Reply to Comment on: SUN Classification Criteria for Vogt-Koyanagi-Harada Disease

Specificity is the ability of a test or criteria to exclude those without the disease. It is a probabilistic, not an absolute, measure. The Standardization of Uveitis Nomenclature (SUN) “Classification Criteria Project” machine learning optimized accuracy rather than sensitivity and specificity per se. The accuracies of the SUN criteria for panuveitides overall and for early-stage Vogt−Koyanagi−Harada (VKH) disease were excellent, 93.3% and 92.3%, respectively. 1 Standardization of Uveitis Nomenclature (SUN) Working Group Development of classification criteria for the uveitides. Am J Ophthalmol. 2021; 228: 96-116 Abstract Full Text Full Text PDF PubMed Scopus (25) Google Scholar ,2 Standardization of Uveitis Nomenclature (SUN) Working Group. Classification criteria for Vogt-Koyanagi-Harada disease. Am J Ophthalmol. 2021; 228: 205-211 Abstract Full Text Full Text PDF PubMed Scopus (17) Google Scholar Precision, the machine learning equivalent to specificity, was 94.4% for early-stage VKH. 1 Standardization of Uveitis Nomenclature (SUN) Working Group Development of classification criteria for the uveitides. Am J Ophthalmol. 2021; 228: 96-116 Abstract Full Text Full Text PDF PubMed Scopus (25) Google Scholar The SUN VKH criteria included both full-fledged early-stage VKH with panuveitis and neurologic involvement and the more limited “Harada disease” presentation with only exudative detachments; but the criteria for the latter required an exudative detachment with the classic features on fluorescein angiography and/or optical coherence tomography for classification as early-stage VKH. Comment on “SUN Classification Criteria for Vogt-Koyanagi-Harada Disease”American Journal of OphthalmologyVol. 244PreviewWe read with interest the article by the Standardization of Uveitis Nomenclature (SUN) Working Group regarding classification criteria for Vogt−Koyanagi−Harada (VKH) disease.1 The SUN Working Group has made a remarkable effort to provide several insights into the classification of uveitic entities. They used machine learning algorithms to define classification criteria in order to standardize the classification of inflammatory diseases for research purposes. As it is clearly stated by the authors, these criteria are classification criteria and they are not intended to be diagnostic criteria. Full-Text PDF