[R-sig-ME] Estimating indirect contributions in a mixed-treatment comparison (a.k.a. network meta-analysis)

[Jouve, Thomas]

Dear readers,
Several packages, including metafor, use the mixed model framework to adjust for mixed-treatment meta-analysis, also known as network meta-analysis.
These analyses combine direct (traditional) effects seen in direct treatment comparisons with indirect effects obtained by simultaneously considering studies with a common treatment arm.
These packages have several options to display their results and output a large number of matrices, such as treatment effects, variance of these effects...
However, I fail to extract the respective direct and indirect treatment contributions. The most recent GRADE recommendations indeed require both estimates.
Koenig et al. (Stat Med, 2013, Visualizing the flow of evidence...) released a paper with a derivation of direct and indirect effects for the case of fixed-effect models. Does anyone have a clue to help me derive indirect treatment effects in a random-effect model ? Possibly using metafor output of the rma.mv function ?
Thanks in advance for your help and comments.
Thomas JOUVE

--
Thomas JOUVE
Interne de néphrologie - Centre d'Investigation Clinique
CHU Grenoble

	[[alternative HTML version deleted]]