RETINAL VASCULITIS SEVERITY ASSESSMENT

Supplemental Digital Content is Available in the Text. Retinal vasculitis is assessed by fluorescein angiography, but there is no agreed-on scheme for grading its severity in clinical practice/trials. The authors propose a new straightforward grading scheme using wide-field fluorescein angiography. The scheme demonstrates good to excellent interobserver and intraobserver reliability and a correlation between its severity score and vision.

R etinal vasculitis (RV) is an inflammation of the retinal vessels, 1 which constitute the inner blood-retinal barrier. It has multiple causes, including infection, autoimmune diseases, inflammation, and neoplastic disorders, and can be idiopathic. 2,3 Retinal vasculitis potentially leads to visual loss because of macular edema and vascular occlusion.
Assessing RV can be challenging. Fundoscopic findings include perivascular hemorrhage, perivascular infiltrate, vessel sheathing, vessel caliber abnormalities, nonperfused retina, and retinal neovascularization 4 but can be subtle or absent when RV is subclinical. Currently, fluorescein angiography, preferably wide field, is the standard method to assess RV. 5 Retinal vasculitis manifests as vascular leakage, occlusion of vessels, or vessel wall staining on the angiogram. 6 Leakage or breakdown of the inner blood-retinal barrier can affect arterioles, venules, or capillaries. It is generally accepted that vascular leakage, either segmental or widespread, 7 is an indicator of inflammation and disease activity. 8 Inflamed vessels, either capillary or larger vessels, can occlude, resulting in retinal nonperfusion, and if extensive lead to neovascularization. Although an important indicator of severity, the presence of occlusion or nonperfusion is not a reliable indicator of active disease because it persists after inflammation has resolved.
Current clinical practice lacks agreement on a suitable severity classification or grading system for RV. Recent clinical trials have merely used the presence or absence of "active RV" as an endpoint, ignoring all information on severity. [9][10][11] The Standardization of Uveitis Nomenclature working group announced in 2005 that "there was consensus that the definition of RV required more work." 12 Later, the Angiography Scoring for Uveitis Working Group (ASUWOG) developed a 40-point scoring scheme to assess fluorescein angiography for posterior uveitis in standard images. 13 The ASUWOG scheme has not been widely adopted despite showing high interobserver reliability even in wide-field fluorescein angiography (WFFA) images. 14 It has only been used in two published studies of Behcet uveitis 8 including monitoring response to adalimumab. 15 The ASUWOG scoring scheme is too detailed for practical clinical use or even for clinical trials.
In this study, we aimed to develop a grading scheme for RV that was meaningful and easy to use. We decided to focus on vascular leakage and occlusion as the main criteria for WFFA imaging and their extent in the peripheral or central retinal zones. We tested the reliability of our novel grading scheme by analyzing agreement between same and different graders and correlated severity scores with visual acuity.

Study Design
This was a single-center, retrospective clinical study of patients diagnosed with RV and treated at St Paul's Eye Unit, Liverpool University Hospitals NHS Trust, Liverpool, United Kingdom, between 2009 and 2022. The study was conducted in accordance with the Declaration of Helsinki and was approved by the NHS Research Ethics Committee of the North of Scotland (reference number 19/NS/0186).
Two investigators identified patients attending St Paul's Eye Unit with RV based on electronic medical records. The diagnosis of uveitis was based on the Standardization of Uveitis Nomenclature criteria. 12 Eligible patients were older than 18 years with known RV in one or both eyes and had previous WFFA. Patients with proliferative diabetic retinopathy, retinal vascular disease other than RV, or other ocular diseases, which would interfere with WFFA assessment, were excluded. Informed consent was obtained from each patient.
Demographic data collected included age, sex, and race. We also collected the final etiological diagnosis of uveitis and the details of systemic immunosuppressant and local (ocular and periocular) treatment. Visual acuity was gathered as corrected visual acuity at the time of imaging and at 1-year follow-up. The corrected visual acuity was converted from Snellen to logarithm of the minimum angle of resolution for statistical analysis.

New Retinal Vasculitis Clinical Grading for Wide-Field Fluorescein Angiography Images
Our grading scheme was based on graded scoring for the degree of leakage and occlusion on WFFA per eye of a patient. First, the quality of images was assessed for focus and clarity and any obscuration from media opacity and graded as ungradable (0), where it was not possible to assess leakage from the retinal vessels or retinal perfusion with confidence, or when less than two retinal quadrants were seen (see below). For gradable images, image quality was noted as poor (1), fair (2), or good (3).
We divided the fundus into peripheral and central zones. The central zone was defined by a circle centered on the foveal center with distance from the central fovea to the nasal optic disk edge as its radius. This zone was chosen to include all of the macula, the optic disk, and the temporal vascular arcades, all of which are critical to maintaining central vision. Vascular leakage or occlusion in this zone is more likely to have consequences for vision and is therefore deemed more severe than changes in the peripheral zone. All of the visible retina external to the central zone was defined as the peripheral zone. Both zones were divided into four quadrants based on horizontal and vertical lines intersecting at the fovea ( Figure 1).
Retinal vascular leakage or "leakage" was defined as progressive hyperfluorescence coming from retinal vessels, either segments or full lengths of vessels. 13,17 Leakage is usually evident early in the fluorescein imaging and progresses to cover a larger area around vessels in later shots. Staining, defined as vessel wall hyperfluorescence evident from mid-run onward without expansion, 17 was not included because it is deemed not indicative of active disease. It did not include leakage at the level of the retinal pigment epithelium, which forms the outer blood-retinal barrier, leakage from optic disk, new vessels, or hyperfluorescence due to fluid-filled cysts in macular edema alone.
The leakage score was based on the number of quadrants showing the presence of leakage and their zone. Leakage in the peripheral zone was graded as L1, when 1 to 2 quadrants were involved, or L2, when 3 to 4 quadrants were involved. Leakage in the central zone was graded as L3 (1-2 quadrants) or L4 (3-4 quadrants) regardless of leakage in the peripheral zone. Thus, a leakage score of increasing severity from L0 to L4 was derived.
The occlusion score was based on the presence of occluded retinal vessels resulting in nonperfusion or ischemia, the zone affected, and the area of ischemia. Occlusion in the peripheral zone was graded as O1, when nonperfusion covered an area of less than three disk areas, or O2, when more than three disk areas. Nonperfusion in the central zone was graded as O3, when less than three disk areas, or O4, when more than three disk areas. The cutoff of three disk areas was derived from the ASUWOG grading scheme. All the grading criteria are summarized in Figure 1. In addition, we included further questions on the predominant vessel type involved, the presence of cystoid macular edema (CMO) (on WFFA alone), and the presence of optic disk leakage (Table 1). This was to explore the utility of these assessments.

Evaluation of the Grading Scheme
One investigator selected two images from each eye and masked the clinical data and shuffled the participant order for intraobserver testing. This investigator was excluded from the grading exercise. The graders were two uveitis specialists (Consultants 1 and 2), a trained senior ophthalmic image grader, and an ophthalmology higher specialist trainee (resident). All the graders were trained by the same example image sets before grading and undertook a pilot exercise on scoring for images independently. For intraobserver reliability, the ophthalmic image grader was asked to grade the same image set again after a gap of 1 month. For the purpose of analyzing interobserver reliability, the first grading of the ophthalmic image grader was used. The grading outcomes of Consultant 1 were used for subsequent analysis.

Statistical Analysis
All analyses were performed using SPSS 22.0 (SPSS Inc, Chicago, IL). The data are shown as mean, percentage, and SD. Student t-tests or chi-square tests were applied to compare the variables. For intraobserver and interobserver agreement, we calculated the intraclass correlation coefficient (ICC) for ordinal data (leakage score, occlusion score, and quality of image) and Fleiss' kappa statistics for categorical data (presence of cystoid macular edema, presence of optic disk leakage, and predominant vessel type involved). A single-measurement, absolute agreement with a two-way mixed effects model was applied for the ICC. The interpretation of both ICC and Fleiss' kappa was as follows: less than 0.40 as poor, 0.40 to 0.59 as fair, 0.60 to 0.74 as good, and 0.75 to 1.00 as excellent. 18 Univariable and multivariable generalized linear regression analyses were used to calculate the association of the new grading scheme with the corrected visual acuity. The generalized estimating equations accounted for the elimination of the intereye correlation. Factors showing a P value less than 0.1 in univariable analysis were consequently analyzed in the multivariable analysis. A P value less than 0.05 was considered statistically significant for all analyses.

Results
A total of 99 eyes from 50 patients with RV were recruited for this study. Table 1 summarizes the patients' characteristics and diagnoses. Of the 50 patients, 29 (58%) were female, most were White British (38, 76%), and the mean age at diagnosis was 46.4 years (interquartile range = 36.8-59). Thirty-eight (76%) patients had bilateral RV. Of 99 eyes, there were 87 eyes with RV and 12 without. The most
The interobserver reliability was investigated for four graders scoring independently. The result revealed good agreement for the leakage score (ICC 0.655, 95% CI 0.491-0.769) and excellent agreement for the occlusion score (ICC 0.750, 95% CI 0.679-0.813). It is noted that the ICC between Consultants 1 and 2 has the highest agreement for the leakage (ICC 0.795, excellent).
For additional questions assessing WFFA, the intraobserver reliability showed good agreement for CMO (kappa 0.638, 95% CI 0.635-0.640), poor agreement for optic disk leakage (kappa 0.293, 95% CI 0.291-0.296), fair agreement for the quality of WFFA imaging (kappa 0.565, 95% CI 0.449-0.609), and poor agreement for the predominant vessel type involved (kappa 0.266, 95% CI 0.265-0.267). The results of interobserver reliability are summarized in Table 2. As assessing CMO using WFFA is not standard practice, we additionally reviewed macular optical coherence tomography (OCT) scans acquired within 2 weeks (before or after) of WFFA in the study eyes for the presence/absence of CMO. Macular OCT scans were available in 82 study eyes. The result indicated fair agreement between assessing CMO by WFFA and OCT (kappa 0.444, 95% CI 0.317-0.571).

Association of Retinal Vasculitis Scoring With Visual Acuity
We compared the RV scoring with visual acuity using generalized linear regression analyses with generalized estimating equations (Table 3, Figure 4). The univariable analysis showed that worse visual acuity was associated with higher leakage scores (b = 0.105, P = 0.001) as well as older age (b = 0.005, P = 0.04), male sex (b = 0.162, P , 0.05), and presence of optic disk leakage (b = 0.202, P = 0.027). The multivariable analysis showed only higher leakage score (b = 0.090, P = 0.008) and older age (b = 0.006, P = 0.016) as statistically significant associations with visual acuity.

Discussion
There is currently no accepted grading scheme for RV, resulting in unquantified clinical assessment, poor communication between clinicians, and only a binary outcome for clinical trials in uveitis (present or absent) meaning much lost data. We present a pragmatic grading scheme based on WFFA, which is not time consuming and encompasses the important concepts of vascular leakage (activity) and occlusion (damage). We demonstrate that it has intraobserver and interobserver reliability across different cadres of observers and relates to function in the form of visual acuity.
At present, there are no Standardization of Uveitis Nomenclature criteria for assessing the severity of RV. 12 The ASUWOG scoring system 13 has been developed but has 40 points of assessment and is not practical for use clinically or in clinical trials. It has not been widely taken up and has only been tested in Behcet disease.
The recent decades have made significant progress in noninvasive imaging methods, specifically OCT and OCT-A, but these are not yet suitable alternatives to using WFFA for the assessment of RV, 19,20 which remains the standard method for diagnosis and severity assessment. 6 In this current study, we emphasize two major angiographic features of RV, leakage and occlusion, and we propose grading their involvement within a peripheral zone and a central zone on a scale of severity. Our investigation of intraobserver and interobserver agreement indicates that this grading scheme has good or excellent concordance, which means that it is reliable and reproducible between observers. Our interobserver agreement in the leakage score has a higher agreement between two uveitis consultants rather than an ophthalmic grader or trainee. This suggests that the accuracy of the assessment improves with clinical experience, and additional training may improve less experienced cadres.
Both univariable and multivariable analyses found that the leakage but not occlusion score was associated with visual acuity. The relationship between leakage and vision is meaningful in that it is measuring disease activity with an impact on visual function. Severity of vascular leakage thus may potentially serve as a clinical biomarker, indicating the need for more intensive treatment. 21,22 Our current analysis did not show an association between occlusion grade and visual acuity. Occlusion will only directly affect visual acuity if it affects the fovea, which will result in severe loss, so there is not an expected direct relationship with generalized severity. However, greater areas of occlusion and involvement of the posterior pole rather than the periphery have a greater chance of having a devastating impact on acuity. This is supported by previous research, which indicated a relationship between the foveal avascular zone size but not peripheral ischemia with poor visual outcome 5 in eyes with RV secondary to uveitis.
We found poor interobserver agreement for optic disk leakage. This feature may be related to overall uveitis activity or specific optic nerve involvement and RV. We suggest that it not be included in the grading of RV severity. We also assessed the concordance of observers to identify the predominant type of vessel involved in inflammation, which is said to aid the diagnosis of the underlying disease. Unfortunately, this only had poor agreement, and we suggest that this should not be included in a RV severity grading scale.
Besides clinician grading, automated computerized algorithms developed to grade retinal imaging are promising in many retinal diseases. 23 Automated algorithms could quantify areas affected by leakage and occlusion on continuous scales with weighting for central versus peripheral involvement. Attempts at This study was designed to retrospectively use images obtained in routine clinical practice from patients presenting with RV to a tertiary uveitis center. It was limited by using observers all from one center, albeit from different cadres. There were more examples of leakage than occlusion, as is encountered in clinical practice. The images used were not equally distributed in terms of scores for leakage and occlusion; in particular, there were more L4 images than L1, L2, or L3. This is in part a result of the simplicity of the scheme and also due to the limited sample size, but the relationship of the leakage grade to visual acuity supports this division of grades. Visual acuities were uniformly collected using Snellen charts and with the patients' own correction or pinhole, rather than refracted (best-corrected) visual acuity.
In conclusion, we have proposed a new severity grading scheme for RV focused on leakage and occlusion on retinal angiography and found it to have suitable intraobserver and interobserver reliability. The leakage score relates to visual acuity, which supports its use as a biomarker for disease activity. Prospective studies could further validate this scoring scheme for RV and how it relates to treatment decisions and outcomes.