Natural history studies of branch retinal vein occlusion (BRVO) show that even without treatment, 58% to 76% of patients experience some improvement in visual acuity (VA) after the spontaneous resolution of cystoid macular edema (CME).1 However, a subset of BRVO eyes do not regain vision even after treatment and resolution of CME.2,3 In untreated eyes that do regain some vision, improvement beyond 20/40 is uncommon.4,5 Several optical coherence tomography (OCT) features have been evaluated as surrogate predictors of VA outcomes. Reduction in central foveal thickness as well as integrity of the external limiting membrane (ELM), ellipsoid zone (EZ), and foveal bulge are associated with better VA outcomes in BRVO.6 Additionally, fluorescein angiography (FA) parameters such as size of parafoveal nonperfusion have been associated with VA even after resolution of CME, implicating macular ischemia as a potential mediator of VA outcome.7 Such imaging biomarkers for predictive VA stratification may improve patient counseling and better direct treatment strategies.
Disorganization of the retinal inner layers (DRIL) is another OCT parameter that reflects disruption of the normal structure of the inner retina. DRIL has been shown to be a robust surrogate marker for VA in eyes with uveitis-associated CME,8,9 diabetic macular edema,10–12 BRVO,13 and central retinal vein occlusion (CRVO).13,14 Mimouni et al.13 recently investigated the role of DRIL in 136 treatment-naïve eyes with RVO-related CME in a retrospective cohort study, finding that greater DRIL length was associated with worse VA both concurrently and subsequently during an 8-month period. Our group recently described that in treatment-naïve eyes with acute CRVO, DRIL length correlated with worse VA at 6 months and was predictive of worse VA subsequently through more than 2 years of follow-up.14 In addition, the extent of ischemia on ultra-widefield fluorescein angiography (UWFFA) at baseline was predictive of the DRIL development through more than 2 years of follow-up.
Such a relationship between the severity of retinal ischemia at baseline and the development of DRIL with long-term follow-up has, however, not been investigated in BRVO. This investigation was designed to examine the correlation of DRIL and other OCT-derived parameters with VA and the severity of baseline ischemia on UWFFA in treatment-naïve eyes with acute BRVO.
Patients and Methods
Institutional review board approval was obtained through Duke University School of Medicine in Durham, North Carolina. The Duke University investigational review board determined that informed consent was not necessary for this study because of its retrospective nature and absence of risks to subjects studied. All research adhered to the tenets of the Declaration of Helsinki, and all work using patient information was performed in compliance with the Health Insurance Portability and Accountability Act.
Using the Duke Enterprise Data Unified Content Explorer system, patients diagnosed with a BRVO (International Classification of Disease, Ninth Edition, code 362.36) between January 1, 2009, and July 1, 2016, were identified. Of these patients, only those with BRVO onset of 3 months or less prior to presentation, 1 year or more of available follow-up, no history of prior treatment, UWFFA at baseline, and spectral-domain OCT imaging at baseline and through follow-up were included. Patient medical records were reviewed for demographic data; treatment course; imaging findings; and clinical outcomes at the baseline, 6-month, 12-month, and most recent (final) visit. Multimodal imaging was reviewed as described below.
SD-OCT Analysis: Imaging was performed using the registered volume scans (with the large majority being 61-line raster and rarely 31-line raster) capturing the 25° × 30° area (Spectralis; Heidelberg Engineering, Heidelberg, Germany). The central line scan centered on the foveal depression was identified, and the central 1,000 μm (1-mm) portion of this B-scan was analyzed by two masked graders (JJY and DB). An overlay measuring 1,000 μm centered on the foveal depression was placed over the foveal scan to determine the central 1-mm portion. DRIL length was defined as the horizontal extent (μm) for which one or more boundaries between the inner retinal layers (ganglion cell layer and inner plexiform layer complex, inner nuclear layer and outer plexiform layer) were not separately identifiable and has been previously described (Figures 1–3).10–12
Optical coherence tomography showing foveal cystoid macular edema (CME) at baseline (A). The yellow box represents the central 1,000 μm and the discontinuities in the white (ganglion cell layer-inner plexiform layer [GCL-IPL] and inner nuclear layer [INL] interface), red (INL and outer plexiform layer [OPL] interface), and yellow (OPL and outer nuclear layer interface) lines represent the disorganized retinal inner layers (DRIL). The white double-ended arrow indicates the DRIL length. Visual acuity (VA) was 20/80 at baseline. Following anti-vascular endothelial growth factor intravitreal injections there was resolution of CME and at 13 months of follow-up, VA improved to 20/25 with resolution of DRIL (B).
Optical coherence tomography showing foveal cystoid macular edema (CME) at baseline (A). The yellow box represents the central 1,000 μm and the discontinuities in the white (ganglion cell layer-inner plexiform layer [GCL-IPL] and inner nuclear layer [INL] interface), red (INL and outer plexiform layer [OPL] interface), and yellow (OPL and outer nuclear layer interface) lines represent the disorganized retinal inner layers (DRIL). The white double-ended arrow indicates the DRIL length, which is greater than 500 μm. Visual acuity (VA) was 20/80 at baseline. Following anti-vascular endothelial growth factor intravitreal injections, there was resolution of CME and at 14 months of follow-up, VA improved to 20/32 with resolution of DRIL (B).
Optical coherence tomography showing foveal cystoid macular edema (CME) at baseline (A). The yellow box represents the central 1,000 μm and the discontinuities in the white, (ganglion cell layer-inner plexiform layer [GCL-IPL] and inner nuclear layer [INL] interface), red (INL and outer plexiform layer [OPL] interface) and yellow (OPL and outer nuclear layer interface) lines represent the disorganized retinal inner layers (DRIL). The white double-ended arrow indicates the DRIL length. Visual acuity (VA) was 20/64 at baseline. Following anti-vascular endothelial growth factor intravitreal and steroid injections there was persistence of CME and at 12 months of follow-up, DRIL length had increased (B) with worsening of VA to 20/200. There is also slightly increased disruption of the external limiting membrane at follow-up.
The central 1 mm (1,000 μm) of the central line scan was also reviewed for the following variables: central subfield thickness (μm), total intraretinal cyst area (mm2), extent of ELM disruption (μm), and extent of EZ disruption (μm). Total intraretinal cyst area was calculated by tracing the outline of each individual cyst using the built-in caliper tool in the Spectralis software and calculating the sum of individual cysts within the central 1 mm. It was included if the total value was greater than or equal to 0.01 mm2 (measurement limits on the Spectralis software). Additionally, subfoveal choroidal thickness (μm) was measured on enhanced depth OCT images using previously defined boundaries.15
Measurements obtained by the two graders were compared and any value with more than 5% difference was arbitrated by a third masked grader (AST). For measurements with 5% or less difference, the final measurement was obtained by averaging the measurements of the two graders.
UWFFA Analysis: An early phase, mid-phase, and late-phase UWFFA image obtained on the Optos 200 Tx camera (Optos, Dunfermline, UK) was reviewed in a masked fashion for foveal avascular zone (FAZ) enlargement and ischemic index (IsI) calculation by two graders (AST and SY). IsI calculations were performed as previously described.16,17 Briefly, the mid-phase image was imported into Adobe Photoshop (Version CC 2017; Adobe Systems, San Jose, CA). Contrast and brightness were adjusted to help clearly define a perfused/nonperfused junction. Next, areas of capillary nonperfusion were manually marked. The IsI (%) was calculated by measuring what percentage of imaged pixels representing the fundus were nonperfused (Figure 4). In eyes with a large amount of intraretinal hemorrhage, the areas corresponding to intraretinal hemorrhage were excluded from IsI calculations as previously described.16 We did not have any eyes with vitreous hemorrhage that obscured the fundus vasculature. IsI values obtained by the two graders were compared and any values with more than 2% gross difference were reviewed jointly to obtain a consensus value. For IsI values with 2% or less difference, the final measurement was obtained by averaging the measurements of the two graders.
Representative ultra-widefield fluorescein angiogram (UWFFA) of an eye with a superotemporal branch retinal vein occlusion included in this study. Mid-phase UWFFA images were analyzed for areas of nonperfusion (A). The nonperfused pixels (blue) were measured (B) and divided by the total number of imaged pixels representing the fundus. This value was multiplied by 100 to calculate the ischemic index (%).
Analyses were performed using SAS 9.3 (SAS Institute, Cary, NC). Recorded Early Treatment of Diabetic Retinopathy (ETDRS) VA was converted to logarithm of the minimum angle of resolution (logMAR) VA. Descriptive statistics were computed. Nonparametric tests were utilized since the data were not normally distributed. Univariate and multivariate analyses were performed to assess the effect of individual OCT and UWFFA parameters on VA. The relationship between DRIL and other imaging variables were assessed using simple linear regression. Multivariate linear regression for continuous variables was used to analyze the effect of DRIL on VA adjusting for other imaging findings. Although we performed multiple statistical tests, Bonferroni correction was not applied because it increases the risk of type II error, which is not ideal given the exploratory nature of this study. P values less than or equal to .05 were considered statistically significant.
Our cohort comprised 30 eyes of 30 patients (18 female, 12 male) with a mean age of 70.2 years ± 10.2 years (range: 49 years to 89 years). Mean follow-up was 25.4 months ± 11.0 months (range: 11.6 months to 47 months). Eighty percent of patients were white (n = 24), 16.7% were black (n = 5), and 3.4% self-identified as other (n = 1). Superotemporal location of BRVO was the most common (70%), followed by inferotemporal BRVO (30%). Mean logMAR VA at presentation was 0.56 ± 0.56 (range: 0 to 2.6; mean: 20/73 Snellen equivalent). At baseline, DRIL was present in 15 of 30 eyes (50%), with a mean DRIL length of 443.1 μm ± 460.4 μm (range: 0 μm to 1,000 μm). Both the prevalence and length of DRIL remained fairly stable throughout the follow-up period (50% at baseline to 46.7% at final follow-up). At baseline, CME was present in 22 eyes (73%), with intraretinal cysts present in 12 eyes (40%), and subretinal fluid in three eyes (10%). At the final visit, CME was present in nine eyes (30%), with cysts present in five eyes (16.7%), and subretinal fluid in none. Additional baseline and follow-up OCT parameters along with VA are shown in Table 1.
Visual Acuity and OCT Parameters at Baseline and Follow-Up Visits
Inter-reader Pearson correlation coefficients ranged from 0.90 to 1.0 (0.90 for horizontal DRIL length, 1.0 for intraretinal cyst area, 0.95 for length of ELM disruption, and 0.95 for length of EZ disruption). Mean IsI at baseline was 12.5% ± 7.9% (range: 1.3% to 26.8%). An enlarged FAZ was not present in any eye at baseline. Figures 1 to 3 provide representative OCT examples with change in DRIL length and corresponding change in VA through follow-up.
Association Between OCT Parameters and VA
Nonparametric bivariate correlation analysis was used to examine the relationship between the OCT parameters and VA at each visit in a cross-sectional manner. There were no significant associations between OCT parameters and VA at baseline. However, DRIL length and presence of CME were associated with worse VA at the 12-month and final visits. Extent of ELM and EZ disruption correlated with VA at the 6-month, 12-month, and final visits. Parameter estimates and P values for these associations are provided in Table 2.
Linear Regression Between OCT Parameters and logMAR VA at Baseline and Follow-Up Visits
Next, the relationship between the OCT parameters at the baseline visit and VA at follow-up visits was analyzed (Table 3). Baseline DRIL length was associated with worse VA at 12 months (P = .029). In addition, DRIL length at 12 months was associated with worse VA at the final visit; P = .011). Extent of EZ disruption at both baseline and 6 months was associated with worse VA at the 12-month and final follow-up visits. Extent of ELM disruption at baseline was associated with worse VA at 12 months. Both ELM and EZ disruption at 12 months were associated with worse VA at final follow-up. Choroidal thickness, central subfield thickness, or presence of subretinal fluid (SRF) did not impact VA at any time point. Although the presence of CME at baseline did not impact VA at the 6-month, 12-month, or final visits, presence of CME at the 12-month visit was associated with poorer final VA (P = .041). IsI at baseline was not significantly correlated with VA at any time point. Parameter estimates and P values for these associations are provided in Table 3.
Regression Analysis of OCT and UWFFA Parameters at Baseline and Visual Acuity at Follow-Up Visits
Separate regression analyses for DRIL length showed significant associations with presence of CME, total intraretinal cyst area, and extent of ELM and EZ disruption at each visit in a cross-sectional manner (Table 4). Greater central subfield thickness was associated with increased DRIL length at all visits except for the final visit. Presence of SRF and choroidal thickness did not impact DRIL length. Regression analysis for prediction of subsequent DRIL length (Table A available at www.healio.com/OSLIRetina) showed a positive correlation with baseline DRIL length impacting the DRIL length at 6 months (P = .0019) and 12 months (P = .0002). Baseline DRIL length was not, however, associated with final DRIL length (P = .19).
Relationship Between DRIL Length (μm) and Other OCT Parameters
DRIL length at 12 months was also associated with presence of CME (P = .014), as well as extent of ELM (P = .034) and EZ disruption (P = .011) at the same visit. DRIL length at the final visit was associated with DRIL length at 6 months (P = .012) and 12 months (P = .0001), presence of CME at 6 months (P = .0014) and 12 months (P = .0062), ELM disruption at 6 months (P = .0006) and 12 months (P = .0032), and EZ disruption at 6 months (P = .007).
Association of UWFFA Parameters With VA and OCT Parameters: IsI at baseline was not predictive of VA or DRIL length at any time point.
Multivariate Regression: Associations Between Imaging Findings and Final VA
A multivariate regression model including all OCT parameters and CME presence was used to predict final VA. At the final visit, central subfield thickness, DRIL length, and ELM disruption were associated with poorer VA (Table 5). With each 100 μm increase in the DRIL length at the final visit, there was an estimated decline in VA of 0.046 logMAR. At the final follow-up visit, DRIL length was significantly associated with poorer VA even after controlling for presence of CME (P = .0035) (Table B available at www.healio.com/OSLIRetina). There were, however, no significant associations between baseline OCT parameters and final VA.
Multiple Linear Regression Analysis of Final logMAR VA Dependent on OCT Variables and Presence of CME at the Final Follow-Up Visit
The first treatment for the majority of eyes was intravitreal bevacizumab (Avastin; Genentech, South San Francisco, CA) with no treatment being the second most common practice. During the course of the 2-year follow-up, intravitreal bevacizumab (mean: 4.13 injections ± 6.47 injections) was more commonly used than ranibizumab (Lucentis; Genentech, South San Francisco, CA) (mean: 2.1 injections ± 6.36 injections) or aflibercept (Eylea; Regeneron, Tarrytown, NY) (mean: 3.83 injections ± 6.75 injections). Overall, study eyes received a mean of 5.43 ± 4.47 anti-vascular endothelial growth factor (VEGF) injections in the first year and 10.33 ± 10.37 injections through the entire follow-up duration. Four eyes received an intravitreal steroid injection. Six eyes underwent sectoral scatter laser photocoagulation and five underwent grid-pattern laser. Univariate regression of total number of injections administered during the first year and by final follow-up showed no correlations with DRIL length at either time point.
We used OCT-derived anatomic parameters and regression models to assess the impact of DRIL on VA through more than 2 years of follow-up and to study its association with baseline UWFFA features in treatment-naïve eyes with acute BRVO. In our study, DRIL length at baseline was predictive of worse VA at 12 months and DRIL length at 12 months was predictive of worse final VA. Presence of CME at baseline and extent of baseline disruption of the ELM and EZ were associated with DRIL length at 12 months. In multivariate analyses, DRIL length was associated with worse VA at the final visit while controlling for other OCT parameters. We did not find an association between baseline IsI values on UWFFA and DRIL length through the course of follow-up.
In a combined cohort of both BRVO and CRVO eyes, DRIL length has previously been shown to correlate with VA in a univariate analysis, although it was not significant in a multivariate model.18 Additionally, Mimouni et al.13 reported both the association and predictive value of DRIL with worse VA in eyes with macular edema associated with all types of RVO treated with bevacizumab. In treatment-naïve eyes with acute CRVO, our group recently reported that the extent of DRIL at 6 months was associated with worse VA at the 12-month and final visits, and DRIL was the only OCT feature concurrently associated with worse VA at final follow-up.14 However, literature investigating DRIL in BRVO specifically and correlating DRIL with ischemia on fluorescein angiography is scant. We report not only the association with and predictive value of DRIL on VA, but also the lack of association between DRIL and severity of baseline ischemia.
DRIL has been proposed to represent disruption of the synaptic connections and cell bodies of amacrine, bipolar, and horizontal cells within the inner retina and is thought to be a manifestation of neuroglial degeneration due to ischemia, inflammation, or both.8,10,19 DRIL length is a highly specific and moderately sensitive predictor for macular capillary nonperfusion in eyes with diabetic retinopathy (DR),20,21 and DRIL length has been positively correlated with the size of the FAZ in both DR and RVO.18 Further, in BRVO eyes with CME that resolved after treatment, VA has been associated with both the parafoveal nonperfusion area and the vascular density of the superficial and deep capillary plexus.6,21 It is thus hypothesized that the vascular insult in eyes with BRVO may lead to cell death in the inner retinal layers and the subsequent appearance of DRIL on OCT. In our cohort, however, none of the eyes had enlargement of the FAZ on UWFFA, and the extent of macular ischemia was predominantly outside the area of the FAZ, which may explain why it did not impact vision and also the lack of an association between DRIL and macular ischemia.
In a cohort of eyes with treatment-naïve, acute CRVO, we previously found that IsI on baseline UWFFA was predictive of the final DRIL length.14 The lack of an association between DRIL and ischemic parameters in BRVO in this study may be explained by the lesser extent of ischemic retina in eyes with BRVO compared to CRVO, suggesting that the amount of ischemic retina itself may play a contributory role. It is also possible that ischemia in CRVO may be a more diffuse process, whereas in BRVO ischemia may be more focal, and if located away from the FAZ, may not always affect VA. In addition, the ischemic insult is acute and usually more severe in CRVO compared to BRVO, and thus the full extent of DRIL may not manifest at the time of presentation. Perhaps there exists a threshold below which ischemia does not significantly impact the development of DRIL. However, this needs to be further evaluated. Further, as fluorescein angiography does not provide the same depth of insight into capillary nonperfusion in the superficial and deep retinal capillary plexus as OCT angiography, FAZ enlargement may have been underestimated in our study, potentially confounding the analysis. Indeed, after controlling for FAZ size, Balaratnasingam et al.18 found that the VA-DRIL relationship in a combined series of BRVO and CRVO eyes was not significant, suggesting a strong association between FAZ area on OCT angiography and VA. This relationship warrants further investigation in future studies.
Our finding that presence of CME at baseline was associated with DRIL length at 12 months suggests that in BRVO, inflammation, rather than ischemia, may be the primary driver of DRIL development. It has been suggested that if edema falls below the threshold of the elasticity limits of the neurosensory retina, neuronal synapses and thereby VA may be preserved on edema resolution.23 However, edema beyond the elastic capacity may cause the bipolar axons to snap, leading to a potentially irreversible loss of neuronal signaling from photoreceptors to ganglion cells. This may explain the persistence of VA loss and DRIL even following CME resolution in a subset of BRVO eyes, suggesting a component of irreversible damage to the inner retinal layers. As has been recognized before, however, histologic correlations of DRIL are required to better understand the pathophysiology and subsequent implications of this finding.
We also report that extent of ELM and EZ disruption correlated with worse VA at both concurrent and subsequent visits. Das et al.24 recently described the significant association between DRIL and disruption of the ELM and EZ in diabetic macular edema. Although the associations of ELM and EZ with both VA and capillary nonperfusion are well established,6,7,18,25 few studies have reported the predictive value of these parameters. Ischemia of the deep plexus is better detected on OCT angiography, although there are currently unresolved challenges in accurate anatomic segmentation of the deep retinal vascular bed.26
Strengths of this study include the long follow-up in a real-world setting, enabling the detection of the predictive value of DRIL during a period of more than 2 years, along with the strict inclusion criteria and robust image grading methodology using masked graders with excellent agreement. Unique to this study is evaluation of the relationship between the severity of retinal ischemia as measured on UWFFA and DRIL.
The primary limitations of this study are inherent to the retrospective nature and relatively modest sample size. Subsequently, there may be inadequate power in detecting predictive value of DRIL and a larger margin of error in our findings. We also did not correct for non-linear peripheral distortion or adjust for normal perfused retina in our UWFFA images while calculating the ischemic index, which may have potentially impacted the correlations of this parameter.27
In conclusion, we report the utility of DRIL as an OCT biomarker predictive of worse VA in treatment-naive eyes with acute BRVO through more than 2 years of follow-up, even after controlling for CME as well as EZ and ELM disruption. The lack of association between DRIL and baseline ischemic index on UWFFA suggests that DRIL development in BRVO may be more influenced by the amount of edema rather than ischemia. DRIL length and associated CME, ELM, and EZ disruption, therefore, are better predictors of eventual vision than is the IsI in patients with acute, treatment-naive BRVO. However, larger, prospective studies with longer follow-up and a more detailed evaluation of ischemia using OCT angiography are critical in establishing a rigorous independent association of DRIL with future VA and validating the use of DRIL as a predictive imaging biomarker in BRVO.