I have multiple outcome measures (HDL, LDL, TG & Total cholesterol) that are measured before, at the end, and a month after an intervention and I would like to perform meta-analysis.
First of all, a meta-analysis is performed for each outcome separately. A network meta-analysis would be carried out if you are comparing more than 2 treatments across many different studies, but this would, again, be done separately for each outcome. Hence the 'network' refers to the network of treatments rather than outcomes. If you have some variables that you believe could be confounding the meta-analysis or network meta-analysis, you may conduct a meta-regression, but these confounders are not themselves outcomes. Meta-regressions can be carried out fairly easily on WinBugs.
Multivariate meta-analysis is an extension of the classical meta-analysis with one outcome to more than one outcomes. Multivariate meta-analysis allows you to incorporate the correlation that might be present among multiple outcomes.
Network meta-analysis extends the classical meta-analysis with one relative treatment effect to more than one relative treatment effects on the same outcome as long there are network is connected. E.G suppose there a n1 studies comparing A-B treatment effect and n2 studies comparing A-C treatment effect and no studies have studied B-C treatment effect, then a network meta-analysis would enable you to estimate it indirectly.
At the relative-effects levels, network meta-analysis can be thought of as multivariate meta-analysis where some of the treatment-effects are missing at random.
See Ian R. W 2009. Multivariate random-effects meta-analysis with Stata
Ian R. W 2012. Consistency and inconsistency in network meta-analysis: model estimation using multivariate meta-regression
Your data seem amenable to a standard meta-analysis as there is a pre-post element and you may want to compute a mean gain.
In terms of network meta-analyses, there are three frameworks through which a such a meta-analysis is performed:
a)Bayesian modelling
b)Multivariate frequentist modelling
c)Generalised pairwise modelling
The last is a new approach that is very simple and straightforward for researchers and a free software is available to implement this (MetaXL) at www.epigear.com