Anna Brouwer

Chapter 2 40 If available, the mean deviation (MD) (n = 247 eyes) was recorded from the most recent visual field analysis (median 5 months, IQR 0 - 17 months) before the ERG measurement. Fluorescein angiograms (FA) (n=333 eyes) were scored by an experienced ophthalmologist (JdB) who was blinded regarding ERG results, using the FA scoring system of the Angiography Scoring for Uveitis Working Group (ASUWOG) 12 as an indication of the severity of previous inflammation. FAs with the highest overall score until ERGmeasurement (median 10months, IQR 3 - 64months) were used for statistical analysis. When FAs were performed on the same day as the ERG (n = 9 patients), there was at least 1 hour between the FA and ERG. 13 Statistical analysis For statistical analysis SPSS version 25.0 (SPSS Inc., Chicago IL) was used. Only uveitis eyes were included. First, ERG results were compared to reference values, and differences between groups A and group B were investigated. AMann-Whitney U test was used for continuous variables and the Pearson chi squared test or the likelihood ratio for categorical variables. Using Tukey’s fences k = 3, possible outliers were identified. Second, the most characteristic ERG abnormality result was selected and its relationship to clinical parameters was determined by using the Pearson chi-square test or the likelihood ratio for categorical variables, and a Student t-test or a Mann- Whitney U test for continuous variables, depending on the normality of data. Normality was tested using Kolmogorov-Smirnov normality test. A Spearman rank correlation coefficient was used to find possible correlations among variables. Third, a multiple imputation was performed to address missing values. Herein all statistically significant variables of the univariate analysis were included, as well as several auxiliary variables and the outcome variable. The automated program of SPSS was used, but continuous variables were restricted to avoid unrealistic imputed variables. The number of imputations was equal to the maximum percentage of missing data plus 1. Fourth, a generalized estimating equation (GEE) analysis was performed using the imputed data to correct for paired sampling of eyes and possible collinearity of parameters. Imputed data were used because a GEE analysis excludes eyes with one or more missing values, which would have led to biased results. Ordinal variables such as the amount of cells were transformed into categorical values (i.e. in the case of vitritis) to avoid complicating the model with multiple dummy variables. P values below 0.05 were considered statistically significant. All tests were 2-tailed.