Calculating I2 from a Bayesian Model Network metanalysis

Question

I am conducting a network metanalysis using both Bayesian (using JAGS run throug R with the package R2JAGS) and frequentist model and, as a form of heterogeneity, I would like to calculate I2.

However, after computation, I have complete different levels between the Bayesian and the frequentist model (as an example, the frequentist model have 0% I2 whereas the Bayesian have as much as 90% for the same outcome and same comparisons).

How is this possible? I don't understand if it is normal or is an error from the code I'm using.

Here is the code for the Bayesian model (is a random effect design-by-treatment interaction model):

    model{                            
for(i in 1:ns){                     
  h[i,1] <- 0    
  delta[i,1] <- 0           
  mu[i] <- log(p[i,1])
  p[i,1] ~ dunif(0,1)     
  for (k in 1:na[i]) {             
    r[i,k] ~ dbin(p[i,k],n[i,k]) 
   }
  for (k in 2:na[i]) {             
    log(p[i,k]) <- mu[i] + min(delta[i,k], -log(p[i,1]))
    delta[i,k] ~ dnorm(md[i,k],taud[i,k])
    md[i,k] <-  d[t[i,k]] - d[t[i,1]] + w1[i,k]*fw1 + w2[i,k]*fw2 + sh[i,k]
    taud[i,k] <- tau.d *2*(k-1)/k
    h[i,k] <- delta[i,k] - (d[t[i,k]] + d[t[i,1]] + w1[i,k]*fw1 + w2[i,k]*fw2)
    sh[i,k] <- sum(h[i,1:(k-1)])/(k-1)
   }
 } 


sd ~ dunif(0,2)    
tau.d <- pow(sd,-2)  

fw1 ~ dnorm(0, tau.d)
fw2 ~ dnorm(0, tau.d)  

d[1]<-0
for (k in 2:nt){  d[k] ~ dnorm(0,.1) }

for (c in 1:(nt-1)) {  
  for (k in (c+1):nt)  { 
      lnRR[c,k]<- d[c] - d[k]
      RR[c,k]<- exp(d[c]-d[k])
      lnRR[k,c]<- d[k] - d[c]
      RR[k,c]<- exp(d[k]-d[c])
   }  
 }

rk_bad<-rank(d[])

And here is the code I use for calculating I2 from the posterior distribution:

tau2<-res$BUGSoutput$sims.matrix[,"tau.d"]
I2.cons<-tau2/(tau2+mean(na.omit(sei_All_cause^2)))
median(tau2)
median(I2.cons)

Thanks in advance for the suggestions.

Calculating I2 from a Bayesian Model Network metanalysis

0 Answers0