Figure 1. A multiple mediation model illustrating potential mediatory roles of adiponectin, leptin, and CRP concentrations in the smoking-diabetes association. This figure demonstrates how we estimated the indirect effects through the proposed mediators of being in one smoking category (X₁, X₂, X₃, or X₄) on DM incidence (Y) relative to the never smokers category. The ‘c’ paths represent the overall association—relative total effect—of being in a smoking category with DM incidence, and are equivalent to the unstandardized regression coefficients (β). Mediation is said to occur when (part of) this overall association is explained by a hypothesized mediator variable: adiponectin (M₁), leptin (M₂), and CRP (M₃). Associations of the potential mediators with each smoking category (a-paths) and their associations with DM incidence independent of one another and of the smoking category (paths b₁, b₂, and b₃) are assessed simultaneously. The product of regression coefficients a’s and b’s (ab’s) quantifies the relative indirect effect of being in one smoking category, relative to the reference, on DM incidence through the mediators. For instance, the relative indirect effect of being an ex-smoker (X₁) relative to never smokers on DM incidence via adiponectin (M₁), leptin (M₂), and CRP (M₃) is given by products a₁₁b₁, a₂₁b₂, and a₃₁b₃, respectively. The part of the relative total effect (c₁) not explained by these mediating paths is the relative direct effect (c′₁) of being an ex-smoker on DM incidence. Similarly, the relative total direct and indirect effects of being in the other smoking categories on DM incidence relative to the never smokers can be represented by the respective paths linking the independent, dependent, and potential mediator variables.
DM, diabetes mellitus; CRP, C-reactive protein.