The usual metan command is used to compare intervention A to B and to
pool several trials together. The same exercise is done to compare
intervention B to intervention C. So the code for a random effects
direct comparison would be:
* meta-analysis of trials of AvB (dummy= 1 if trials are A vs B)
metan event_t noevent_t event_c noevent_c if dummy==1, or random nograph
* This creates a point estimate and associated standard error for the
difference between A and B, which can be stored:
local logor1=log($S_1)
local se1=$S_2
* Similarly, meta-analysis of trials of AvC (dummy=0 if trials are A vs C)
metan event_t noevent_t event_c noevent_c if dummy==0, or random nograph
local logor2=log($S_1)
local se2=$S_2
But I am interested in comparing trials B vs C
So according to the authors above I write for computation of log OR,
OR and se for indirect comparison as follows:
local logor_bc_ic=`logor1'-`logor2'
local or_bc_ic=exp(`logor_aic')
local se_bc_ic=sqrt(`se1'^2+`se2'^2)
all this works fine but my question is twofold:
1) how do I interpret the results of this indirect comparisons? Assume
I have logor_bc_ic=logor1-logor2=0.8, how would I read that?
2) assume I want to do this same exercise but instead of using OR, I
want to use other efficacy measures such as relative risks or risk
differences. What I have done so far is to change the code and instead
of writing or, I would write rr or rd above and when it comes to the
indirect comparisons I do not convert the rd results into logs. This
gives me a result but I wonder why in the case of or and rd the result
of the indirect comparison is the same as actually subtracting the
(pooled) results from AvB minus AvsC, while in the case of rr,
something completely differently and (apparently) unrelated to the
numbers of the direct comparison appear on the screen.
Thank you very much for your help.
Maria A.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
