Drug safety analyses in a rheumatoid arthritis registry: application of different approaches regarding timing of exposure and confounder measurement

Patient registry data serves an increasing role in drug safety and comparative effectiveness research, but registry databases often do not contain confounder information measured at the same time that treatments begin. This study evaluated a set of approaches for estimating confounder values at treatment initiation using actual data from a rheumatoid arthritis (RA) registry to examine the risk of infection associated with different biologic DMARDs (bDMARDs). We examined the risk of infection associated with starting a TNF inhibitor (TNFi) versus any of the other non-TNFi bDMARDs. Different confounder assessment approaches were tested. All approaches were tested in Cox proportional hazard regression models that used a propensity score (PS). The confounder of interest was the disease activity score (DAS28-CRP). The confounder assessment approaches utilized different temporal relationships between the DAS28-CRP measurement and the start of the treatments of interest. We included 219 subjects with RA with 269 initiations of either a TNFi or a different bDMARD or both. Among this group, 305 infections were reported and confirmed through chart review. The hazard ratio (HR) for the risk of infection associated with use of a non-TNFi bDMARD ranged from 1.17 to 3.03 using 13 different approaches; only the approach with the highest HR produced results significantly different than one, but this approach included the fewest subjects and infections. The relative risk of infection for TNFi and other non-TNFi bDMARDs was similar using various approaches regarding which DAS28-CRP score should be used as the baseline measure in adjusted analyses.