Further meta-analysis

 

Other software, such as WinBUGS/OpenBUGS, can be useful for exploring a meta-analysis beyond the standard functions of StatsDirect. You might work with a statistician to build a Bayesian model of your meta-analysis and run simulations of it in WinBUGS.

 

A typical fixed effects odds ratio BUGS model:

 model

 {

 for( i in 1 : Num ) {

  rc[i] ~ dbin(pc[i], nc[i])

  rt[i] ~ dbin(pt[i], nt[i])

  logit(pc[i]) <- mu[i]

  logit(pt[i]) <- mu[i] + d

  mu[i] ~ dnorm(0.0,1.0E-6)

  }

 d ~ dnorm(0.0,1.0E-6)

 or <- exp(d)

 }

 

A typical random effects odds ratio BUGS model:

 {

 for( i in 1 : Num ) {

  rc[i] ~ dbin(pc[i], nc[i])

  rt[i] ~ dbin(pt[i], nt[i])

  logit(pc[i]) <- mu[i]

  logit(pt[i]) <- mu[i] + delta[i]

  mu[i] ~ dnorm(0.0,1.0E-5)

  delta[i] ~ dnorm(d, tau)

  }

 d ~ dnorm(0.0,1.0E-6)

 # Choice of priors for random effects variance

 #tau ~ dgamma(0.001,0.001)

 #sigma <- 1 / sqrt(tau)

 tau<-1/(sigma*sigma)

 sigma~dunif(0,10)

 delta.new ~ dnorm(d, tau)

 or <- exp(d)

 }

 

Warning – do not attempt this sort of modelling without the guidance of an expert statistician.

 

Meta-regression methods (such as the metareg macro for Stata) can be used to explore origins of heteroenegity in meta-analyses. To perform meta-regression you need: the effect estimate for each study; a confidence interval or standard error for the effect; and covariables that might explain the origins of the differences between the studies. There are numerous ways to go wrong in the use and interpretation of meta-regression, therefore please seek the help of a statistician (Thompson and Higgins, 2002).