Traditional meta-analysis usually focuses on pair-wise of treatment, and calculating effect size from mean and SD, or others. But when it comes to multiple treatments, for instance, three distinct drugs on a disease, it seems to be useless. So how do I contrast multiple interventions simultaneously, combining a synthesized effect-size across different groups?
I agree with Elmer! This is exactly how you'd go about evaluating the effectiveness of multi-valued treatments in meta-analysis. In Stata there is a user-written program called mvmeta (findit mvmeta). In addition, there is a user-written command that will plot out the relationships and other relevant metrics (see attached link).
I hope this helps
Ariel
http://www.mtm.uoi.gr/index.php/stata-routines-for-network-meta-analysis
Agreed, network meta-analysis is the tool you require! Lots of helpful papers out there on how to do them (see http://www.nicedsu.org.uk/Evidence-Synthesis-TSD-series(2391675).htm), software to do them (see the same refs as well as http://www.ncbi.nlm.nih.gov/pubmed/25267416, http://www.ncbi.nlm.nih.gov/pubmed/24098547,http://www.ncbi.nlm.nih.gov/pubmed/25541687, amongst several other good papers), and how to report them (http://www.ncbi.nlm.nih.gov/pubmed/26030634). You'll also find a great repository of related methods papers at www.cmimg.org. Best of luck!
The GPM framework for network meta-analysis in the software MetaXL (freely available from www.epigear.com) is a framework that finally brings the analysis platform to the mainstream researcher. In addition it avoids the multitude of problems associated with Bayesian and multivariate frequentist methods.
In terms of the Bayesian approaches, they are not of much use to the mainstream researcher because this involves writing a directed acyclic graph (DAG) model for general-purpose Markov chain Monte Carlo (MCMC) software such as WinBUGS. In addition, prior distributions have to be specified for a number of the parameters, (the choices being arbitrary, more so for priors on heterogeneity in random-effects models) and the data have to be supplied in a specific format. Together, the DAG, priors, and data form a Bayesian hierarchical model. To complicate matters further, because of the nature of MCMC estimation, overdispersed starting values have to be chosen for a number of independent chains so that convergence can be assessed (Brooks and Gelman 1998). Currently, there is no software that automatically generates such Bayesian models, although there are some tools to aid in the process and indeed the only attempt at automation requires that arm-level outcome data are available, and this is usually unavailable (van Valkenhoef, Lu et al. 2012).
Therefore, we created the GPM framework for network meta-analysis for the mainstream researcher and this can be downloaded freely from www.epigear.com
See details in the User Guide attached
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis