The Standardization of Uveitis Nomenclature (SUN) project

Abstract

The uveitides are a collection of over 30 diseases characterized by intraocular inflammation. Previous work demonstrated that the agreement among uveitis experts on diagnosis was modest at best with some pairs of experts having chance alone agreement on selected diseases. The Standardization of Uveitis Nomenclature (SUN) is a17-year collaboration among experts in uveitis, ocular image grading, informatics, and machine learning to improve clinical and translational uveitis research. The SUN “Developing Classification Criteria for the Uveitides” project used a rigorous, multi-phase approach to develop classification criteria for 25 of the most common uveitic diseases. The project’s phases were: 1) informatics; 2) case collection; 3) case selection; 4) machine learning; and 5) consensus review and publication. The results were classification criteria with a high degree of accuracy (93.3% to 99.3% depending on anatomic class of the uveitis), the goal of which is to form the basis for future uveitis research.

1. INTRODUCTION

The uveitides are a collection of over 30 diseases characterized by intraocular inflammation (Table 1).¹ They may be due to intraocular infection (e.g., herpes simplex virus anterior uveitis, varicella zoster virus anterior uveitis, acute retinal necrosis, etc.), due to a systemic infection (e.g., syphilitic uveitis), associated with a systemic disease (e.g., spondyloarthritis/HLA-B27-associated anterior uveitis, juvenile idiopathic arthritis [JIA]-associated anterior uveitis, and sarcoidosis-associated uveitis), or eye-limited and presumably immune-mediated (e.g., birdshot chorioretinitis, serpiginous choroiditis). In addition to diseases that are classifiable as a specific entity, uveitides may lack distinguishing clinical features leading to their classification as undifferentiated.^1,2

Table 1:

Selected Uveitic Diseases^†

Anatomic class	Infectious‡	Systemic Disease Associated	Eye-limited
Anterior	Cytomegalovirus anterior uveitis	Juvenile idiopathic arthritis-associated anterior uveitis	Fuchs uveitis syndrome
	Herpes simplex virus anterior uveitis	Spondyloarthritis/HLA-B27-associated anterior uveitis	Undifferentiated anterior uveitis
	Varicella zoster virus anterior uveitis	Tubulointerstitial nephritis with uveitis
	Syphilitic anterior uveitis	Sarcoidosis-associated anterior uveitis
Intermediate	Syphilitic intermediate uveitis	Multiple sclerosis-associated intermediate uveitis	Pars planitis
		Sarcoidosis-associated intermediate uveitis	Intermediate uveitis, non-pars planitis type (undifferentiated intermediate uveitis)
Posterior	Acute retinal necrosis	Sarcoidosis-associated panuveitis	Acute posterior multifocal placoid pigment epitheliopathy
	Cytomegalovirus retinitis		Birdshot chorioretinitis
	Syphilitic posterior uveitis		Multiple evanescent white dot syndrome
	Toxoplasmic retinitis		Multifocal choroiditis with panuveitis
	Tuberculous posterior uveitis		Punctate inner choroiditis
			Serpiginous choroiditis
			Undifferentiated choroiditis
			Undifferentiated retinal vasculitis
Panuveitis	Syphilitic panuveitis	Behçet disease uveitis	Sympathetic ophthalmia
	Tuberculous panuveitis		Undifferentiated panuveitis with choroiditis
		Sarcoidosis-associated panuveitis	Undifferentiated panuveitis with retinal vasculitis
		Vogt-Koyanagi-Harada disease (Early-stage and late-stage)

^†Adapted from Standardization of Uveitis Nomenclature (SUN) Working Group. Development of classification criteria for the uveitides. Am J Ophthalmol 2021;228:96–116. Used with permission.

^‡Infectious uveitides refer to those with evidence of active infection. They do not include auto-inflammatory or auto-immune diseases triggered by a prior infection (e.g. reactive arthritis-associated uveitis).

In the past, the field of uveitis was dominated by the approach termed the “etiologic diagnosis of uveitis,” in which the goal of diagnosis was to find the “cause” of the uveitis. If no cause was identifiable, the patient was diagnosed with “idiopathic uveitis”. This approach was fraught with logical inconsistencies and inconsistent clinical management. The purported “causes” included infectious causes (e.g., herpes family viruses, syphilis), associated systemic diseases (e.g., juvenile idiopathic arthritis [JIA], sarcoidosis), syndromes with several causes (e.g., acute retinal necrosis), and defined uveitic diseases (e.g., birdshot chorioretinitis). A child with undifferentiated chronic anterior uveitis with no associated systemic disease would be considered “idiopathic”, whereas a child with an identical anterior uveitis associated with JIA would be considered not “idiopathic”, even though both are of unknown cause (i.e., “idiopathic”). Clinically patients underwent extensive laboratory testing (e.g. “uveitis surveys”) looking for the “cause” of the uveitis, including tests with low sensitivity, specificity, and positive predictive value, such as serum angiotensin converting enzyme and lysozyme levels, and HLA typing other than HLA-B27 and HLA-A29.^1,3–5 Diagnoses often were made based on abnormal results for these tests, despite the absence of evident non-ocular clinical disease, in an effort to find the “underlying cause” of the uveitis. There was limited agreement on terminology and multiple grading systems for the severity of the inflammation each with a different number of steps (e.g. the grading schema for anterior chamber cells ranged from 6 to 9 steps), leading to difficulties with interpreting the term “improvement” in the uveitis.² The lack of agreement on terminology led Rosenbaum and Holland to write an editorial in 1996 entitled “Uveitis and the Tower of Babel”,⁶ which challenged the field of uveitis to develop a consistent terminology.

Prior to the Standardization of Uveitis Nomenclature (SUN) “Developing Classification Criteria for the Uveitides” project several groups had proposed criteria for the diagnosis of several of the uveitides.^7–13 In general, these criteria were developed using informal consensus techniques and with limited evaluation of their performance against other diseases in the differential diagnosis. For several of these diseases’ criteria, problems identified on evaluation led to revisions.^10,14–16 In the field of rheumatology there is a distinction between diagnostic criteria (often developed for clinical use) and classification criteria (developed for research use). Diagnostic criteria often emphasize sensitivity in an effort to be as inclusive as possible for the diagnosis of the disease. Classification criteria strive to optimize sensitivity and specificity, but when a trade-off is needed, they emphasize specificity in order to ensure that the phenotype defined represents a homogeneous group of patients.¹⁷ The emphasis on specificity is critical in clinical research, such as genomic research, translational research, and pathogenesis research, where including a substantial proportion of patients without the disease being studied could confound the results. In most of the criteria developed previously for the uveitides, the authors did not make this distinction, and agreement on diagnosis of the uveitides among uveitis experts was modest at best. In 2018 an evaluation of the agreement on diagnosis among uveitis experts for 25 of the more common uveitides revealed an overall κ of 0.39, considered moderate agreement at best.¹⁸ The κ statistic is a measurement of agreement that ranges from −1.00 (perfect disagreement) to +1.00 (perfect agreement), and a κ of 0 represents chance agreement. The disease-specific κ’s ranged from 0.23 to 0.79; on only one disease did the disease-specific κ fall into the substantial range; and for selected pairs of experts the agreement was ~0, indicating chance agreement.¹⁸ Because of these problems the Standardization of Uveitis Nomenclature (SUN) Working Group was brought together to improve clinical and translational research in the field of uveitis by standardizing terminology, inflammation grading, outcome measures, and by developing classification criteria for the more common uveitides.

2. THE FIRST STANDARDIZATION OF UVEITIS NOMENCLATURE WORKSHOP AND ITS IMPACT

The SUN Working Group began in 2004 when 50 uveitis experts came together to begin the process of standardizing the approach to clinical uveitis research. It has since expanded to nearly 100 members with expertise in uveitis, ophthalmic image grading, informatics, and machine learning. The First SUN Workshop was held in 2004, and the results of it were published in 2005.² The First SUN Workshop used formal consensus techniques^2,19 to develop a consensus on the anatomic classification of the uveitides, reporting structural complications of uveitis, grading schema for anterior chamber cells, anterior chamber flare, and vitreous haze, as well as outcome measures, including visual acuity thresholds for reporting visual acuity results, definition of successful corticosteroid sparing, definition of inactive uveitis, and definitions of improved and worsened inflammation severity.² In the first two years after its publication, the article reporting the results of the First SUN Workshop was the most cited article published by the American Journal of Ophthalmology (Liesegang T, personal communication). Over the next 17 years the article was cited approximately 2600 times (exaly.com accessed 11 September 2022) or on average over 150 times/year. A subsequent independent evaluation of the SUN grading schema demonstrated that the κ’s for the interobserver agreement within one grade for anterior chamber cells were in the almost perfect range and that the interobserver agreements within one grade for anterior chamber flare and for vitreous haze were in the substantial range,²⁰ results which supported their use in clinical research. Over time there have been a few slight modifications to the results of the First SUN Workshop. For example, the First SUN Workshop agreed that successful corticosteroid-sparing should be inactive uveitis at a dose of prednisone <10 mg/day,² but subsequent data have led to changing the definition of this outcome to inactive uveitis at a dose of prednisone <7.5 mg/day.^21–24

The next subject tackled by the SUN Working Group was development of classification criteria for 25 of the more common uveitides.

3. DEVELOPMENT OF CLASSIFICATION CRITERIA FOR THE UVEITIDES

The SUN “Development of Classification Criteria for the Uveitides” project was conducted in five phases: 1) informatics, 2) case collection, 3) case selection, 4) machine learning, 5) consensus review and publication.²⁵

3.1. Informatics

The first phase of the SUN uveitides’ classification criteria project was the development of a standardized terminology and nosology for the uveitides conducted in 2009 through 2010 and using formal consensus techniques.^25–27 The initial terminology was developed using a modified “green field” approach coupled with web-based surveys and teleconferences using a modified Delphi technique.²⁶ The terms were mapped into ontologic dimensions for each uveitic disease. In 2010 the SUN Working Group met at the Second SUN Workshop and formalized these terms and mappings using nominal group techniques. The outcome was a supermajority consensus on how to describe each of the uveitic diseases using 190 terms and phrases mapped into 10 dimensions.^26,27 This work product enabled the development of a standardized, hierarchical, “menu-driven” case collection tool for the second phase of the SUN uveitides classification criteria project.²⁵

3.2. Case Collection

The second phase of the SUN uveitides classification criteria project was conducted from 2010 through 2016 and consisted of collecting cases of the 25 uveitides under consideration into a customized data base using the customized, informatics-based case collection tool. Seventy-six clinician investigators from five continents entered cases. In order to minimize selection bias and maximize the use of more modern imaging, investigators were instructed to enter consecutive cases of the diseases working from the most recent case backwards in time. Investigators were instructed to enter the data from the presentation visit or if there was disease evolution, the visit at which the diagnosis became known. The target for case collection was 150–250 cases of each disease. Once approximately 250 cases of an individual disease were collected, case collection for that disease was closed. Because they entered into the differential diagnosis of several classes of the uveitides and because of their protean morphology, more than 250 cases of sarcoidosis-associated uveitis (383 cases) and of tubercular uveitis (358 cases) were collected. In addition, because of their different clinical presentations, cases of early-stage Vogt-Koyanagi-Harada (VKH) disease and of late-stage VKH disease were collected separately.²⁵

Investigators were instructed to submit images relevant to the diagnosis into the database. These included fundus photographs for posterior and panuveitides, and fluorescein angiograms, optical coherence tomograms, etc. as appropriate. These images were used by the case selection committees to assist with case selection and were graded by an independent Image Reading Center for features such as numbers of lesions, predominant lesion size, lesion location, lesion morphology, etc. Image Reading Center data were used preferentially in the machine learning. Image results relevant to the disease criteria were reviewed by a clinician dedicated to image management, and any discrepancies adjudicated. Images themselves were not subjected to machine learning, only the results of the image grading, which was integrated into the clinical database.²⁵

The product of the case collection phase was a preliminary database of 5766 retrospectively-collected, de-identified cases of the 25 uveitic diseases under consideration for use in the case selection phase.²⁵

3.3. Case Selection

Due to the absence of a “gold standard” for the diagnosis of each disease, and because of the limited agreement among investigators on disease diagnosis,¹⁸ it was decided to use formal consensus techniques to select cases from the preliminary database into a final database for the machine learning phase. Case selection was deemed necessary, as the goal of the SUN classification criteria project was the development of classification criteria for research use in which a homogeneous group of patients is needed.¹⁷ The case selection phase occurred from 2016 through 2017. Cases in the preliminary database were reviewed by committees of nine uveitis experts for inclusion into the final data base for use in machine learning. These committees were geographically and “school of thought” dispersed. Five committees worked in parallel, addressing cases in the differential diagnosis of each other, generally by anatomic class. The core membership of each committee was the same for each disease within an anatomic class, but there was some variability of membership on a disease-specific basis depending on availability. Case selection proceeded in two steps: online voting and consensus conference calls. During the online voting committee members reviewed the cases individually and voted to “accept” the case for inclusion into the final database or to “reject” it from the final database. Cases getting a supermajority (>75%) agreement on acceptance were included in the final database, and cases getting a supermajority agreement on rejection were excluded from the final database. Cases without a supermajority agreement on either acceptance or rejection were tabled for consensus conference calls.²⁵

Consensus conference calls were accomplished using nominal group techniques, a formal consensus technique that minimizes “dominant personality” effects. The technique involves a round of time-limited comments from each committee member followed by anonymous voting on acceptance or rejection. Cases with a supermajority agreement for acceptance were included in the final database, and cases with a supermajority agreement for rejection were excluded from the final database. Cases obtaining neither a supermajority agreement for acceptance nor for rejection were subject to a second round of discussion and voting. If after the second round of voting no supermajority agreement was obtained, the case was tabled and not included in the final database.^2,19,25

Of the 5766 cases collected, 4046 (70%) were selected for inclusion in the final database. Supermajority agreement on inclusion in or exclusion from the final database was achieved on 99% of the collected cases with only1% of cases being tabled.²⁵ The cases selected came from all of the geographic regions with some regional variation by disease prevalence (Table 2).²⁵ For example, Asian cases were more common in panuveitides (due to the greater prevalence of VKH disease and Behçet disease in Asia) and in infectious posterior and panuveitides (due to the greater prevalence of tubercular uveitis in Asia). When a single region produced a disproportionate number of cases, comparative analyses of features by region was performed to ensure that the criteria were applicable across regions.^25,28–31

Table 2:

Regional Origin of Cases Selected for the Final Database in the SUN Classification Criteria Project¹

Uveitic Class	Anterior Uveitides	Intermediate Uveitides	Posterior Uveitides	Panuveitides	Infectious Posterior & Panuveitides
Number cases	947	452	735	846	1066
Region	Regional Origin of Selected Cases (% of cases in class)
Asia	14	2	9	34	27
Australia	7	3	2	4	5
Europe	44	49	31	26	25
North America	34	39	57	31	31
South America	1	7	1	5	12

¹Adapted from Standardization of Uveitis Nomenclature (SUN) Working Group. Development of classification criteria for the uveitides. Am J Ophthalmol 2021;228:96–116. Used with permission.

3.4. Machine Learning

Machine learning to develop the individual disease criteria was conducted from 2018 through 2019. The final database of 4046 cases was split into a training set (~85% of cases) and a validation set (~15% of cases) for each disease. Database modifications necessary to perform machine learning included changing any “check all that apply” questions into a series of binary “yes/no” or “present/absent” questions. Because of the retrospective nature of data collection and because of the Bayesian approach to testing currently advocated, not all laboratory or imaging data were available on all cases. Therefore, an “evidence for” approach was adopted, in which data supporting the diagnosis were used to make the diagnosis, and missing data were treated as negative, an approach which mimics clinical care. However, because syphilitic uveitis and sarcoidosis-associated uveitis are in the differential diagnosis of all uveitic classes, relatively complete data were available for these two diseases.²⁵

Machine learning generally was performed by anatomic class, as the diseases in each class represent the typical differential diagnosis. For example, anterior uveitides and posterior uveitides are distinguished by the presence of chorioretinal lesions in the latter, so that the two classes would have been automatically separated by chorioretinal lesions if machine learning had been performed on both classes together. For those diseases which crossed anatomic class (e.g. syphilitic uveitis, sarcoidosis-associated uveitis, tubercular uveitis), appropriate cases were included in the relevant anatomic classes for the machine learning (and hence could be in more than one class). Therefore, the numbers of cases used in the machine learning were: anterior uveitides, 1083; intermediate uveitides, 589; posterior uveitides, 1068; panuveitides 1012; and infectious posterior/panuveitides 803 cases, respectively.²⁵

The uveitic diagnosis is patient-specific; therefore, eye-specific information was coalesced into patient-specific information. If a feature was present in either eye, it was considered present for the individual. If a feature was present in both eyes and there was differential severity, the “worse” eye was used. If there were multiple options for a feature (e.g. predominant lesion size), it was taken as the larger of the two ranks.²⁶

Machine learning was used on the training set to determine the criteria that minimized the misclassification rate, which was defined as the proportion of cases classified incorrectly when compared to the consensus diagnosis. Four classification methods were considered, listed with their tuning parameter and R package name in parenthesis: 1) classification and regression trees (CART) with cost-complexity pruning and cp=0.01 (rpart); 2) random forests with default parameters (randomForest); 3) multinomial logistic regression with lasso regularization and the 1 standard error value chosen for lambda (gimnet); and 4) support vector machines with radial kernel and tuning performed on a gird of coast and gamma values (e1071). In order to develop parsimonious sets of criteria and avoid over-fitting, the Boruta algorithm (R package Boruta) was used. The Boruta algorithm is an all relevant feature algorithm that uses random forests and compares diagnostic accuracy to the number of features included, enabling it to generate classification criteria that are both parsimonious and accurate.^25,32 All four methods produced similar results (increasing the confidence in the results), although CART provided a slightly worse performance. Multinomial logistic regression with lasso regularization was used to identify classification rules as linear combinations of features, which were restated as equivalent Boolean classification rules. The Quine-McCluskey algorithm as extended by Dusa and Thiem was used to construct a minimal set of Boolean expressions for each uveitic disease.^25,33

To optimize the performance of the criteria, an iterative approach was used for feature engineering on the training set. Clinically-relevant “or” variables were combined into a single “evidence of” variable. For example, chest radiograph results and chest computed tomography results were combined to produce a variable identifying bilateral hilar adenopathy on chest imaging (chest radiograph or chest computed tomography). This variable then was combined with the results of a tissue biopsy demonstrating non-caseating granulomata to produce an “evidence of sarcoidosis” variable. All such feature engineering was performed only on the training set and without reference to the diagnosis.²⁵

After the criteria were developed on the training set, they were evaluated on the validation set to determine the accuracy of the criteria (proportion of cases identified correctly versus the consensus diagnosis). Accuracies by anatomic class (Table 3) were: anterior uveitides 96.7%; intermediate uveitides 99.3%, posterior uveitides 98.0%; panuveitides 94%; and infectious posterior or panuveitides 93.3%, respectively.

Table 3:

Accuracies of Machine Learning Classification Criteria on Validation Set and of Masked Examiner Using Final Rules on 10% Random Sample of Cases^†

Uveitis Anatomic Class	Number Diseases	Validation Set Machine Learning Classification Criteria		Masked Examiner Evaluation of Final Rules on 10% Random Sample
		Accuracy	95% CI	Accuracy	95% CI
Anterior uveitides	9	96.7	92.4, 98.6	96.5	91.4, 98.6
Intermediate uveitides	5	99.3	96.1, 99.9	98.4	91.5, 99.7
Posterior uveitides	9	98.0	94.3, 99.3	99.2	95.4, 99.9
Panuveitides	7	94.0	89.0, 96.8	98.9	94.3, 99.8
Infectious posterior or panuveitides	5	93.3	89.1, 96.3	98.8	93.4, 99.9

^†Adapted from Standardization of Uveitis Nomenclature (SUN) Working Group. Development of classification criteria for the uveitides. Am J Ophthalmol 2021;228:96–116. Used with permission.

The Boolean expressions created by the machine learning phase then were translated into English-language “final rules”. Several formats for the final rules were considered, including: 1) achieving a minimal number of criteria from a list of criteria (e.g. the 1982 revised criteria for the classification of systemic lupus erythematosus,³⁴ the Systemic Lupus International Collaborating Clinics criteria for systemic lupus erythematosus,³⁵ and the 1987 revised American College of Rheumatology [ACR] criteria for the classification of rheumatoid arthritis36); 2) a point system (e.g. the 2010 ACR/European League Against Rheumatism [EULAR] criteria for rheumatoid arthritis³⁷ and the 2019 EULAR/ACR criteria for systemic lupus erythematosus³⁸); and 3) criteria plus exclusions (e.g. the International League Against Rheumatism [ILAR] criteria for juvenile idiopathic arthritis³⁹). The SUN criteria fit best with an approach similar to the ILAR criteria for JIA, and that approach was used.²⁵ The accuracy of the “final rules” was tested on a 10% random sample of each disease from the final database without the feature engineered variables by a masked observer. The masked observer’s results were compared to the machine learning results and to the consensus diagnoses. The accuracy of the final rules within anatomic class versus the consensus diagnoses (Table 3) were: anterior uveitides 96.5%; intermediate uveitides 98.4%; posterior uveitides 99.2%; panuveitides 98.9%; and infectious posterior or panuveitides 98.8%, respectively, indicating excellent performance.²⁵

The final product of the machine learning phase was a set of classification criteria for each of the 25 diseases under consideration (Supplemental Tables 1 through 26, available online at https://onlinelibrary.wiley.com/doi/10.1111/ceo.14175.^{28–31,40–60}

3.5. Consensus review and publication

The classification criteria developed after machine learning were reviewed by the SUN Executive Committee, the writing committees for the individual disease-specific manuscripts, the SUN Steering Committee, and the entire SUN Working Group. The SUN Working Group met in December 2019 to review the data, the criteria, and the draft manuscripts, resulting in over 80 separate suggestions for further analyses and possible revisions of the manuscripts. Additional analyses were performed in the first quarter of 2020, and the manuscripts revised accordingly. Publication of the 26 articles, one documenting the process and 25 the disease-specific criteria, were published in the August 2021 issue of the American Journal of Ophthalmology.^{25,28–31,40–60}

4. TERMINOLOGY AND CRITERIA

The SUN Working Group used several guidelines in its nosology. Infectious uveitides were defined as those for which there is evidence of active viral or microbial replication in the disease process and for which antiviral or antimicrobial treatment appears needed.^1,25 These diseases included cytomegalovirus anterior uveitis, herpes simplex virus anterior uveitis, varicella zoster virus anterior uveitis, acute retinal necrosis, cytomegalovirus retinitis, toxoplasmic retinitis, syphilitic uveitis, and tubercular uveitis.^{28,31,40,41,55–58} Diseases in which a prior infection triggers an auto-inflammatory (or autoimmune) reaction but in which microbial replication is not part of the pathogenesis were not considered infectious, as the treatment is not dependent on antiviral or antimicrobial therapy. Uveitis associated with epidemic reactive arthritis is an example of this latter phenomenon.⁴⁴ Post-rubella Fuchs uveitis syndrome also was considered a post-infectious disease and not an infectious disease, as the preponderance of current evidence suggests prior rubella infection may play a role, not active rubella replication.⁴² However, it is possible that newer molecular techniques could lead to a revision of this classification. In general, infectious diseases were named by the infection and the site of the infection (e.g. herpes simplex virus anterior uveitis, varicella zoster virus anterior uveitis, cytomegalovirus retinitis); the exception was the acute retinal necrosis syndrome as it represents a morphologic syndrome with two recognized viral causes.⁵⁵

The criteria for infectious uveitides generally allowed for diagnosis based on morphologic grounds as well as on virologic or microbiologic grounds (e.g. culture, pathology, or polymerase chain reaction techniques to identify microbes or viruses).^{25,40,41,55,56,58} However, the morphologic diagnosis had to have a high degree of accuracy on its own to be included.^{40,41,55,56,58} The decision was made by the SUN Working Group to enable the SUN criteria to be used for retrospective studies in which invasive procedures (e.g. paracentesis or vitrectomy) for microbes or viruses were not performed and from regions in which these invasive procedures are not performed routinely. The exceptions to permitting morphologic diagnoses only for infectious uveitides were cytomegalovirus anterior uveitis, syphilitic uveitis and tubercular uveitis, where evidence of infection was required, as these diseases have protean manifestations and the clinical presentation is not disease-specific.^31,57

Conversely, diseases associated with a systemic disease, but which are present in a minority of patients with the systemic disease, were termed “systemic disease-associated” uveitis. These diseases included JIA-associated chronic anterior uveitis, spondyloarthritis/HLA-B27-associated anterior uveitis, multiple sclerosis-associated intermediate uveitis, and sarcoidosis-associated uveitis.^43,44,47,59 Uveitides associated with a systemic disease in which the uveitis defined the disease (e.g. tubulointerstitial nephritis with uveitis syndrome, VKH disease) or were present in the large majority of patients (e.g. Behçet disease uveitis) omitted the word “associated”.^29,30,45

Traditional names for the uveitides were used. Acute posterior multifocal placoid pigment epitheliopathy (APMPPE) was used even though evidence now suggests that it may be a choriocapillaritis and not a primary disease of the pigment epithelium.⁴⁹ This decision was made to increase recognition of these criteria and permit their ease of use. Renaming of APMPPE likely would require a consensus conference of uveitis and retina experts. The term multifocal choroiditis with panuveitis (MFCPU) was retained even though the disease is a posterior uveitis (primary site of inflammation is the choroid) with a variable anterior uveitis and vitritis.⁵² There has been a sentiment among some to combine MFCPU and punctate inner choroiditis (PIC) into a single disease termed “multifocal choroiditis”, based on the following: 1) both diseases have chorioretinal spots and atrophic scars, 2) both diseases have choroidal neovascularization as a common structural complication; and 3) the occasional overlap of the two diseases in which a patient may have what appears to be PIC in one eye and MFCPU in the other.^52,53 The SUN Working Group elected to define the two diseases separately for several reasons. First, the term multifocal choroiditis is a confusing one; it can refer to either: 1) a clinical finding on examination (e.g. multifocal choroiditis in a patient with sarcoidosis); or 2) a class of diseases, the multifocal choroiditides (which include APMPPE, birdshot chorioretinitis, MFCPU, PIC, serpiginous choroiditis, etc.). To add to its use a specific disease would only increase the confusion as to what is being referenced. Second, the two diseases largely appear distinct with only occasional overlap.^{51,52,61–63} The spot sizes and predominant location differ; PIC typically does not have anterior uveitis or vitritis; other than choroidal neovascularization, the inflammatory structural complications seen in MFCPU (e.g. macular edema, optic neuropathy, posterior synechiae) are not seen in PIC; and the courses differ. Multifocal choroiditis with panuveitis is a chronic disease requiring immunosuppression to control the process, whereas PIC has a variable course including possible spontaneous remission, recurrent episodic disease, or chronic disease.^{51,52,61–63} The consensus of the SUN Working Group was that MFCPU and PIC should be classified as separate diseases and the occasional overlap cases described as such (i.e., MFCPU/PIC overlap).^51,52

5. IMPACT AND FUTURE EXPECTATIONS FOR THE SUN CLASSIFICATION CRITERIA

Although the SUN classification criteria were designed as classification criteria for research use, they also may provide guidance for clinicians in clinical care. In 2022 Mudie et al.⁶⁴ analyzed the concordance of clinical diagnoses with SUN diagnoses among the cases seen at a large uveitis practice. Of the 1143 patients seen by the practice between 1 January 2013 and 31 December 2020, 522 were diseases for which disease-specific criteria have been developed and published by the SUN Working Group. Among these 522 patients, there was agreement between the clinician caring for the patient and the SUN classification in 94.3%,⁶⁴ suggesting clinical utility for the SUN classification criteria. Furthermore, among the cases not included in the comparative analysis, 508 had a diagnosis included in the SUN classification schema (Table 1), namely an undifferentiated uveitis.⁶⁴ The SUN Working Group has recommended that uveitis cases not meeting criteria for a specific disease be classified as undifferentiated with the course, laterality, and anatomic class (e.g. undifferentiated bilateral chronic anterior uveitis) or anatomic class with posterior involvement details (e.g. undifferentiated bilateral panuveitis with choroiditis or undifferentiated bilateral panuveitis with retinal vasculitis).² Classifying these cases as undifferentiated uveitides, would result in 97% agreement on the classification of these cases, again suggesting clinical relevance for the SUN classification criteria.

Unlike the original SUN publication,² which has had over 15 years to evaluate its impact, the SUN classification criteria are too recently published to evaluate their impact. Nevertheless, the goal for their development is that they will be used widely to identify uveitic diseases for clinical and translational research in both current and future research in the field of uveitis.

Finally, classification criteria evolve over time. The criteria for rheumatologic diseases, such as rheumatoid arthritis,^36,37 systemic lupus erythematosus,^34,35,38 and spondyloarthritis,^65–69 all have been revised. As such it is possible, or even likely, that the SUN classification criteria will need revision at some time in the future, as newer imaging techniques become available and are applied to the uveitides and as genomic and other molecular biologic techniques (e.g. metagenomic deep sequencing) are applied to the uveitides. However, evaluation of these techniques should avoid circular reasoning and should use rigorous and validated definitions of the phenotype, such as the SUN classification criteria, in their evaluation of these techniques, findings, and testing results.