Figure 2

Causal diagram representing the four-way decomposition of the association between upper-level nonmanual workers, remote work, and coronary heart disease incidence. The exposure (upper-level nonmanual worker), mediator (remote work), confounders (sex, age, educational attainment, and household income), and outcome (coronary heart disease [CHD] incidence) were defined. Logistic regression models were fitted with a four-way decomposition for the outcome (as a function of the exposure, the mediator, exposure-mediator interaction, and confounders) and the mediator (as a function of the exposure and confounders). The total effect (TE) of the exposure, in the presence of the mediator with which the exposure may interact, on the outcome is decomposed into four components due to just mediation (pure indirect effect, PIE), just interaction (reference interaction, INTref), both mediation and interaction (mediated interaction, INTmed), and neither mediation nor interaction (controlled direct effect, CDE): i.e., TE = CDE + INTref + INTmed + PIE. The overall proportion of the total effect explained by the mediating effect (“total indirect effect”) was estimated using these four components, as follows: (INTmed + PIE)/TE.