Maxime Verhoeven

88 Chapter 5 in diagnosis might result, and damage might already have occurred. These patients might have passed this ‘window of opportunity’, and thus MTX treatment might have less (additional) effect. 12 Actually, indeed in IPD of this study, early RA patients with lower baseline DAS28 more often had joint damage at baseline (Table S7), suggesting a late diagnosis of RA because of the milder symptoms. For established RA patients, effectiveness of TCZ-monotherapy approximates that of TCZ+MTX for the subgroup with longer disease duration at baseline (i.e., high-level subgroup). This might indicate that, with a longer disease duration more often MTX is (no longer) effective, when a csDMARD has shown to be insufficiently effective, consequently the effectiveness of the combination treatment predominantly relies on the added bDMARD, here TCZ. 13 Baseline GC use may affect radiographic progression, however, this was not different between subgroups for disease duration (55% in high-level group vs. 58% in low-level), as well as for the subgroups, indicating that baseline GC use probably has not biased the difference we found between the (sub)groups in our study. Overall results were in line for total SvdH and erosions scores, but less clear for JSN scores. This may be due to the fact that JSN is a slow process, occurring far beyond the 2 year follow-up, and can additionally be influenced by genetic and mechanical factors (e.g., osteoarthritis), 14 whereas erosion formation is mainly inflammatory driven and more specific for RA. The current study naturally has limitations. First, radiographs were assessed by different readers in the original trials, and radiographs were not re-assessed specifically for the current analysis; a substantial portion of radiographic data was missing, although the influence on our subgroup analysis was probably limited, as patients with and without information on radiographic progression were not different regarding baseline joint damage, disease duration and disease activity (as well as regarding age, gender and rheumatoid factor positivity). Second, although we used IPD of multiple studies, the total sample size is still relatively low for detecting effect modification. For the SURPRISE trial, unfortunately data sharing was not possible due to legal considerations. 8 However, despite the relatively low sample size, effect modifiers were detected. Even when we tested both interactions in the model (i.e., baseline SvdH*treatment & baseline DAS28*treatment) in early RA, both predictors were still modifiers. Outcomes of the analyses, based on 4 strata (Supplementary Table S7), were in line with reported outcomes. Third, radiographic progression is nowadays less, due to the availability of better treatment, resulting in absolute chances of preventing radiographic progression varying between 12% and 0%, indicating that differences in preventing radiographic progression may not be clinically relevant in all subgroups. However, using combined data may thus be a suitable means to identify relevant treatment effect estimates in subgroups. We also considered to apply the minimal clinically important difference as