Selection Collider Bias in Rheumatology
Conditions affecting the selection of individuals for a study are a common source of bias in epidemiological and medical research. A recent trend in epidemiology has been to link selection conditions to the concept of collider variables. The effects of selection and associated collider variables depend on the composition of the study population, the study selection process, the target of inference, and methods of data collection and analysis. We give here an introduction to bias arising from selection effects, termed selection collider bias, and discuss ways of avoiding it. We provide illustrations related to two apparent paradoxes in rheumatology that involve smoking. The first paradox is that smoking, a risk factor for rheumatoid arthritis (RA), appears to be protective for disease progression in RA. The second is that smoking is a risk factor for psoriatic arthritis (PsA), but in some studies involving persons with psoriasis it appears to reduce the risk of PsA. Using data from the Gladman-Krembil Psoriatic Arthritis Research Program we examine the effect of smoking on the development of PsA in individuals with psoriasis and resolve the paradox. collider variables, confounding, directed acyclic graphs, observational data, selection effects.
Authors
- Jerald F. Lawless (ORCID: https://orcid.org/0000-0002-3192-0470)
- Richard J. Cook (ORCID: https://orcid.org/0000-0002-1414-4908)
- Lily Zou
Institutions
- University of Waterloo (CA)
Publication Details
- Journal
- The Journal of Rheumatology
- Published
- 2026-11-15
- DOI
- https://doi.org/10.3899/jrheum.2026-0177
- Primary Topic
- Advanced Causal Inference Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00