Table 6.
Mediation analysis using a structural equation modeling.
| F5L family study | MARTHA study | |||
|---|---|---|---|---|
| triglycerides <1.12 mmol/L N = 86 | triglycerides ≥1.12 mmol/L N = 113 | triglycerides <1.12 mmol/L N = 222 | triglycerides ≥1.12 mmol/L N = 102 | |
| Effect size of triglyceride levels without BMI adjustment | −0.12 | −0.26 | 0.022 | −0.25 |
| 95% CI | (−0.31, 0.081) | (−0.37, −0.14) | (−0.11, 0.15) | (−0.44, −0.062) |
| P-value | 0.24 | 8.4 × 10−6 | 0.74 | 0.0091 |
| Effect size of triglyceride levels with BMI adjustment | −0.097 | −0.25 | 0.038 | −0.24 |
| 95% CI | (−0.30, 0.11) | (−0.36, −0.13) | (−0.099, 0.17) | (−0.42, −0.046) |
| P-value | 0.34 | 2.4 × 10−5 | 0.59 | 0.036 |
Mediation analyses were performed using structural equation modelling in both F5L family and MARTHA studies assuming the pathway to be: BMI (B) → triglyceride levels (T) → methylation levels (C). Models were adjusted for age and sex and cell type proportion accordingly in each dataset. RMSEA <0.08 in general shows good fit. RMSEA, root mean square approximation. The non-linear relationship was modelled with a GAM model adjusted on age, sex and cell type proportions (RUV component in the F5L family study). A random effect was added to the model for the analysis of the F5L family study data to adjust for the relatedness among the family members.