Background: Recently, using a counterfactual framework, a causal mediation analysis has been formalized to decompose the total effect of a time-fixed exposure on an outcome into four components that can be loosely defined as being components due to mediation only, interaction only, mediated interaction and neither. The interpretation of the estimated effect sizes is challenging when these components of the total effect are of the opposite sign compared with each other. Particularly, a negative mediated interaction might be intuitively difficult to conceptualize and, so far, lacks an easy-to-understand biological or mechanical interpretation. Methods: In this paper, we focus on negative mediated interaction, and propose an interpretation using biological examples. For negative mediated interaction to be present, the effect of interaction on the outcome and the effect of the exposure on the mediator should be in opposite directions. Results: In this article, we give examples of biological and biochemical processes that may exhibit negative mediated interaction, such as drug treatment in clinical practice, allosteric effects of enzymes, different adaptations in the cardiovascular system and its effect on brain health, and antibiotic drug-drug interactions. Conclusions: We aim to make researchers realize that negative-effect estimates might reflect relevant biological processes in the mechanism under study.

Additional Metadata
Keywords biology, causal inference, clinical medicine, four-way decomposition, interaction, mediation
Persistent URL dx.doi.org/10.1093/ije/dyy252, hdl.handle.net/1765/119083
Journal International Journal of Epidemiology
Citation
Mutlu, U, Ikram, M.A, & Ikram, M.K. (2019). Clinical interpretation of negative mediated interaction. International Journal of Epidemiology, 48(4), 1286–1293. doi:10.1093/ije/dyy252