Background Confounding by indication occurs when factors associated with the choice of therapy are also predictors of the studied outcome, and is generally considered the major limitation of non-randomized comparisons of different therapies. In RA, age, sociodemographic factors, disease activity, and medical history may influence the choice of e.g. a non-TNFi over a TNFi bDMARD. The degree of bias is difficult to assess since most studies have limited data on potential confounders, and residual confounding cannot be separated from true treatment effects.
Objectives To quantify the expected confounding by indication caused by baseline differences in patient characteristics at initiation of different bDMARDs in RA.
Methods All RA patients in the Swedish Rheumatology Register (SRQ) were linked to nationwide registers to assess whether a comprehensive list of covariates predicted the choice of bDMARD to such a degree that different rates of adverse events (AEs) would be expected. Among considered confounders were demographic variables, RA related factors (including RF, disease duration, HAQ, DAS28 w. components) and medical history (health care utilization and 20 specific conditions).
We used historical data on the 5 year risk of several AEs among RA patients starting any bDMARD 2005–2009 to predict the risk among RA patients (bionaïve and switching from a first TNFi) starting specific bDMARDs 2010–2014. Risk was modelled in logistic regressions, as a function of baseline characteristics but assuming no effect from the therapy itself.
Results Patients starting non-TNFi were older than those starting a TNFi, had lower socioeconomic status, higher disease activity and more often a history of diseases including malignancy, serious infections, and diabetes. These factors were in general also significant predictors of AEs, and the predicted proportions of all AEs were substantially higher for non-TNFi compared to TNFi bDMARDs, highest for rituximab. Within the TNFi-group, only minor differences were seen. Age was a strong confounder, and standardizing to the age/sex-distribution of the TNFi group reduced the group difference dramatically.
Conclusions Even if there are no true differences in risk by bDMARD, confounding by indication will make the non-TNFi drugs appear less safe than the TNFi. A simple adjustment for age and sex will reduce this confounding dramatically, but residual confounding is still expected to give higher rates of AEs, and comparisons should be adjusted for medical history and comorbidities when possible.
Acknowledgements The ARTIS registry has been, or is, supported by agreements with Abbvie, BMS, MSD, Pfizer, Roche, Samsung Bioepis, and UCB.
Disclosure of Interest T. Frisell: None declared, E. Baecklund: None declared, K. Bengtsson: None declared, D. Di Giuseppe: None declared, H. Forsblad-d'Elia: None declared, J. Askling Grant/research support from: Abbvie, UCB, Pfizer, Merck, Samsung, Roche, Lilly