Maxime Verhoeven

46 Chapter 3 To compare the DAS28 (as well as m-CDAI and HAQ) over time between the TCZ arms and the MTX+Pred arm, a linear mixed effects model was used. A random intercept at study level and at patient level was used to account for clustering of measurements within trials and within patients over time, respectively (for DAS28 as well as other time varying outcomes only). The MTX-strategy was similar in both trials and was used as reference arm in the analysis. The rationale of this approach is that within this reference arm, the effect of study (i.e., population) can be estimated assuming that differences in outcomes between patients from both trials can be regarded due to differences in trial population (as they are treated according to the same strategy). This is then used to correct for potential differences between studies and therefore, the three treatment strategy arms can be compared more validly. Treatment arm was categorised as TCZ, TCZ+MTX, MTX+Pred and MTX. Treatment arm and time were included in all analyses, and all models were corrected for DAS28 and HAQ at baseline, as well as for smoking status and RF-status and other variables that proved relevantly different between trials. We used a natural logarithm transformation for the m-CDAI in the analyses as this score was not normally distributed. 16 A logistic mixed effect model was used for the binary outcomes (occurrence of remission and safety outcomes). For SvdH scores over time, a Poisson mixed effect model was used to handle the skewed distribution of these data. Results of the 40 imputed datasets were pooled using Rubin’s Rule. 17 A separate analysis over the short term only (i.e., first 3 months) was performed, for DAS28/m-CDAI based outcomes, to specifically compare the early response. The influence of missing data imputation on the findings was investigated in a sensitivity analysis using the non-imputed data over 24 months. Analyses were performed in SAS version 9.4 and R version 3.4.3. RESULTS In total 553 patients were included in both trials. During the trials, 146 patients dropped out (80 in U-Act-Early and 66 in CAMERA-II), mostly due to an adverse event. Overall dropout rates were not different between the trials, neither for safety nor for inefficacy reasons. Statistically significant differences in baseline characteristics between the trials were observed for alcohol use, RF-status, baseline DAS28 and its individual components, Table 1. An overview of baseline characteristics separately for each treatment arm is provided in Supplementary Table S1 and Supplementary Table S2 provides baseline characteristics separately for the MTX arms that were used as reference.