Maxime Verhoeven

45 An indirect comparison of treat-to-target strategies in early RA Assessments For the current analyses, individual patient data from the U-Act-Early and CAMERA-II trials was used. DAS28 (using ESR) over 24 months was defined as the primary outcome. To account for the specific reduction of acute phase reactant (APR) by TCZ, a disease activity outcome not containing APR was used as secondary outcome. For this, we used a modification of the Clinical Disease Activity Index (m-CDAI) 13 as one of the components of the CDAI (VAS physician) was not available in the CAMERA-II dataset. We used the same formula and rescaled the resulting score to the original range using the formula: 76*(TJC28+SJC28+VASpatient/66). Other secondary effectiveness outcomes were (1) DAS28 remission (DAS28<2.6), (2) m-CDAI remission (m-CDAI<2.8), (3) cumulative occurrence and persistence of remission as calculated with the continuity rewarded (ConRew) score, 14 (4) physical function using health assessment questionnaire (HAQ) scores and (5) radiographic progression, all over 24 months. Radiographs of both trials were scored by the same professional reader, using the Sharp van der Heijde (SvdH) method. Safety outcomes were: occurrence of ≥1 event for infection and for elevated levels (above upper limit of normal) of alanine aminotransferase (ALT; ≥55 units per litre) and aspartate aminotransferase (AST; ≥40 units per litre), 15 and drop out due to an adverse event (AE). These outcomes were selected because these were deemed clinically relevant and data was similarly collected in both trials. Statistical analyses Continuous variables were described using means with standard deviations (SD) or medians with interquartile ranges (IQR), as appropriate. Frequencies and proportions were calculated for categorical variables. Differences between the trial populations in baseline characteristics were evaluated and tested using t-test for continuous outcomes and chi-square test or Fisher exact test for categorical outcomes, or non-parametric alternatives where appropriate. To account for missing data in relevant baseline data, multiple imputation was used. This was done separately for the dataset of each study (using the same approach and predictor variables), to allow for possible modification of the predictor effects in the imputation model by trial. We used predictive mean matching to impute data, except for smoking status and rheumatoid factor (RF) status for which logistic regression was used. For all imputations the following predictor variables were used: age, DAS28, HAQ, SvdH score and smoking status, all at baseline; gender, treatment strategy, RF-status, ConRew score, and occurrence of ≥1 event for elevated ALT, elevated AST, and infection, respectively, and drop out due to an AE. Multiple imputation was performed in R using the MICE package, yielding 40 imputed datasets. 3