Maxime Verhoeven

57 An indirect comparison of treat-to-target strategies in early RA In Supplementary Table S3 the results are shown of the sensitivity analyses using non- imputed data. The outcomes were generally in line with the outcomes based on imputed data. Statistical significant differences were shown for DAS28 over time, which were in favour of the TCZ strategy arms compared to the MTX+Pred strategy arm. Missing data was mostly found in the CAMERA-II trial (especially for the confounders RF status and smoking status), as this resulted in exclusion of patients with a more favourable prognosis in the CAMERA-II trial (as apparent for instance from the lower DAS28, SJC, TJC and HAQ in excluded patients 5.6 vs 5.7, 10 vs. 11, 10 vs. 11, and 1.06 vs. 1.11), more favourable outcomes were found in the TCZ treatment strategies (U-Act-Early) by using the non-imputed data. Therefore we consider the analysis imputing these baseline characteristics as the more valid results. Supplementary Table S3 Difference in effectiveness and safety outcomes between strategy arms over 24 months based on non-imputed data. TCZ+MTX vs. MTX+Pred TCZ vs. MTX+Pred Effectiveness DAS28, mean difference (95%CI) -0.51 (-0.84 to -0.19) -0.42 (-0.75 to -0.09) m-CDAI, proportional mean difference (95%CI) 1 0.07 (-0.20 to 0.33) 0.22 (-0.05 to 0.49) DAS28 remission, RR (95%CI) 1.10 ( 1.00 to 1.22) 1.09 ( 0.98 to 1.21) m-CDAI remission, RR (95%CI) 0.97 (0.88 to 1.08) 0.92 (0.83 to 1.02) ConRew, mean difference (95%CI) 3.00 (-2.68 to 8.65) 4.20 (-1.50 to 9.91) HAQ score, mean difference (95%CI) 0.01 (-0.14 to 0.18) 0.08 (-0.08 to 0.24) SvdH score, IRR (95%CI) 0.87 ( 0.36 to 2.11) 1.43 ( 0.60 to 3.44) Safety Occurrence of ≥1x elevated ALT, RR (95%CI) 1.89 ( 0.72 to 5.01) 1.09 ( 0.37 to 3.18) Occurrence of ≥1x elevated AST, RR (95%CI) 1.06 ( 0.25 to 4.60) 0.41 ( 0.08 to 2.13) Occurrence of ≥1 infection, RR (95%CI) 2.62 ( 0.52 to 13.28) 2.61 ( 0.53 to 13.66) Drop out due to AE, RR (95%CI) 1.22 ( 0.36 to 4.15) 1.34 ( 0.40 to 4.48) Random effects for the variables in the model were evaluated and retained if the model fit was improved, as judged by a decrease in Bayesian information criterion or a change in one of the estimates of treatment effect of at least 10%. Random effects for the following variables were used in the models: DAS28, HAQ, RF-status, smoking status, all at baseline and time. Negative values indicate lower values for continuous outcomes (i.e., DAS28, m-CDAI, HAQ) in TCZ-based strategy arms compared to the MTX+Pred arm. (I)RR >1 indicate higher occurrence of the outcome (i.e., DAS28 remission, m-CDAI remission, drop out due to AE) in TCZ strategy arms compared to the MTX+Pred strategy arm. 1 log transformed values. Multiplying the above regression coefficients by 100 leads to a difference on a modified percentage scale. 16 Outcomes based on adjusted analyses using non-imputed data. TCZ+MTX= tocilizumab + methotrexate strategy arm; MTX+Pred= methotrexate + prednisone strategy arm; TCZ= tocilizumab + placebo strategy arm. DAS28= disease activity score assessing 28 joints; CI= confidence interval; m-CDAI= modified clinical disease activity index; RR= relative risk; ConRew= continuity rewarded; HAQ= health assessment questionnaire; SvdH= Sharp van der Heijde; IRR= incidence rate ratio; ALT= alanine aminotransferase; AST= aspartate aminotransferase; AE= adverse event. 3