The understanding of the interaction term probably presents one of the greatest challenges for the clinical researcher. The foe arises from its application, since it could refer different situations. Firstly, we could mean interaction when assessing for the joint influence of two, or more, exposure causes on their risk of disease, which is a biological or causal interaction. It could also refer the variation in the effect measure for the exposure cause under study across levels of another factor, which is an effect modification. Finally, we could also refer the necessity for a product term in a linear-based regression model, that is the statistical interaction.
Usually, both, biological interaction and effect modification are addressed for by fitting an interaction term in a chosen regression model. This would imply that the relation between the response variable and the cause exposure and/or other factor variables is no longer additive. However, this departure from additivity could also depend on the statistical model used.
The aim of this presentation is to emphasize the friendly use of statistical interactions as the basis for assessing biological interactions and/or effect modifications. Practical examples will be used to illustrate the proper use of linear-based regression models for both situations in rheumatology research.
Disclosure of Interest None declared