Fundus Autofluorescence in Age-related Macular Degeneration Study11
Table 1
 
Geographic Atrophy Patterns as Originally Described With Fundus Autofluorescence in the Fundus Autofluorescence in Age-related Macular Degeneration Study11
Specific Classification General Classification Simplified Classification
None None None + Focal
Focal Focal None + Focal
Banded Banded Banded + Diffuse
Branching Diffuse Banded + Diffuse
Fine granular Diffuse Banded + Diffuse
Fine granular with peripheral punctate spots Diffuse Banded + Diffuse
Reticular Diffuse Banded + Diffuse
Trickling Diffuse Banded + Diffuse
Patchy Patchy Patchy + Undetermined
Undetermined Undetermined
Phenotypic characterization may contribute to unraveling the mechanisms of disease and opens the opportunity for genotype–phenotype correlations.13 It may also help to monitor disease progression14 and to predict individual response to therapy should treatment become available.15 An example is the clinicopathological correlation found in the different types of neovascularization in AMD,1619 which has improved patient management. In GA, the current classification based on FAF patterns predicts rates of growth. However, it was developed on subjective grounds11,20,21 and is complex (with low to moderate interobserver agreement),22 and histological or genetic correlation is still not available. Whether GA patterns are stable in a given individual across time is also currently unknown. 
Latent class analysis (LCA) is a statistical method that unravels the relationship between a group of observed variables and a set of unobserved (i.e., latent) variables.23 In this study, the latent class model connects the patterns or phenotypes on FAF described in the FAM Study (the observed variables, defined from the classification assigned by the observers) to a set of unobserved variables, which have to be defined afterward. Simply stated, LCA creates groups that have something in common that may not be readily apparent and that must be identified later on. An important principle of standard LCA modeling is that the classes emerging from the model are assumed to have a causal relationship with the observations. The new categories (the latent classes) arise from data and therefore are free of any subjective criteria. 
The purpose of this study was threefold: (1) to re-evaluate the classification of GA patterns as seen with FAF objectively, using LCA to see what patterns have in common, with the aim of attaining a simplified classification; (2) to determine if the new patterns (classes) change with time in a given individual; and (3) if these changes occur, to evaluate if they do so in a predictable manner (i.e., from one specific class to another). 
Methods
Subjects
The patients included in this study were participants in a prospective, longitudinal natural history study of subjects with GA, the GAIN (Characterization of GA progression in patients with AMD, NCT01694095). It has been conducted at the Institut de la Màcula i de la Retina (Hospital Quirón Teknon, Barcelona, Spain) since December 2009, and its main purpose is to identify risk factors for progression of already present GA. 
The particular study described herein included all participants in the GAIN cohort visited from December 2009 until February 2013 for whom at least two FAF images were available at least 1 year apart. If several images were available, those at baseline and at the last follow-up visit were selected. We included patients of either sex 50 years or older with GA secondary to AMD. Geographic atrophy was defined as a uni- or multifocal area of RPE atrophy on a 35° fundus photograph (TRC 50DX IA, IMAGEnet; Topcon Corporation, Tokyo, Japan) centered on the fovea; at least one of the atrophic lesions had to be larger than 0.5 disc areas (1.27 mm2). Both eyes of each patient were eligible for the study. Eyes were excluded if RPE atrophy was deemed to be secondary to other causes (macular dystrophy, high myopia, and so on); if there was a history of wet AMD or any other macular disease thought to interfere with interpretation of FAF images (i.e., diabetic retinopathy) in the study eye; if there was any history of laser treatment in the macular area, intravitreal injection, or intraocular surgery (excluding phacoemulsification) in the study eye; or if poor image quality precluded the interpretation of the FAF pattern. The study followed the tenets of the Declaration of Helsinki; it was approved by the local Ethics Committee, and all patients signed an informed consent after explanation of the nature and possible consequences of the study. 
Procedures
All patients underwent a complete ophthalmic examination that included best-corrected visual acuity, intraocular pressure, fundus biomicroscopy, and imaging after pupil dilation with one drop of 1.0% tropicamide and 10% phenylephrine. Fundus autofluorescence imaging (λ = 480 nm, approximate emission 500–700 nm) was acquired with Spectralis HTRA+OCT (Heidelberg Engineering, Heidelberg, Germany). Fluorescein angiography was performed if choroidal neovascularization was suspected in the study eye or according to the status of the fellow eye. For FAF, a certified operator captured high-resolution (1536 × 1536 pixels), 30° × 30° field of view images centered on the fovea with a minimum automatic real time (averaging) of 10 frames. In all cases, the area of atrophy was measured by a single observer (MB) using the Region Finder software, version 2.4.3.0 (Heidelberg Engineering). 
Five observers from three different centers (four retinal specialists and one optometrist), with experience in the use of FAF, independently assigned a pattern to each eye according to the Specific classification described in the FAM Study11 (with 10 categories), which automatically implied a precise pattern in the General and Simplified classifications (see Table 1). The agreement between the optometrist and three of the four retinal specialists had been previously shown to be similar to that among the retinal specialists between them.22 Baseline FAF images were exported to PowerPoint (Microsoft, Redmond, WA, USA) slides and were randomly presented to each observer, who determined the specific pattern for each eye according to the classification used in the FAM Study.11 Two to 3 weeks later, a second set of PowerPoint slides with the images of the last follow-up visit was sent to the observers for determination of the pattern. Review of results from the previous set of images was not allowed. Therefore, for each eye there were two independent assignments (baseline and follow-up) for each of the five observers. 
Statistical Analysis
After the classification made by each of the five observers on baseline and the last follow-up images using the FAM description, LCA analysis was used to determine the optimal number of classes in which the observations could be grouped. Afterward, the groups were described according to a series of a priori–defined characteristics of FAF images. A brief description of the methods follows, and a more detailed explanation is provided in the Supplementary Material
As an initial exploratory analysis, we used the intraclass correlation coefficient (ICC) to evaluate the consistency of the different classifications described in the FAM Study (Specific, General, and Simplified) between observers with a two-way random-effects model.24 
Then we used LCA to obtain a more stable classification system (see Supplementary Fig. S1). The emerging classes represent a categorical classification scheme that maximizes the probability of finding the distribution of responses provided by the observers. In our model, five observers assessed each patient with three classification systems (Specific, General, and Simplified) on two occasions (baseline and follow-up). The process is data driven, and no initial assumptions were made about the number of emerging classes. Different results involving two to seven latent classes were compared. The model with the best balance between lowest sample size–adjusted Bayesian information criterion (SSBIC, a criterion for model selection based on the likelihood function that penalizes for an increase in the number of variables) and highest entropy (a value between 0 and 1, where higher values represent better ability of the model to discriminate between classes) was chosen.25,26 
As a validation test we applied the LCA model to the follow-up data in order to obtain (new) classes at the time of the last visit. With the classification of observers at the follow-up time, we computed the concordance of the initial and final latent classifications by means of cross-tables and Cochrane's Q index.27 
Finally, to describe the characteristics of the resulting LCA-based classes we compared the age, sex, median size of atrophy, and growth distribution between each category using the Kruskal-Wallis test for continuous variables and the likelihood-ratio χ2 test for categorical variables. We also analyzed their FAF features according to a prespecified set of characteristics: the presence of hyperautofluorescence surrounding the area of atrophy (yes versus no), its distribution (elsewhere versus none or adjacent to the area of atrophy), the color of the atrophic area (grayish versus black), the number of atrophic lesions (more than one versus one), and their location in relation to the fovea (extrafoveal versus foveal). 
Univariate statistics were also provided to describe the characteristics of the patients included in the study. We used MPlus, version 6.10 (Muthén and Muthén, Los Angeles, CA, USA), and Stata IC, version 11.1 (StataCorp LP, College Station, TX, USA), for statistical analysis. A two-tailed P value < 0.05 was considered statistically significant. 
Results
We screened 171 subjects, 59 of whom (75 eyes) met the eligibility criteria and were included in the present study. There were 38 females (64.4%), with a median (mean) age of 79 (78.1) years old (interquartile range [IQR], 73–82; range, 52–97 years), 36 right eyes (48%), and a median (mean) follow-up of 19 (18.9) months (IQR, 15–22; range, 12–32 months). The median (mean) growth was 1.65 (1.59) mm2/year (IQR, 0.88–2.10; range, 0.05–4.14 mm2/y). All were Caucasian. 
The concordance of the different classification systems across the five observers showed values of ICC below 0.7 except for the Simplified category (0.99; see Supplementary Table S1). When agreement between classifications was analyzed (Supplementary Table S2), the Trickling category produced the highest concordance between the different categories, while the Undetermined had the lowest. The None and Focal categories were classified in the Diffuse category in many instances, while the Patchy category was seldom used. 
Latent class analyses were consistent in extracting a five-category system (five classes) with all original classifications. The five-category classification minimized SSBIC and obtained excellent relative entropies (≥0.97; see Supplementary Table S3). The agreement between each original classification and the emerging latent classes varied. Class agreement between the Specific and General systems was 0.64. The agreement between Specific and Simplified was lower (0.40), and so was that between General and Simplified (0.38), which is an expected result as we were obtaining more latent classes than categories in the Simplified classification, thus producing cross-classification errors. Table 2 shows the distribution of patterns according to the General classification in the FAM Study allocated to each new class resulting from LCA. 
Table 2
 
Percentage of Cases in the Original Classification Allocated to Each New Class
Table 2
 
Percentage of Cases in the Original Classification Allocated to Each New Class
None Focal Banded Diffuse Patchy Undetermined
1 53.8 7.7 7.7 30.8 0.0 0.0
2 4.8 38.2 7.1 35.7 7.1 7.1
3 4.8 33.3 14.3 47.6 0.0 0.0
4 2.8 13.9 8.3 63.9 0.0 11.1
5 0.0 3.5 58.5 34.5 0.0 3.5
There was high agreement between the classification at baseline and at the follow-up examination. The class of an individual did not change with time, as there was no significant change in the assigned class categories on both occasions as measured by Cochrane's Q (Specific: Q = 0.35, P = 0.55; General: Q = 0.15, P = 0.70; Simplified: Q = 0.50, P = 0.06). The Simplified system showed the highest value of the Q statistic, indicating a higher number of discrepancies. This was expected due to the previous results on the loss of information and nonspecificity of the Simplified system. 
To further validate the results, we used a longitudinal model to analyze growth of GA by latent class (see Supplementary Fig. S2). Baseline area of atrophy, time, and latent class predicted the area of atrophy at the last follow-up visit (all P < 0.001). Rate of growth was lower for class 1 and relatively similar for classes 2 to 5 as shown by their similar slopes. 
Table 3 shows the demographic and FAF imaging characteristics of each class derived from LCA. There were statistically significant differences in baseline area of atrophy (with progressively larger areas from class 1 to 5), annual growth (patients in class 1 experienced the smallest enlargement), presence of hyperautofluorescence (again, class 1 included the smallest percentage of patients with increased FAF), its location (most patients with class 4 showed diffuse FAF while patients in class 1 did not show FAF beyond atrophy), and the percentage of patients with multifocal lesions (classes 3 and 4 had many patients with this feature, while unifocal lesions were most common in class 1). 
Table 3
 
Characteristics of Each Class Based on Demographic and Autofluorescence Features
Table 3
 
Characteristics of Each Class Based on Demographic and Autofluorescence Features
1,n = 7, 9.3% 2,n = 18, 24.0% 3,n = 10, 13.3% 4,n = 23, 30.7% 5,n = 17, 22.7% P Value
Age, y 81 75 79 80 80 0.49
Sex, % females 57.1 55.6 70 65.2 70.6 0.88
Baseline area, mm2 1.92 2.75 4.58 10.73 12.7 <0.001
Growth, mm2/y 0.43 1.11 1.75 1.77 1.79 0.002
FAF*, %
 AF+ 14.3 77.8 90.0 95.7 100 <0.001
 AF elsewhere 0.0 38.9 50.0 69.6 29.4 0.003
 Gray color 0.0 27.8 20.0 43.5 29.4 0.12
 Multifocal 28.6 50.0 80.0 87.0 58.8 0.01
 Extrafoveal 14.3 38.9 50.0 60.9 35.3 0.17
From the values shown in Table 3, one may attempt to define major features for each class (see Fig.): 
Figure.
 
Representative fundus autofluorescence images of each class derived from LCA. Four examples of class 1 (top row), 2 (second row), 3 (third row), 4 (fourth row), and 5 (fifth row). Some common features could be identified as predominant in each class, such as lesion size, despite some heterogeneity. The last column shows examples of cases in each class whose characteristics are not consistent with the main features of the class to which they belong (see also Supplementary Material).
Figure.
 
Representative fundus autofluorescence images of each class derived from LCA. Four examples of class 1 (top row), 2 (second row), 3 (third row), 4 (fourth row), and 5 (fifth row). Some common features could be identified as predominant in each class, such as lesion size, despite some heterogeneity. The last column shows examples of cases in each class whose characteristics are not consistent with the main features of the class to which they belong (see also Supplementary Material).
  •  
    Class 1: small areas of foveal atrophy without increased FAF;
  •  
    Class 2: small to moderate areas of atrophy with increased FAF, usually juxta-atrophic;
  •  
    Class 3: moderate areas of atrophy with increased FAF, predominantly multifocal;
  •  
    Class 4: large areas of atrophy, predominantly multifocal, with increased FAF elsewhere; and
  •  
    Class 5: large areas of atrophy with increased juxta-atrophic FAF.
Discussion
Using the FAF classification from the FAM Study11 and a model based on LCA, we found that the original classification could be summarized in just five categories (classes). These classes were closely related to different variables, notably area of atrophy, and they did not change within individuals after a median follow-up of 1.5 years. The classification was robust, as shown by the consistent results found using different original classifications, different time points (baseline and follow-up), and entropy (excellent ability of the model to discriminate between classes). 
The emerging classes were different in terms of area of atrophy, GA growth, percentage of cases with increased FAF and FAF elsewhere, and percentage of subjects with multifocal lesions (P < 0.01 each). However, the characteristic that more clearly differentiated between classes was area of atrophy. 
Fundus autofluorescence patterns were shown to be associated with rates of growth in the FAM Study,11 but if they were true phenotypes they should be expected to be independent of GA size. The fact that they are closely related in the present study suggests that rather than being true disease phenotypes, FAF patterns reflect different stages of the disease, different periods along the continuum of disease enlargement. Here, we provide a possible explanation for the changes in the distribution of high FAF (clinically seen as GA “phenotypes”) with disease enlargement. 
It has been shown that there is mobilization of RPE cells at the borders of atrophy.28 Rudolf et al.29 found that RPE cells at the junction between healthy and diseased retina were morphologically abnormal and that increased FAF was caused by disorganized, vertically superimposed lipofuscin-containing RPE rather than excessive lipofuscin accumulation within individual cells. On the other hand, the physiological distribution of increased FAF in the fundus in healthy subjects increases progressively with increasing eccentricity, reaching a maximum at approximately 10° (3 mm from the foveal center) and decreasing thereafter. Whatever the mechanism causing GA, areas of atrophy in the fovea mobilize RPE cells at the junction whose FAF is mostly masked by luteal pigment, creating a distribution of FAF that is clinically identified as patterns with low FAF (mostly None or Focal). As GA enlarges, it eventually reaches perifoveal areas of the macula, where the vertical superimposition of these altered, physiologically highly autofluorescent RPE cells would induce FAF patterns clinically characterized by extensive FAF (patterns Banded or Diffuse). In this situation, elevated FAF would be a consequence of atrophy enlargement, not a cause of it. This mechanism indirectly calls into question the central role of lipofuscin accumulation in GA pathogenesis, as other histologic and imaging studies have pointed out.6,7,29,30 However, research designed specifically to address this issue will respond to this question more adequately. 
We did not detect any changes in GA class over a period of 1.5 years, but this time frame may be insufficient to study the transitions of a given patient from one class to another as lesion grows. In fact, Fujinami et al.31 found predictable transitions between FAF subtypes (from subtype 1 to 2, and from subtype 2 to 3) in patients with Stargardt disease, a phenotypically similar disorder to GA, after a mean follow-up of 9.1 years. This suggests that a longer follow-up may reveal transitions between classes in GA, as would be expected if our hypothesis is correct. 
Since the results of this study suggest that the distribution of FAF accompanies different areas of atrophy and therefore different stages along a continuum of the disease, no new classification is proposed. Despite our findings, the FAM classification11 remains a useful reference to provide individual prognosis, since it can stratify patients by their predicted progression. In fact, the relationship between FAF patterns and GA growth as evaluated by each of the five observers in the present study was similar to that reported in the FAM Study (see Supplementary Fig. S3), minimizing the possibility of image misclassification as a cause of our results. 
The LCA method allows the identification of unobserved features behind a set of multivariable measured variables, which often show an intricate interrelationship making the assessment of their ultimate effect difficult. Despite its potential, this method is used infrequently in the field of ophthalmology,3236 but it offers interesting applications. Other approaches, such as cluster analysis, have also been used to unmask the characteristics that define groups of individuals.37 
One of the limitations of the current study is the relative small sample size (75 eyes), which may cause overfitting of the LCA model. We cannot exclude the possibility that there are true GA phenotypes with low prevalence, which may have been diluted within the classes described herein. Longer follow-up would be needed to evaluate how the change in area affects class adjudication in any given subject. Therefore, these results are preliminary and should be validated by an independent research group with different observers, a different sample of patients with GA, and longer follow-up. Finally, there are other potential sources of increased FAF aside from lipofuscin build-up within the RPE38 (impaired photoreceptor cells, thinned retina, and so on); a better understanding of FAF dynamics could help to differentiate between them. 
In summary, using a data-driven approach, LCA, we found that FAF patterns are closely related to area of atrophy in patients with GA secondary to AMD. This relationship suggests that FAF patterns reflect different stages of the disease and call into question the existence of many phenotypes in GA and also the role of lipofuscin in disease progression. The emerging classes do not replace the existing classification for prognostic purposes. 
Acknowledgments
The authors thank Fabio Trindade, MD, PhD, and Miguel Zapata, MD, PhD, for their collaboration in the determination of fundus autofluorescence patterns. 
Disclosure: M. Biarnés, None; C.G. Forero, None; L. Arias, None; J. Alonso, None; J. Monés, Alcon (F, R), Allergan (F, R), Allimera (F), Bayer (F, R), Novartis (F, R), Ophthotech (F) 
References
Fleckenstein M Schmitz-Valckenberg S Sunness J Holz F. Geographic atrophy. In: Holz F Pauleikhoff D Spaide R Bird A eds. Age-Related Macular Degeneration. 2nd ed. Berlin: Springer; 2013: 121–138.
McLeod DS Grebe R Bhutto I Merges C Baba T Lutty GA. Relationship between RPE and choriocapillaris in age-related macular degeneration. Invest Ophthalmol Vis Sci. 2009; 50: 4982–4991. [CrossRef] [PubMed]
Ambati J Anand A Fernandez S An animal model of age-related macular degeneration in senescent Ccl-2- or Ccr-2-deficient mice. Nat Med. 2003; 9: 1390–1397. [CrossRef] [PubMed]
Zhou J Kim SR Westlund BS Sparrow JR. Complement activation by bisretinoid constituents of RPE lipofuscin. Invest Ophthalmol Vis Sci. 2009; 50: 1392–1399. [CrossRef] [PubMed]
Sparrow JR Cai B. Blue light-induced apoptosis of A2E-containing RPE: involvement of caspase-3 and protection by Bcl-2. Invest Ophthalmol Vis Sci. 2001; 42: 1356–1362. [PubMed]
Ablonczy Z Higbee D Anderson DM Lack of correlation between the spatial distribution of A2E and lipofuscin fluorescence in the human retinal pigment epithelium. Invest Ophthalmol Vis Sci. 2013; 54: 5535–5542. [CrossRef] [PubMed]
Smith RT Bernstein PS Curcio CA. Rethinking A2E. Invest Ophthalmol Vis Sci. 2013; 54: 5543. [CrossRef] [PubMed]
Sparrow JR Dowling JE Bok D. Understanding RPE lipofuscin. Invest Ophthalmol Vis Sci. 2013; 54: 8325–8326. [CrossRef] [PubMed]
Delori FC Dorey CK Staurenghi G Arend O Goger DG Weiter JJ. In vivo fluorescence of the ocular fundus exhibits retinal pigment epithelium lipofuscin characteristics. Invest Ophthalmol Vis Sci. 1995; 36: 718–729. [PubMed]
Holz FG Bellman C Staudt S Schutt F Volcker HE. Fundus autofluorescence and development of geographic atrophy in age-related macular degeneration. Invest Ophthalmol Vis Sci. 2001; 42: 1051–1056. [PubMed]
Holz FG Bindewald-Wittich A Fleckenstein M Progression of geographic atrophy and impact of fundus autofluorescence patterns in age-related macular degeneration. Am J Ophthalmol. 2007; 143: 463–472. [CrossRef] [PubMed]
Rosenfeld PJ Legarreta J. Preclinical and phase I drugs in development for dry AMD: an overview. Retinal Physician. November 1, 2009.
Miller JW. Age-related macular degeneration revisited—piecing the puzzle: the LXIX Edward Jackson memorial lecture. Am J Ophthalmol. 2013; 155: 1–35, e13. [CrossRef] [PubMed]
Riley RD Hayden JA Steyerberg EW Prognosis Research Strategy (PROGRESS) 2: prognostic factor research. PLoS Med. 2013; 10: e1001380. [CrossRef] [PubMed]
Trusheim MR Berndt ER Douglas FL. Stratified medicine: strategic and economic implications of combining drugs and clinical biomarkers. Nat Rev Drug Discov. 2007; 6: 287–293. [CrossRef] [PubMed]
Lafaut BA Bartz-Schmidt KU Vanden Broecke C Aisenbrey S De Laey JJ Heimann K. Clinicopathological correlation in exudative age related macular degeneration: histological differentiation between classic and occult choroidal neovascularisation. Br J Ophthalmol. 2000; 84: 239–243. [CrossRef] [PubMed]
Lafaut BA Aisenbrey S Vanden Broecke C Bartz-Schmidt KU. Clinicopathological correlation of deep retinal vascular anomalous complex in age related macular degeneration. Br J Ophthalmol. 2000; 84: 1269–1274. [CrossRef] [PubMed]
Freund KB Zweifel SA Engelbert M. Do we need a new classification for choroidal neovascularization in age-related macular degeneration? Retina. 2010; 30: 1333–1349. [CrossRef] [PubMed]
Couch SM Bakri SJ. Review of combination therapies for neovascular age-related macular degeneration. Semin Ophthalmol. 2011; 26: 114–120. [CrossRef] [PubMed]
Holz FG Bellmann C Margaritidis M Schütt F Otto TP Völcker HE. Patterns of increased in vivo fundus autofluorescence in the junctional zone of geographic atrophy of the retinal pigment epithelium associated with age-related macular degeneration. Graefes Arch Clin Exp Ophthalmol. 1999; 237: 145–152. [CrossRef] [PubMed]
Bindewald A Schmitz-Valckenberg S Jorzik JJ Classification of abnormal fundus autofluorescence patterns in the junctional zone of geographic atrophy in patients with age related macular degeneration. Br J Ophthalmol. 2005; 89: 874–878. [CrossRef] [PubMed]
Biarnés M Monés J Trindade F Alonso J Arias L. Intra and interobserver agreement in the classification of fundus autofluorescence patterns in geographic atrophy secondary to age-related macular degeneration. Graefes Arch Clin Exp Ophthalmol. 2012; 250: 485–490. [CrossRef] [PubMed]
Formann AK Kohlmann T. Latent class analysis in medical research. Stat Methods Med Res. 1996; 5: 179–211. [CrossRef] [PubMed]
Shrout PE Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull. 1979; 86: 420–428. [CrossRef] [PubMed]
Akaike H. On entropy maximization principle. In: Krishnaiah PR ed. Applications of Statistics. Amsterdam: North-Holland; 1977: 27–41.
Schwarz GE. Estimating the dimension of a model. Ann Stat. 1978; 6: 461–464. [CrossRef]
Conover WJ. Practical Nonparametric Statistics. 3rd ed. New York: Wiley; 1999.
Bird AC Phillips RL Hageman GS. Geographic atrophy: a histopathological assessment. JAMA Ophthalmol. 2014; 132: 338–345. [CrossRef] [PubMed]
Rudolf M Vogt SD Curcio CA Histologic basis of variations in retinal pigment epithelium autofluorescence in eyes with geographic atrophy. Ophthalmology. 2013; 120: 821–828. [CrossRef] [PubMed]
Hwang JC Chan JWK Chang S Smith RT. Predictive value of fundus autofluorescence for development of geographic atrophy in age-related macular degeneration. Invest Ophthalmol Vis Sci. 2006; 47: 2655–2661. [CrossRef] [PubMed]
Fujinami K Lois N Mukherjee R A longitudinal study of Stargardt disease: quantitative assessment of fundus autofluorescence, progression, and genotype correlations. Invest Ophthalmol Vis Sci. 2013; 54: 8181–8190. [CrossRef] [PubMed]
Keenan JD See CW Moncada J Diagnostic characteristics of tests for ocular Chlamydia after mass azithromycin distributions. Invest Ophthalmol Vis Sci. 2012; 53: 235–240. [CrossRef] [PubMed]
Aspinall PA Johnson ZK Azuara-Blanco A Montarzino A Brice R Vickers A. Evaluation of quality of life and priorities of patients with glaucoma. Invest Ophthalmol Vis Sci. 2008; 49: 1907–1915. [CrossRef] [PubMed]
Gao F Miller JP Miglior S The effect of changes in intraocular pressure on the risk of primary open-angle glaucoma in patients with ocular hypertension: an application of latent class analysis. BMC Med Res Methodol. 2012; 12: 151. [CrossRef] [PubMed]
Niederkohr RD Levin LA. A Bayesian analysis of the true sensitivity of a temporal artery biopsy. Invest Ophthalmol Vis Sci. 2007; 48: 675–680. [CrossRef] [PubMed]
Bandeen-Roche K Muñoz B Tielsch JM West SK Schein OD. Self-reported assessment of dry eye in a population-based setting. Invest Ophthalmol Vis Sci. 1997; 38: 2469–2475. [PubMed]
Nunes S Ribeiro L Lobo C Cunha-Vaz J. Three different phenotypes of mild nonproliferative diabetic retinopathy with different risks for development of clinically significant macular edema. Invest Ophthalmol Vis Sci. 2013; 54: 4595–4604. [CrossRef] [PubMed]
Sparrow JR Yoon KD Wu Y Yamamoto K. Interpretations of fundus autofluorescence from studies of the bisretinoids of the retina. Invest Ophthalmol Vis Sci. 2010; 51: 4351–4357. [CrossRef] [PubMed]
Footnotes
 MB and CGF contributed equally to the work presented here and should therefore be regarded as equivalent authors.
Figure.
 
Representative fundus autofluorescence images of each class derived from LCA. Four examples of class 1 (top row), 2 (second row), 3 (third row), 4 (fourth row), and 5 (fifth row). Some common features could be identified as predominant in each class, such as lesion size, despite some heterogeneity. The last column shows examples of cases in each class whose characteristics are not consistent with the main features of the class to which they belong (see also Supplementary Material).
Figure.
 
Representative fundus autofluorescence images of each class derived from LCA. Four examples of class 1 (top row), 2 (second row), 3 (third row), 4 (fourth row), and 5 (fifth row). Some common features could be identified as predominant in each class, such as lesion size, despite some heterogeneity. The last column shows examples of cases in each class whose characteristics are not consistent with the main features of the class to which they belong (see also Supplementary Material).
Table 1
 
Geographic Atrophy Patterns as Originally Described With Fundus Autofluorescence in the Fundus Autofluorescence in Age-related Macular Degeneration Study11
Table 1
 
Geographic Atrophy Patterns as Originally Described With Fundus Autofluorescence in the Fundus Autofluorescence in Age-related Macular Degeneration Study11
Specific Classification General Classification Simplified Classification
None None None + Focal
Focal Focal None + Focal
Banded Banded Banded + Diffuse
Branching Diffuse Banded + Diffuse
Fine granular Diffuse Banded + Diffuse
Fine granular with peripheral punctate spots Diffuse Banded + Diffuse
Reticular Diffuse Banded + Diffuse
Trickling Diffuse Banded + Diffuse
Patchy Patchy Patchy + Undetermined
Undetermined Undetermined
Table 2
 
Percentage of Cases in the Original Classification Allocated to Each New Class
Table 2
 
Percentage of Cases in the Original Classification Allocated to Each New Class
None Focal Banded Diffuse Patchy Undetermined
1 53.8 7.7 7.7 30.8 0.0 0.0
2 4.8 38.2 7.1 35.7 7.1 7.1
3 4.8 33.3 14.3 47.6 0.0 0.0
4 2.8 13.9 8.3 63.9 0.0 11.1
5 0.0 3.5 58.5 34.5 0.0 3.5
Table 3
 
Characteristics of Each Class Based on Demographic and Autofluorescence Features
Table 3
 
Characteristics of Each Class Based on Demographic and Autofluorescence Features
1,n = 7, 9.3% 2,n = 18, 24.0% 3,n = 10, 13.3% 4,n = 23, 30.7% 5,n = 17, 22.7% P Value
Age, y 81 75 79 80 80 0.49
Sex, % females 57.1 55.6 70 65.2 70.6 0.88
Baseline area, mm2 1.92 2.75 4.58 10.73 12.7 <0.001
Growth, mm2/y 0.43 1.11 1.75 1.77 1.79 0.002
FAF*, %
 AF+ 14.3 77.8 90.0 95.7 100 <0.001
 AF elsewhere 0.0 38.9 50.0 69.6 29.4 0.003
 Gray color 0.0 27.8 20.0 43.5 29.4 0.12
 Multifocal 28.6 50.0 80.0 87.0 58.8 0.01
 Extrafoveal 14.3 38.9 50.0 60.9 35.3 0.17
Spanish Abstract
Supplementary Tables and Figure Information
Supplementary Figures
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×