Table 6.

Relative impacts of individual characteristics on the estimated effects of having BMI ≥ 30 versus BMI < 30 on HDL (ln mmol/l), according to a regression model with interactions

	Univariate (95% CI)	Multivariate (95% CI)
Interaction effects
Risky alcohol consumption	0.911 (0.737–1.125)	0.906 (0.730–1.124)
Smoker	0.897 (0.825–0.975)	0.907 (0.832–0.988)
Physical activity score (10,000 units, centered)	1.013 (0.955–1.073)	1.009 (0.952–1.070)
Energy intake (1000 kcal/day, centered)	0.978 (0.935–1.023)	0.964 (0.915–1.015)
Primary education	1.000 (ref)	1.000 (ref)
Secondary education	1.011 (0.888–1.151)	1.018 (0.893–1.161)
Tertiary education	1.090 (1.009–1.177)	1.085 (1.003–1.174)
Male	1.025 (0.963–1.091)	1.057 (0.983–1.135)
Age 45–49	0.960 (0.881–1.046)	0.967 (0.884–1.058)
Age 50–59	1.000 (ref)	1.000 (ref)
Age 60–68	1.012 (0.952–1.076)	0.999 (0.936–1.065)
Main effect of obesity	–	0.857 (0.810–0.906)

Heterogeneous effects of obesity on HDL levels (ln mmol/l), based on interaction models where all variables previously used in “step 1–3” were entered as main effects and either one or all variables previously used in “step 4” were interacted with obesity. Since univariate models (including only one interaction term) were run separately, there is no common main effect of obesity to be reported. Effects are multiplicative, as the outcome is measured in logarithms and we have exponentiated coefficient estimates