Biomarkers that predict future visual acuity (VA) in eyes with baseline diabetic macular edema (DME) would substantively improve risk assessment, management decisions, and selection of eyes for clinical studies targeting DME.
To determine whether baseline or early change in the novel spectral domain–optical coherence tomography (SD-OCT) parameter disorganization of the retinal inner layers (DRIL) is predictive of VA in eyes with center-involved DME.
Design, Setting, and Participants
At a tertiary care referral center for diabetic eye disease, a retrospective, longitudinal cohort study obtained demographics, VA, and SD-OCT images from baseline, 4-month, and 8-month visits in 96 participants (120 eyes) with diabetes mellitus and baseline center-involved DME (SD-OCT central subfield thickness, ≥320 µm for men and ≥305 µm for women). Exclusion criteria included substantial media opacity, cataract surgery within 6 months, and nondiabetic retinal pathology affecting VA. On SD-OCT, the 1-mm-wide retinal area centered on the fovea was evaluated by masked graders for DRIL extent, cysts, hyperreflective foci, microaneurysms, cone outer segment tip visibility, and external limiting membrane or photoreceptor disruption and reflectivity.
Main Outcomes and Measures
Visual acuity and SD-OCT–derived retinal morphology.
Greater DRIL extent at baseline correlated with worse baseline VA (point estimate, 0.04; 95% CI, 0.02-0.05 per 100 µm; P < .001). An increase in DRIL during 4 months was associated with VA worsening at 8 months (point estimate, 0.03; 95% CI, 0.02-0.05 per 100 µm; P < .001). A multivariate model that included a 4-month change in VA, DRIL, and external limiting membrane disruption was predictive of an 8-month VA change (r = 0.80). Each approximately 300-µm DRIL increase during 4 months predicted a 1-line, 8-month VA decline. When DRIL increased at least 250 µm at 4 months, no eyes had VA improvement of at least 1 line at 8 months. When DRIL decreased at least 250 µm at 4 months, no eyes had VA decline of at least 1 line at 8 months, and 77.7% had VA improvement of at least 1 line.
Conclusions and Relevance
Disorganization of the retinal inner layers in the 1-mm foveal area is associated with VA, and change in DRIL predicts future change in VA. Early change in DRIL prospectively identifies eyes with a high likelihood of subsequent VA improvement or decline. Therefore, DRIL warrants further study as a robust, readily obtained, and noninvasive biomarker of future VA response in eyes with DME.
Diabetic macular edema (DME) is a sight-threatening manifestation of diabetic retinopathy, affecting almost 30% of individuals with more than 20 years of diabetes mellitus.1 Standard treatment for DME involves repetitive, invasive intraocular injections, which place heavy burdens on the patient, physician, and health care reimbursement. No reliable methods exist to determine which individuals with DME will gain or lose vision over time, making such predictive biomarkers a major unmet need.2 These tools would substantively enhance patient counseling, improve risk stratification, advance clinical management, and influence selection of eyes for clinical studies targeting DME.
The noninvasive, readily performed imaging modality of spectral domain–optical coherence tomography (SD-OCT) provides reliable, high-resolution imaging of retinal anatomy and quantification of central retinal thickness. Previous investigations using OCT demonstrated that central retinal thickness is only modestly correlated with current visual acuity (VA) or change in VA.3 Various other SD-OCT anatomic findings have been studied, but these have not demonstrated adequate correlation to be useful as reliable predictors of VA in individual eyes with DME.4-9
Our group has investigated disorganization of the retinal inner layers (DRIL) as a surrogate marker that seems to be correlated with current VA in individuals with existing or resolved center-involved DME.10 Disorganization of the retinal inner layers affecting 50% or more of the central 1-mm-wide zone centered on the fovea (foveal DRIL) is associated with worse VA. This holds true even in eyes with reduced vision despite edema resolution or, conversely, in eyes with good vision despite concurrent edema. The strong association of foveal DRIL with VA in a previous cross-sectional study10 supported longitudinal investigation of foveal DRIL as a potential biomarker of future VA in eyes with current DME.
In this study, we longitudinally evaluated baseline foveal DRIL and other SD-OCT parameters to assess how they predict VA during 8 months in eyes with center-involved DME.
The study protocol was approved by the Joslin Diabetes Center Institutional Review Board and adhered to the tenets of the Declaration of Helsinki. Institutional review board approval was obtained to review the medical record data of the patients described herein. This was a single-site, retrospective cohort study conducted at the Beetham Eye Institute of the Joslin Diabetes Center, a tertiary referral center for diabetes mellitus care. All Beetham Eye Institute patients who underwent baseline and 8-month (±1 month) follow-up SD-OCT imaging (Spectralis; Heidelberg Engineering) between November 1, 2011, and December 31, 2012, were identified. A standard imaging protocol was used consisting of 49 B-scans spanning a 20 × 20° frame, with a mean of 16 automatic real-time images per scan in high-resolution mode. All imaging was performed by clinical study–certified imagers.
Eligible participants were 18 years or older with a history of diabetes mellitus (type 1 or type 2) and baseline center-involved DME (SD-OCT central subfield thickness [CST], ≥320 µm for men and ≥305 µm for women).11 Exclusion criteria were significant media opacity that precluded adequate images, cataract surgery within 6 months before baseline or during the 8-month follow-up period, and a history of nondiabetic pathology that might substantially affect VA. Each patient’s medical record was reviewed for baseline diabetic retinopathy severity level and age, sex, duration of diabetes mellitus, and most recent glycated hemoglobin level preceding baseline imaging. Early Treatment Diabetic Retinopathy Study VA (converted to logarithm of the minimum angle of resolution [logMAR] VA for analyses) was recorded for baseline, 4-month, and 8-month follow-up visits.
For each study eye, the scan passing through the foveal center was selected along with 3 B-scans immediately above and below, for a total of 7 scans (spaced 120 µm apart, encompassing a vertical height of 0.72 mm). A 1-mm-wide area centered on the foveal depression was analyzed for each B-scan using a standard template. Image analysis was performed by an experienced grader (M.M.L.) who was masked to VA, additional scan results, and all other clinical information. The following lesions were evaluated in the 1-mm-wide area centered on the fovea: microaneurysms, epiretinal membranes, subretinal fluid (SRF), small hyperreflective dots,5 DRIL (see the next paragraph), visibility of the cone outer segment tips (COSTs), large irregularly shaped hyperreflective foci consistent with hard exudates, intraretinal cysts (small, is <250 μm in horizontal width; medium, is ≥250 μm and <500 μm in horizontal width; and large, is ≥500 μm in horizontal width), and external limiting membrane (ELM) and ellipsoid zone (EZ [formerly inner segment–outer segment photoreceptor junction]) disruption and reflectivity.
Disorganization of the retinal inner layers was defined as the horizontal extent in microns for which any boundaries between the ganglion cell–inner plexiform layer complex, inner nuclear layer, and outer plexiform layer could not be identified (Figure 1). The finding of DRIL is assessed independently of and is not graded differently in the presence of retinal edema, intraretinal cysts, or any other SD-OCT–evident pathology. The DRIL extent was graded in each of 7 B-scans, and these gradings were averaged across 7 scans to derive a global DRIL measure for each eye at each visit. Reproducibility of DRIL grading was previously assessed for 3 independent masked graders (including M.M.L. and J.L.), with each evaluating 63 images. Pearson correlation coefficients for agreement of DRIL extent ranged from 0.80 to 0.86 (J.K.S., unpublished data, 2013).
Analyses were performed using statistical software (SAS, version 9.2; SAS Institute Inc). Bivariate linear regressions were performed to establish the association of VA outcomes with SD-OCT parameters. Methods to assess regression assumptions (eg, linearity, homoscedasticity, and normality of errors) included evaluating scatterplots of residuals vs predicted values and normal probability plots of residuals. Variables significantly associated with VA in unadjusted models were used to create multivariate models in which baseline VA (or early VA change) and early CST change were included as parameters of a priori interest. All analyses used repeated measures to adjust for correlations between right eyes and left eyes from the same individual. P < .05 was considered statistically significant in these exploratory analyses.
The data herein identify DRIL as a robust predictive biomarker of subsequent VA and determine DRIL thresholds beyond which visual improvement or decline is unlikely. Furthermore, this study provides multivariate models using DRIL and additional SD-OCT parameters that predict VA change over time in eyes with center-involved DME.
One hundred twenty eyes of 96 individuals were studied. Participants had a mean (SD) age of 61.0 (12.8) years, diabetes mellitus duration of 23.9 (12.4 years), and most recent glycated hemoglobin level of 7.9% (1.4%) (to convert hemoglobin level to proportion of total hemoglobin, multiply by 0.01) (Table 1). Thirty-five percent had type 1 diabetes mellitus, and 49.0% were female.
All 120 eyes had baseline and 8-month VA and SD-OCT images. Eighty-six eyes (71.7%) also had these data obtained at an interim 4-month (±1 month) visit. Participants with 4-month visits did not differ significantly from the entire cohort, and eyes with 4-month data did not differ significantly from all study eyes in baseline VA, CST, or a history of treatment with anti–vascular endothelial growth factor (VEGF) or laser (Table 1). For study eyes, the mean (SD) logMAR baseline VA was 0.28 (0.25) (approximately 20/40+1), and the mean (SD) baseline CST was 408.3 (91.5) µm. Almost 47% of eyes had received intravitreal anti-VEGF injections, and 62.5% of eyes had undergone macular laser treatment before the baseline visit (Table 1 and Table 2).
Association of Baseline DRIL and Other SD-OCT Parameters With Baseline VA
Greater horizontal DRIL extent was associated with worse baseline VA (point estimate, 0.04; 95% CI, 0.02-0.05 per 100 µm; P < .001) in bivariate analyses (Table 3 and eFigure 1 in the Supplement). Worse baseline VA was also associated with the following: thicker CST (0.10; 95% CI, 0.05-0.14 per 100 µm; P < .001), the presence of large cysts (0.02; 95% CI, 0.00-0.04 per B-scan; P = .02), less EZ reflectivity (−0.94; 95% CI, −1.35 to −0.53 per arbitrary unit; P < .001), fewer B-scans with visible COSTs (−0.03; 95% CI, −0.04 to −0.01 per B-scan; P < .001), and greater extent of ELM and EZ disruption (0.04; 95% CI, 0.02-0.07 per 100 µm; P < .001 and 0.04; 95% CI, 0.02-0.07 per 100 µm; P < .001, respectively). However, in multivariate modeling, only greater DRIL extent (P = .005) at baseline and, to a lesser degree, reduced baseline COSTs visibility (P = .04) remained independently associated with reduced VA even when controlling for baseline CST (P = .19) (Figure 2A).
Association of Change in DRIL and Other SD-OCT Parameters During 8 Months With Change in VA During 8 Months
As expected, worsening VA was associated with increasing CST (0.09; 95% CI, 0.06-0.12 per 100 µm; P < .001). Increasing DRIL extent during 8 months (0.03; 95% CI, 0.02-0.05 per 100 µm; P < .001) was also associated with worsening of VA during 8 months (eFigure 1 in the Supplement). In bivariate analyses, change in the following other SD-OCT parameters was also associated with worsening VA during 8 months: SRF (0.04; 95% CI, 0.02-0.07 per B-scan; P < .001), the presence of large cysts (0.02; 95% CI, 0.00-0.03 per B-scan; P = .01), decreasing COSTs visibility (−0.02; 95% CI, −0.04 to −0.01 per B-scan; P = .002), and decreasing EZ reflectivity (−0.54; 95% CI, −0.90 to −0.18 per arbitrary unit; P = .005) (eTable 1 in the Supplement). After multivariate analyses controlling for baseline VA, only changes in DRIL extent (P = .001), CST (P = .046), SRF (P = .01), and COSTs visibility (P = .003) were independently associated with change in logMAR VA from baseline to 8 months (Figure 2B).
Baseline DRIL and Other SD-OCT Parameters as Predictors of Future VA
Although some baseline SD-OCT parameters were associated with logMAR VA at 8 months in unadjusted analyses, none of these variables remained significantly related to VA at 8 months when included in multivariate models adjusting for baseline VA (eTable 2 in the Supplement). No baseline SD-OCT variables were significantly related to VA change from baseline to 8 months in unadjusted analyses.
In contrast, changes in some SD-OCT parameters during the first 4 follow-up months were predictive of change in VA during 8 months even after adjusting for baseline VA. Worsening VA during 8 months was associated in unadjusted analyses with the following: an increase in large cysts (0.02; 95% CI, 0.00-0.04 per B-scan; P = .049), a 4-month increase in CST (0.08; 95% CI, 0.04-0.12 per 100 µm; P < .001), an increase in DRIL extent (0.03; 95% CI, 0.02-0.05 per 100 µm; P < .001) (eFigure 1 in the Supplement), a decrease in COSTs visibility (−0.03; 95% CI, −0.04 to −0.01 per B-scan; P = .003), and an increase in extent of ELM and EZ disruption (0.04; 95% CI, 0.02-0.06 per 100 µm; P < .001 and 0.05; 95% CI, 0.03-0.08 per 100 µm; P < .001, respectively) (Table 3). However, multivariate linear regression demonstrated that reduction in VA during 8 months remained significantly correlated only with a 4-month increase in DRIL extent (P = .005), an increase in CST (P = .03), and a decrease in COSTs visibility (P = .04) from baseline even when controlling for baseline VA (Figure 2C).
Predictive Models of Future Change in VA Using DRIL and Other SD-OCT Parameters
When adjusting for change in VA from baseline to 4 months, only increase in DRIL extent (0.01; 95% CI, 0.00-0.02 per 100 µm; P = .03) and ELM disruption (0.02; 95% CI, 0.00-0.03 per 100 µm; P = .04) during 4 months were predictive of change in vision from baseline to 8 months. Using this model, the mean (SD) difference between predicted and actual change in logMAR VA was 0.00 (0.10) (r = 0.80) (Figure 2D). Absolute differences between predicted and actual VA change were less than 1 line in 68.6% of eyes, less than 2 lines in 97.7% of eyes, and 2 lines or greater in 2.3% of eyes.
In most eyes, increased DRIL extent at 4 months was associated with an 8-month VA reduction, while decreased DRIL extent was associated with improved VA (Table 4). Of 30 eyes in which DRIL decreased (improved) by more than 50 µm during the first 4 months, stable or improved VA at 8 months was observed in 96.7%. In this group, VA was 12 times more likely to improve by at least 1 line than decline by at least 1 line. Of 30 eyes in which DRIL increased (worsened) by more than 50 µm during the first 4 months, VA was stable or worse at 8 months in 90.0%. The VA was 4 times more likely to decline by at least 1 line than improve by at least 1 line.
When DRIL increased (worsened) by at least 250 µm at 4 months, no eyes had VA improvement of at least 1 line at 8 months. In contrast, when DRIL decreased (improved) by at least 250 µm at 4 months, no eyes had VA decline of at least 1 line at 8 months, but 77.7% had VA improvement of at least 1 line. In eyes in which DRIL extent changed by less than 50 µm during the first 4 months, 65.4% had VA change of less than 1 line at 8 months.
Point estimates for the effect of DRIL extent on VA outcomes were consistent across bivariate analyses (range, 0.028-0.036). On average, based on these estimates, for every 279-µm to 352-µm increase in DRIL extent, VA will be reduced by 1 line compared with an eye without such change in DRIL extent. More important from a predictive standpoint, for every 299-µm increase in DRIL extent during a 4-month period, VA after 8 months will be decreased by 1 line compared with baseline.
Subgroup analyses revealed consistent trends for increasing DRIL extent to be associated with worsening VA across analyses stratified by the presence or absence of DME treatment history and successful or unsuccessful resolution of DME at 8 months. Analyses stratified by DME treatment history revealed that, in eyes with and without a history of anti-VEGF treatment at baseline, baseline VA was significantly related to baseline DRIL extent. Furthermore, VA change during 8 months was significantly associated with DRIL change during 4 and 8 months regardless of treatment history. When adjusting for baseline VA, change in VA during 8 months was not significantly related to whether or not an eye had received anti-VEGF treatment during the study period, nor was it associated with the number of injections that had been given during the 8 follow-up months.
Results from this study suggest that centrally located DRIL is correlated with VA in eyes with center-involved DME. Most important, change in DRIL extent seems to be a predictive biomarker for future VA outcomes in these eyes. The association of DRIL extent change with VA change is more consistent and robust than that for change in any other SD-OCT parameter that we evaluated, including change in CST. Both baseline DRIL extent and change in DRIL over time were associated with VA outcomes in this cohort even for analyses in which central retinal thickness was not significantly related to VA.
On average, within the central 1-mm-wide area of the fovea, for every DRIL increase of approximately 300 µm from baseline to 4 months, VA is reduced by 1 line at 8 follow-up months compared with baseline. Most eyes (86.1%) in this cohort had an absolute DRIL change of less than 300 µm, so future studies are needed to elucidate the association of larger changes in DRIL with VA. Nonetheless, using our model, the absolute difference between predicted and actual VA change is less than 2 lines in 97.7% of eyes. Paradoxical VA change is defined as worsening VA despite resolving edema or as improving VA despite worsening edema. Highlighting the close correlation between DRIL and VA are the findings that paradoxical VA change exceeding 1 line at 8 months was never observed in eyes with an absolute change in DRIL extent of greater than 500 µm, was never observed with an absolute change in DRIL extent during 4 months of at least 250 µm, and was present in less than 5% of eyes with changes in DRIL extent of greater than 50 µm. These data also suggest that a threshold of DRIL worsening exists beyond which eyes are likely to lose vision and imply that DRIL improvement beyond a certain threshold may reliably portend future VA improvement.
The mechanisms by which DRIL affects VA have yet to be determined, and further histological correlation is needed. However, the inability to segment retinal layer boundaries likely represents an anatomic interruption in the visual transmission pathway. Disruption has been hypothesized to result when bipolar axons snap after their elasticity limit has been exceeded due to edema.12 Disorganization of the retinal inner layers could represent disorganization or destruction of cells within inner retinal layers, including bipolar, amacrine, or horizontal cells, and possibly indicates a disruption of pathways that transmit visual information from the photoreceptors to the ganglion cells. These anatomic changes could also account for less robust associations observed with other SD-OCT parameters, such as changes in photoreceptor outer segments.
Our results demonstrate that DRIL can resolve over time (eFigure 2 in the Supplement) and that resolving DRIL is a good indicator of subsequent VA improvement. Decreasing DRIL presumably represents anatomic improvement toward more normal morphology. Whether reversibility potential declines with increasing duration of DRIL or can be induced to occur is unknown but would have important implications for the timing of treatment initiation or novel therapeutic approaches. Although observed to some extent at all levels of baseline DRIL, a clear tendency was seen for more frequent improvement in DRIL extent with greater DRIL extent at baseline, especially when DRIL extent was greater than 400 µm. This finding was observed during 4 and 8 follow-up months (eFigure 3 in the Supplement).
In addition to the robust association of DRIL with VA outcomes, we observed more modest associations that have been previously reported. Increased COSTs visibility over time on SD-OCT B-scans was associated with an improvement in VA. It is possible that the inability to detect COSTs indicates disorganization or disruption of the photoreceptor layer. This finding is consistent with other studies7,9,13 reporting that poor photoreceptor integrity on SD-OCT is related to worse VA.
Data for this study were obtained as part of routine clinical care rather than as part of a research protocol. Although limitations of this approach are inherent, many are mitigated by the fact that, in this study, all technicians followed standard procedures and were all certified for clinical study protocol refraction, VA measurement, and SD-OCT imaging. A customized electronic health record was used that is specifically designed for standardized data entry, limiting data acquisition errors. Eyes included in this study had different DME durations and underwent different DME treatments before and during study follow-up periods. However, neither a history of DME treatment, anti-VEGF treatment during the study period, nor the success of DME treatment in leading to resolution of edema seemed to affect the association of foveal DRIL with VA in stratified analyses. The P values in this study must be interpreted cautiously in light of multiple comparisons; however, the association of DRIL extent with VA outcomes in this cohort was robust and seemed to be consistent across multiple analyses. It remains unknown at this time whether DRIL may be a similarly robust marker in eyes with edema from origins other than DME.
In summary, DRIL seems to be a correlated predictive biomarker for VA that is independent of central retinal thickness in eyes with baseline center-involved DME. This biomarker may be useful not only as a predictor of VA but also for stratification of eyes with regard to a high likelihood of future VA improvement or decline. Furthermore, the VA improvement observed when DRIL improves suggests both a new therapeutic approach to inducing VA recovery and a potential usefulness as an early marker for therapeutic trials. Therefore, DRIL change may represent a valuable and easily obtained noninvasive biomarker of VA that is applicable to clinical care and research studies and warrants further investigation.
Submitted for Publication: January 13, 2014; final revision received March 26, 2014; accepted March 27, 2014.
Corresponding Author: Jennifer K. Sun, MD, MPH, Beetham Eye Institute, Joslin Diabetes Center, One Joslin Place, Boston, MA 02215 (firstname.lastname@example.org).
Published Online: July 24, 2014. doi:10.1001/jamaophthalmol.2014.2350.
Author Contributions: Dr Sun and Mr Lin contributed equally to this work. Dr Sun had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Sun, Aiello.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Sun, Lin.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Sun, Lin.
Obtained funding: Sun, Lin, Aiello.
Administrative, technical, or material support: Sun, Lammer, Prager, Silva, Aiello.
Study supervision: Sun, Aiello.
Conflict of Interest Disclosures: Dr Sun reported receiving research support in kind from Optovue. No other disclosures were reported.
Funding/Support: This study was supported by the Harvard Medical School Scholars in Medicine Office (Mr Lin), by a grant from the Eleanor Chesterman Beatson Childcare Ambassador Program Foundation (Dr Sun), by grant JDRF 17-2011-359 from the Juvenile Diabetes Research Foundation International (Dr Sun), and by the Massachusetts Lions Eye Research Fund (Drs Sun and Aiello).
Role of the Sponsor: None of the sources of funding support had any role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
DL. The Wisconsin Epidemiologic Study of Diabetic Retinopathy, IV: diabetic macular edema. Ophthalmology
. 1984;91(12):1464-1474.PubMedGoogle ScholarCrossref
et al; Diabetic Retinopathy Clinical Research Network. Relationship between optical coherence tomography–measured central retinal thickness and visual acuity in diabetic macular edema. Ophthalmology
. 2007;114(3):525-536.PubMedGoogle ScholarCrossref
et al. Relationship between optical coherence tomography retinal parameters and visual acuity in diabetic macular edema. Ophthalmology
. 2010;117(12):2379-2386.PubMedGoogle ScholarCrossref
C; Diabetic Retinopathy Research Group Vienna. Optical coherence tomographic hyperreflective foci: a morphologic sign of lipid extravasation in diabetic macular edema. Ophthalmology
. 2009;116(5):914-920.PubMedGoogle ScholarCrossref
U; Diabetic Retinopathy Research Group Vienna. A systematic correlation between morphology and functional alterations in diabetic macular edema. Invest Ophthalmol Vis Sci
. 2010;51(12):6710-6714.PubMedGoogle ScholarCrossref
et al. Relationship between photoreceptor outer segment length and visual acuity in diabetic macular edema. Retina
. 2010;30(1):63-70.PubMedGoogle ScholarCrossref
Y. Association between external limiting membrane status and visual acuity in diabetic macular oedema. Br J Ophthalmol
. 2013;97(2):228-232.PubMedGoogle ScholarCrossref
WR. The association between percent disruption of the photoreceptor inner segment–outer segment junction and visual acuity in diabetic macular edema. Am J Ophthalmol
. 2010;150(1):63-67.e1. doi:10.1016/j.ajo.2010.01.039.PubMedGoogle ScholarCrossref
et al. Spectral domain optical coherence tomography parameters associated with visual acuity in patients with resolved center-involved diabetic macular edema. Invest Ophthalmol Vis Sci
. 2012; ARVO E-Abstract 1338.PubMedGoogle Scholar
et al; Diabetic Retinopathy Clinical Research Network. Retinal thickness in people with diabetes and minimal or no diabetic retinopathy: Heidelberg Spectralis optical coherence tomography. Invest Ophthalmol Vis Sci
. 2012;53(13):8154-8161. PubMedGoogle ScholarCrossref
J. Optical coherence tomography may be used to predict visual acuity in patients with macular edema. Invest Ophthalmol Vis Sci
. 2011;52(5):2741-2748.PubMedGoogle ScholarCrossref
N. Association of pathomorphology, photoreceptor status, and retinal thickness with visual acuity in diabetic retinopathy. Am J Ophthalmol
. 2011;151(2):310-317.PubMedGoogle ScholarCrossref