R help - Mar 2009 - perform subgroup meta-analysis and create forest plot displaying subgroups

Hello, 
I'm using the rmeta package to perform a meta analysis using summary
statistics rather than raw data, and would like to analyze the effects in three
different subgroups of my data. Furthermore, I'd like to plot this on one
forest plot, with corresponding summary weighted averages of the effects
displayed beneath each subgroup.

I am able to generate the subgroup analyses by simply performing 3 separate
meta-analyses with the desired subset of data. However, I can't manage to
plot everything on the same forest plot.

My data and code appear as follows:
data(meta)
subgroup    study    beta    se
subgroupA    site1    -0.35    0.12
subgroupA    site2    -0.34    0.1
subgroupB    site1    -0.28    0.06
subgroupB    site2    -0.29    0.07
subgroupC    site1    0.34    0.03
subgroupC    site2    0.36    0.04

#meta-analysis
A<-meta[c(1,2),]
B<-meta[c(3,4),]
C<-meta[c(5,6),]

Ameta<-meta.summaries(A$beta,A$se,method="fixed",logscale=TRUE,names=A$study)
Bmeta<-meta.summaries(B$beta,A$se,method="fixed",logscale=TRUE,names=B$study)

Cmeta<-meta.summaries(C$beta,A$se,method="fixed",logscale=TRUE,names=C$study)


list(summary(Ameta),summary(Bmeta),summary(Cmeta))

I get my output, but I'm not able to plot the three subgroups (A,B,C) on the
same forest plot using any of the functions I see in rmeta. Any ideas?
Thanks,
Steve Lubitz

Cardiovascular Research Fellow, Brigham and Women's and Massachusetts
General Hospital



      
	[[alternative HTML version deleted]]

Steven Lubitz schrieb:
> Hello, I'm using the rmeta package to perform a meta analysis using
> summary statistics rather than raw data, and would like to analyze
> the effects in three different subgroups of my data. Furthermore, I'd
> like to plot this on one forest plot, with corresponding summary
> weighted averages of the effects displayed beneath each subgroup.
> 
> I am able to generate the subgroup analyses by simply performing 3
> separate meta-analyses with the desired subset of data. However, I
> can't manage to plot everything on the same forest plot.
Maybe I'm wrong but the 'forest'-function (package 'meta', 
<http://cran.at.r-project.org/web/packages/meta/meta.pdf>) should be 
able to do what you want. I guess you could be interested in the 'byvar'
argument.

Bernd

Bernd,
Thanks. I believe this will do what I need, but do you know how I can set up my
meta object so that the meta is performed on each subgroup individually? I can
get the overall meta of all 6 observations, or I can get a separate meta of each
subgroup using the subset command, but I can't get subgroup A, subgroup B,
and subgroup C all into the same object. In order to use the plot feature it
appears I need a byvar, so I think I'll need them all in the same object.

subgroup    study    beta    se
subgroupA    site1    -0.35    0.12
subgroupA    site2    -0.34   
 0.1
subgroupB    site1    -0.28    0.06
subgroupB    site2    -0.29    0.07
subgroupC    site1    0.34    0.03
subgroupC    site2    0.36    0.04
Generic inverse variance meta-analysismetagen(beta, se, studlab=study,
sm="OR")

Thanks,
Steve

--- On Tue, 3/10/09, Weiss, Bernd <bernd.weiss@uni-koeln.de> wrote:
From: Weiss, Bernd <bernd.weiss@uni-koeln.de>
Subject: Re: [R] perform subgroup meta-analysis and create forest plot
displaying subgroups
To: slubitz1@yahoo.com, r-help@r-project.org
Date: Tuesday, March 10, 2009, 2:31 AM

Steven Lubitz schrieb:
> Hello, I'm using the rmeta package to perform a meta analysis
 using
> summary statistics rather than raw data, and would like to analyze
> the effects in three different subgroups of my data. Furthermore, I'd
> like to plot this on one forest plot, with corresponding summary
> weighted averages of the effects displayed beneath each subgroup.
> 
> I am able to generate the subgroup analyses by simply performing
 3
> separate meta-analyses with the desired subset of data. However, I
> can't manage to plot everything on the same forest plot.
Maybe I'm wrong but the 'forest'-function (package 'meta',
<http://cran.at.r-project.org/web/packages/meta/meta.pdf>) should be able
to do what you want. I guess you could be interested in the 'byvar'
argument.

Bernd



      
	[[alternative HTML version deleted]]