Abstract
Purpose :
Accurate differentiation between uveitis, vitreoretinal lymphoma, and endophthalmitis can be difficult due to similar clinical presentations, but is critical for the appropriate treatment of these vision-threatening and potentially life-threatening conditions. Several studies have reported elevated vitreous IL-10 in lymphoma vs. elevated vitreous IL-6 in uveitis and endophthalmitis. It has also been demonstrated that an IL-10/IL-6 ratio > 1 is suggestive of lymphoma and that an IL-10/IL-6 ratio < 1 is suggestive of uveitis. In this study, we attempt to differentiate endophthalmitis from uveitis and lymphoma by aqueous and vitreous IL-6 and IL-10 levels.
Methods :
Patient aqueous and vitreous levels of IL-6 and IL-10 were extracted retrospectively from an NEI ocular immunopathology database. Levels associated with definitive diagnoses of endophthalmitis, uveitis, or lymphoma in the NIH electronic medical record were included in the study. IL-6 and IL-10 levels between groups were compared by the Kruskal-Wallis test and Dunn test after log10 transformation. Using the Python scikit-learn library, a gradient-boosted decision tree was implemented to classify endophthalmitis vs. uveitis/lymphoma from IL-6 and IL-10 levels and tested with 10-fold cross-validation.
Results :
157 patients with definitive diagnoses and aqueous or vitreous IL-6 levels (12 endophthalmitis, 64 uveitis, 81 lymphoma) were extracted from the database. 125 of the 157 also had aqueous or vitreous IL-10 levels (5 endophthalmitis, 40 uveitis, 80 lymphoma). IL-6 levels were highest in the endophthalmitis group and significantly different between all 3 groups (p = 8.8e-5, 0.037, 2.9e-4 for endophthalmitis vs. lymphoma, endophthalmitis vs. uveitis, and lymphoma vs. uveitis respectively). IL-10 levels were highest in the lymphoma group but only significantly different between lymphoma vs. uveitis (p = 7.8e-6). 10-fold cross validation of the gradient-boosted decision tree showed 96±4% accuracy in classifying endophthalmitis vs. uveitis/lymphoma.
Conclusions :
Despite a small sample size, this study demonstrates significant differences in IL-6 and IL-10 between diagnoses as well as 96±4% accuracy in discriminating endophthalmitis from uveitis and lymphoma. Our algorithm shows potential for improving accurate differentiation between uveitis, vitreoretinal lymphoma, and endophthalmitis.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.