Anna Brouwer

Effects of DTL position on amplitude and implicit time An extended ISCEV protocol was measured twice, consisting of stimulus strengths that increase with approximately 0.5 log units and range from 0.3 to 10.0 cds/m 2 for the light-adapted ERG (LA), including a 30 Hz flicker response (LA, 3.0 cds/m 2 ). For analyses, averages were used of two results, each consisting of five sweeps. In cases when sweeps with a trend were accepted during recording, these sweeps were toggled, or removed. After 10 minutes of light adaptation, the first LA ERG was recorded with the DTL electrode in the LLP, and the second with the FP (see Figure 1). The DTL position was checked prior to each measurement and adjusted if necessary. Analyses For statistical analysis R. Studio version 1.0.143 was used. To compare differences between the two DTL positions, a Wilcoxon signed rank test was used (data not normally distributed as tested with histograms, Q-Q-plots and Shapiro-Wilk test). We added a random value between -0.25 and 0.25 ms to each implicit time, because our equipment measures every 0.5 ms. By adding this random value, the implicit time becomes a true continuous variable which has statistical advantages for calculating more reliable confidence intervals. The implicit time difference and amplitude ratio of the two DTL positions were calculated and plotted. To evaluate the effect of the two DTL positions on the percentage of patients that would fall outside normal limits, we compared the results to our reference values. These reference values were obtained with the DTL in LLP and were previously described. 18 To investigate the amount of reliabilitybetween the twomeasurements, intraclass correlation coefficients (ICCs) were calculated (two-way model, type consistency, unit average). For visualization of the amount of agreement between the two positions, Bland-Altman plots were made as well. To evaluate the amount of relative variability of the ERG data at the two DTL positions, we calculated the coefficients of variation (CoV, sd/mean), also known as relative standard deviation. The data were transformed (square root) to fit a normal distribution. This was necessary since the results of uveitis patients can range from normal to abnormal which gives a skewed distribution. 18 Normality was evaluated using Shapiro Wilk tests, histograms and Q-Q-plots. We used the R package cvequality (Version 0.1.3; Marwick and Krishnamoorthy 2018) to test for significant differences in CoV. Using Bonferroni’s correction, we defined p-values of < 0.006 as statistically significant. All significances were two-tailed. 143 7