Table 2. Estimated effects in the mediation mechanism from low education to DNAm to HbA1c.
Method | Estimated Global Indirect Effect | Estimated Direct Effect | Estimated Total Effect | Estimated Proportion Mediated |
---|---|---|---|---|
HIMA | 0.03 | 0.16 | 0.20 | 0.17 |
HDMA | 0.13 | 0.07 | 0.20 | 0.65 |
MedFix | 0.07 | 0.13 | 0.20 | 0.36 |
BSLMM | 0.14 | 0.05 | 0.18 | 1.00 |
Pathway LASSO | 0.13 | 0.05 | 0.18 | 0.74 |
PCMA | 0.15 | 0.02 | 0.17 | 0.91 |
SPCMA | 0.17 | 0.00 | 0.17 | 1.00 |
HILMA | 0.03 | 0.00 | 0.03 | 1.00 |
PMED | 0.00 | 0.20 | 0.20 | 0.00 |
All estimates are adjusted for age, sex, race, and the estimated proportions of residual non-monocytes as fixed effects, along with methylation chip and position as random effects to address potential batch effects. We provide only point estimates, not interval estimates, because some of the methods are either not capable of producing interval estimates or do not provide the code for producing them in their software. Note also that for HIMA, HDMA, MedFix, and pathway LASSO, we used additional screening to reduce the number of mediators in advance for the sake of statistical and computational efficiency, so only n/log(n) ≈ 141 CpG sites were seen by the multivariable outcome model rather than 2,000.