Use treatment effect as input for meta-analysis, not 2 groups

#1
Hey, I'm new to R and currently trying to teach myself how to use it with various guides.

Problem is, none of the guides cover the scenario I'm dealing with, they all assume that for a meta-analysis you have outcome data for both the control and treatment groups.

I would like to use the treatment effect (i.e. the difference between the two group's outcomes) and its standard error as input. I can' figure out which function to use and what arguments to use though.

Any help would be much appreciated.
 
Last edited:

hlsmith

Not a robit
#2
So you have the delta values and SEs for a bunch of studies? What package are you using.


I wound if it is possible to just do the typical MA, with treatment effects per group, then convert to differences. I have not done this before, but to me it seems like you should be able to input the traditional values and then get to the differences. I would be surprised if there isn't a R package for using differences. Also you may want to check out what Cochrane Collaborative Group has for relate software packages for this.
 
#3
So far I've been trying to use meta and metafor.

The thing is, I'm trying to analyse data that was not the primary endpoint in the respective studies (because very vew studies had this as their actual endpoint). So the reporting of the data was incomplete in many instances, and I had to calculate the SE for the differences using correlation coefficients, I did this in MS-Excel. I do in fact have endpoint values for each group, but to those I still need to apply a calculation involving a correlation coefficient to arrive at the final SE for the delta values. It would be easiest if I could just plug the final result from the Excel calculations into R, instead of trying to replicate in R parts of what I already did.

I'll check with the Cochrane Collaborative Group what they offer.
Sorry if I'm not using correct statistical terminology in some cases, I'm a grad student in a medical field and they didn't teach a lot of statistics nor how to use the relevant software. I'm having to learn all of this on my own.
 
#4
Really the details aren't important though.

All I need is for someone to point me to a package or function that accepts delta values (mean effect size of the experimental group minus mean effect size of the control group) and their standard errors (SEM) as input.
From there on I can presumably figure the rest out on my own.
 
#5
In case anyone is interested, I asked about this on the R mailing list and this can be done by using for example:

metagen() in R package meta in the following way:
metagen(delta, SEM)
or
rma() in R package metafor:
rma(yi = delta, sei = SEM)
 

hlsmith

Not a robit
#6
Thanks for the follow-up. I was thinking there should be something out there, especially given the recent interest in network meta-analyses.