OBJECTIVE: Observational data facilitate examination of treatment-effect heterogeneity, but the risk of bias is substantial. We highlight methodological considerations through an analysis of whether smoking affects response to TNF inhibitors (TNFi) in axial spondyloarthritis (axSpA).
METHODS: We used longitudinal data from the British Society for Rheumatology Biologics Register for Ankylosing Spondylitis. Participants fulfilling the ASAS criteria for axSpA, who started their first TNFi were eligible for analysis. To compare the impact of smoking status, weighted generalised estimating equations were used to examine changes in several continuous outcome measures, including BASDAI and ASDAS. Inverse-probability weights were used to account for differences in baseline covariates and excluded participants. We separately assessed response in the first 3 months to account for non-random dropout.
RESULTS: Of 840 participants that started on TNFi, 1,641 assessments from 627 individuals were analysed (69% male, mean age 46 years). 33% were current smokers and 30% ex-smokers. Ex- and current smokers had worse disease than never smokers at baseline. Accounting for these differences, response did not differ according to smoking status. Compared against never smokers, ex-smokers (β=-0.6; 95%CI -1.4, 0.3) and current smokers (β=-0.4; 95%CI -1.1, 0.4) had similar response in BASDAI, and ASDAS (ex: β=-0.1; 95%CI -0.5, 0.3; current: β=-0.01; 95%CI -0.4, 0.4), at 3 months.
CONCLUSIONS: TNFi response did not differ according to baseline smoking status in this UK cohort. Conflicting results from previous studies were likely due to methodological differences. This analysis highlights potential sources of bias that should be addressed in future studies. This article is protected by copyright. All rights reserved.
- axial spondyloarthritis
- ankylosing spondylitis
- treatment response
- inverse probability weights