Ocular Risk Factors for Exudative AMD: A Novel Semiautomated Grading System

Purpose. To evaluate the contribution of the ocular risk factors in the conversion of the fellow eye of patients with unilateral exudative AMD, using a novel semiautomated grading system. Materials and Methods. Single-center, retrospective study including 89 consecutive patients with unilateral exudative AMD and ≥3 years of followup. Baseline color fundus photographs were graded using an innovative grading software, RetmarkerAMD (Critical Health SA). Results. The follow-up period was 60.9 ± 31.3 months. The occurrence of CNV was confirmed in 42 eyes (47.2%). The cumulative incidence of CNV was 23.6% at 2 years, 33.7% at 3 years, 39.3% at 5 years, and 47.2% at 10 years, with a mean annual incidence of 12.0% (95% CI = 0.088–0.162). The absolute number of drusen in the central 1000 and 3000 μm (P < 0.05) and the absolute number of drusen ≥125 µm in the central 3000 and 6000 µm (P < 0.05) proved to be significant risk factors for CNV. Conclusion. The use of quantitative variables in the determination of the OR of developing CNV allowed the establishment of significant risk factors for neovascularization. The long follow-up period and the innovative methodology reinforce the value of our results. This trial is registered with ClinicalTrials.gov NCT00801541.


Introduction
Age-related macular degeneration (AMD) poses a substantial public health problem worldwide [1][2][3][4]. The exudative form is the leading cause of irreversible vision loss in subjects over 65 years of age living in economically developed countries and accounts for >80% of cases of legal blindness associated with the disease [5][6][7]. Risk estimation is crucial in order to provide adequate patient monitoring. Fellow eyes of individuals with unilateral exudative AMD should receive the utmost attention since they are known to have an increased risk of developing choroidal neovascularization (CNV) [4,5,8,9].
The role of retinal imaging in AMD has long been recognized. Several studies have concluded that digital imaging is reliable for the purpose of grading AMD, both in clinical practice and in clinical trials [10,11]. The International Classification and Grading System [12] was created to provide a uniform and internationally accepted nomenclature and grading system for the disease, based on color fundus photography's morphological findings. Implementation of this system in large cohort epidemiological studies helped establishing significant determinants for the development of wet AMD [4,[13][14][15] and led to the creation of a fundus photography-based severity scale by the AREDS group [16]. This simplified scale uses a point scoring system and estimates the 5-year risk of developing advanced AMD. The three variables included are the presence/absence of (1) large drusen (≥125 m), (2) pigmentary changes (hypo-or hyperpigmentation) and (3) neovascular AMD. The major goals of risk estimation are to provide baseline risk categories, to allow tracking of progression, and to define surrogate outcomes for progression to the advanced forms of the disease [17].
The importance of drusen characteristics (such as the area and number of drusen) in the development of wet AMD has been highlighted by several studies [13][14][15]18]. When grading 2 ISRN Ophthalmology is performed with conventional reading center techniques, the depiction of these variables is categorical in nature. Taking the number of drusen as an example, a simplified categorical assessment, as provided by the International Classification and Grading System for AMD [12], would grade one image having 0, 1-9, 10-19, or ≥20 drusen. Statistical analysis models using continuous variables (such as the real number of drusen or the real area of drusen) are generally more robust, hence providing far more reliable results [19].
The advent of computer-assisted grading systems represents a significant advance in retinal imaging. Not only do they make the grading process more efficient, but they also deliver faster data analysis when compared to the more laborintensive conventional reading center methods [18,[20][21][22].
Using a novel semiautomated grading software capable of categorical and continuous variable grouping (Retmark-erAMD), this study aims to evaluate the contribution of the ocular risk factors in the development of CNV in the fellow eyes of patients with unilateral exudative AMD.

Material and Methods
The retrospective, single-center study included 89 consecutive patients with unilateral exudative AMD. Patients with neovascular AMD in one eye (the nonstudy eye) and early age-related maculopathy in the fellow eye (study eye) were enrolled. Inclusion criteria were (a) age ≥ 50 years; (b) any race and either sex; (c) clinical diagnosis of wet AMD in one eye (non-study eye), and the presence of the following characteristics in the second eye (study eye): (1) ≥5 intermediate drusen (≥63 and <125 m), ≥1 large soft drusen ≥ 125 m, and/or confluent drusen, within the central 3,000 m region, and (2) with or without pigmentary changes; (d) ≥3 years of followup after development of CNV in the non-study eye; and (e) baseline color fundus pictures of the study eye at the time of conversion of the non-study eye. Exclusion criteria were (a) presence of other ophthalmologic diseases likely to jeopardize the grading of the color fundus pictures; (b) non-AMDrelated CNV in the non-study eye; (c) clinical or fundoscopic signs of myopic retinopathy or refractive power ≥8D; and (d) evidence of past or present CNV in the study eye.

Grading Fundus Color Photography.
Mydriatic color fundus digital photographs taken at the time of conversion of the non-study eye were used for grading. The pictures were centered on the fovea, with a 30-degree field of view. Grading took place at the Coimbra Ophthalmology Reading Center (CORC) by graders certified by this institution. For the certification process, an agreement between graders was developed as generalized kappa-type statistics, with a kappa value >0.61 required for approval [23].
Grading was performed using the RetmarkerAMD (Critical Health SA), a computer-assisted grading system that follows the guidelines of the International Classification and Grading System for AMD [12]. The system does not automatically detect lesions, but its components allow to (i) differentiate graded from nongraded images; (ii) assess images in full-screen view; (iii) draw free forms of all lesion types over the images; (iv) draw preset circular objects of different color codes and sizes (63 m, 125 m, 175 m, 250 m, or 500 m) to represent drusen, hyperpigmentation, hypopigmentation, geographic atrophy, and exudative lesions; (v) mark drusen as confluent; (vi) zoom in/out the image or display an image area with a higher/lower degree of magnification; (vii) measure distances in the image and control red/blue/green channels (RGB), brightness, and contrast; (viii) superimpose a standard grid to identify eye subfields, and (ix) manually enter patient data.
Each image is calibrated before any measurement is performed. The user manually indicates the optic disc vertical limits along with the center of the macula (fovea), as shown in Figure 1(a). An optic disc with diameter of 1.5 mm is used as reference by the software to set up an overlaying reference grid (diameter 6000 m), where measurements take place. The grid consists of 3 concentric circles and a right-angled cross at 45 ∘ and 135 ∘ to the horizontal, which is adjusted according to the previous calibration ( Figure 1(b)). As per the International Classification and Grading System [12], the diameters of the central, inner, and outer circles are 1000, 3000, and 6000 m, respectively.
Using this semi-automated grading system, the results are displayed on screen in real time, and the final characterization of soft distinct, indistinct, reticular, or crystalline drusen is performed by the operator.
Preliminary results from comparative tests held at our institution have shown that RetmarkerAMD is more effective than 35 mm film manual grading [20]. Aside from being timesaving (overall 35% faster), the software has also proved to be more accurate (overall 32% more lesions identified) [Silva R., oral presentation, EURETINA 2010].

Statistical
Analysis. Data were characterized with descriptive statistics and explored with classification trees using CRT algorithm. Differences between eyes that converted and those who did not were assessed using the Pearson 2 test and Fisher's exact test for nominal variables and the Student's t-test and Mann-Whitney U test for scalar variables. Cumulative hazards for conversion times were calculated. Multivariate logistic regressions were used to calculate the odds ratio, expressing the effects of all of the tested explanatory factors for eye conversion in two models: (1) considering only large drusen (≥125 m) and (2) considering all drusen. The following parameters were considered in the multivariate analysis: real number of drusen, real area occupied by drusen, presence of hyperpigmentation, and age (a) (b) (c) Figure 1: Print screens of real-time grading using RetmarkerAMD. (a) Calibration is achieved after manually identifying the fovea and establishing optic disc diameter (blue dot and arrow, resp.). The software then generates a reference grid (b) that follows the International Classification and Grading System for AMD [12].

Results
Eighty-nine subjects were enrolled in the study. The study eye was the right eye in 47 individuals (52.8%) and the left eye in 42 (47.2%) individuals. The patients were aged between 55 and 92 years (mean ± SD: 74.2 ± 6.86 years), with a slight female predominance-58.4% ( = 52). The average age at conversion of the study eye was 77.5 ± 6.16, years and the time to conversion ranged between 12 and 113 months (median 28.5; IQR 24-43). The follow-up period was 60.9 ± 31.3 months (median 37; IQR 36-74). The occurrence of CNV was confirmed in 42 (47.2%) eyes during followup ( Table 1). The progression of the cumulative risk of conversion is shown in Figure 2. The incidence of CNV, calculated for the follow-up period, was 12.0%/year (95% CI = 0.088-0.162).
In order to increase the robustness of the analysis, we used continuous variables: the real (absolute) number and area occupied by drusen (Table 3). We tested these variables for the totality of drusen and for large drusen (≥125 m) only and      Tables 5 and 6

Discussion
Prior studies (MPS [14,15], AREDS [16]) comprising grading and risk estimation based on morphological assessments of baseline color fundus photography used manual grading and had their grading data grouped according to the guidelines of the International Classification and Grading System for AMD [12]. As such, depiction of variables such as the number (0, 1-9, 0-19, ≥20) and area occupied by drusen (<1% <10%, <25%, <50%, ≥50%) was categorical in nature, which inevitably produces some estimates that can be a source of bias. Although still not widely used in clinical or research settings, the development of semi-automated image analysis systems can significantly reduce the time and expense involved in manual grading. The use of innovative software in digital color fundus photography grading is changing the way reading centers work around the world, as it represents a convenient and versatile way of imaging and grading fundus changes. Friberg et al. [19] used a computer-based algorithm to detect and characterize drusen on digitized images and showed that quantitative detection of drusen can be performed reproducibly and efficiently using this method [19]. When compared to more labor intensive reading center techniques, the results were similar but the algorithm showed higher precision and accuracy [19].
RetmarkerAMD is a revolutionary platform with a userfriendly interface built upon the International Classification and Grading System for AMD's guidelines [12]. The software has already been validated for quantification of AMD features [Simão, S. et al., poster presentation, ARVO 2011] and is currently being used in an ongoing epidemiologic study of AMD in the Portuguese population with an estimated number of 4000 participants (NCT01298674). When compared to manual grading, the software was more effective and less time consuming (nearly 35% faster), allowing identification of 32% more lesions and reducing human error [Silva R.,  oral presentation, EURETINA 2010]. One must not forget, however, that, even though the advantages of using computerassisted grading systems are many, it may also be a source of bias. One example would be image calibration with the generation of the overlaying grid based on a reference optic disc with a diameter of 1.5 mm. Although this estimation of the optic disc diameter has been used in several studies, it does not account for the few cases where anatomic variants of the optic disc are present and may thus constitute a weakness of the software. RetmarkerAMD produces standardized and detailed records in Excel that allows easy data mining and statistical analysis. Variables can be grouped both in a categorical and in a continuous manner. Since statistical models operating with quantitative variables provide more reliable results [19], we took advantage of this unique ability to explore the effect of drusen characteristics in the conversion of the fellow eye. Our study combines the advantages of this innovative and validated semi-automated software with a long-term followup of a cohort consisting purely of fellow eyes of patients with unilateral AMD. With this approach we meant to eliminate the most consistent risk factor for CNV [17,18], consequently enabling better understanding of the role of drusen characteristics in the development of wet AMD.
In order to clarify the role of each determinant independently, we conducted multivariate logistic regressions and calculated the odds ratio (OR) for each covariate. We found that both the real number of (total) drusen (in the central 1000 and 3000 m) and the real number of (large) drusen ≥ 125 m (in the central 3000 and 6000 m) were significant risk factors for the occurrence of neovascular events ( < 0.05). However, the real area occupied by drusen ≥ 125 m proved to be inversely correlated with the development of CNV ( < 0.05). The AREDS group [16] stated that the number of large drusen (≥125 m) is closely related to the total area of drusen and that the latter is a significant risk factor for wet AMD. In a study conducted by Friberg et al. [21], aiming to determine the relationship between the number of drusen and drusen area, the authors found that the number of large drusen does not correlate better with total drusen area than drusen of other sizes, meaning that the number of large drusen is not necessarily a good surrogate for total drusen area. The same authors went further in another study [18] where digital images of 949 eyes of patients included in AREDS [16] and PTAMD [28] studies were retrospectively evaluated by computerized methods, and the variables were analyzed in a continuous manner. The author concluded that total drusen area was not a consistent risk factor for the development of CNV, and although the probability of a neovascular event would increase with increasing drusen area initially, the risk would then invert as a total drusen area of ∼0.75 mm 2 was reached (∼60 large drusen) [18]. This finding might justify our results. One possible explanation would be that the eyes with higher drusen area develop geographic atrophy (GA) instead of CNV. In CAPT [29], even though there was no association between any of the drusen measurements and the development of CNV, a correlation was found between drusen area and the development of GA.
Like drusen size and area, pigmentary changes have also been recognized previously as a risk factor for progression to wet AMD, both in individuals free of advanced AMD bilaterally and in the fellow eye of subjects with unilateral exudative AMD [15,16,29]. The presence of hyperpigmentation is one of the risk factors associated with CNV determined by the MPS group [15] and is also part of the AREDS simplified severity scale [17]. In our study, the presence of hyperpigmentation was inversely correlated with the development of exudative AMD in both models of multivariate analysis. This finding was unexpected. Friberg et al. [18] found that this determinant was not a consistent risk factor for an eye's development of CNV, as the location of the hyperpigmentation would influence its effect. In CAPT [29], hyperpigmentation was associated both with CNV and GA. As it represents degeneration of the retinal pigment epithelium (RPE) and a relatively advanced stage of drusen evolution, it is possible that either CNV or GA can follow its appearance. Since we have not looked for the development of GA in our cohort, we cannot correlate both. The difficulties detecting hyperpigmentation and the relatively few patients where we found it ( = 22) may have influenced our results.
The incidence of CNV in the fellow eyes of subjects with unilateral exudative AMD is known to be high. In our study there was a conversion rate of 12%/year (95% CI 0.088 to 0.162), a result consistent with other studies where it ranges from 6 to 12%/year [15,25,[30][31][32].
From the 89 patients enrolled, 42 (47.2%) developed CNV during the follow-up period (an average of 5.1 years). This number is higher than the ones reported by the extrafoveal trial of the MPS group [14] (25.7% in a cohort of 128 eyes with a 5-year followup) or the SST group [25] (26.9% in a cohort of 364 eyes with a 4-year followup). The long-term followup of our study (ranging from 36 to 148 months) may justify some of this discrepancy. When comparing our results with the AREDS subgroup with unilateral exudative AMD, where the mean followup was 6.3 years, the disparity is less prominent (38.9% in a cohort of 714 eyes).
The effect of systemic (smoking history, hypertension, body mass index, and dietary habits) or genetic factors in the conversion of the fellow eye was not accounted for in our study. Although these might have had some impact, apart from the smoking history, the influence of these determinants in the development of exudative AMD is substantially less important than the contribution of the ocular risk factors [13].
The cumulative incidence of CNV was 7.9% at the 1st year, 23.6% at 2 years, 33.7% at 3 years, 39.3% at 5 years, and 47.2% at 10 years. These results are identical to those identified in other large cohort studies [14,15,[24][25][26][27], summarized in Table 7. Opposing to the studies of Caucasian populations, it is noteworthy to sign that, among Asians [27], the cumulative incidence of CNV in the fellow eye is considerably lower.
In conclusion, RetmarkerAMD allowed the use of quantitative variables in the determination of the OR of developing CNV, thus establishing the real (absolute) number of drusen (in the central 1000 and 3000 m) and the real number of drusen ≥ 125 m (in the central 3000 and 6000 m) as significant risk factors for the conversion of the fellow eye. The long follow-up period and the use of an innovative and validated semi-automated grading software reinforce the value of our results and support the influence of the ocular risk factors in the progression of the fellow eye to exudative AMD.