[R-meta] metafor-mota-analysis with between group moderator

[Matilde Vaghi]

Thank you for very helpful feedback. 

Wolfang, I would appreciate if you could disentangle the following for me . 

My variable of interest is X. 

Meta1 I run a meta-analysis on X [ by using escalc first with SMD for which I obtain yimeta1 and v1meta1; and then rma] and I do see a difference between the two groups. 

meta2: However, I also have another measure which is AGE on which I run the meta-analysis and there is also a difference between the two groups. By using escalc, I obtained the SMD (standardized mean difference [yi]) which is also used for the meta-analysis [res<- rma(yimeta2, vimeta2, data=df)]

As you pointed out, I want to see if the difference between the two groups (with respect to variable X, meta1) might be a confound of the two groups having different age (as assessed with meta-analysis 2). I therefore used the Standardized mean difference obtained with meta2 as a moderator [=SMD age]. [I need to use the standardized mean difference as age is only an example and the variable of interest here (which is not age) is assessed with different scale for each study]. 

I do find a significant value of the intercept but not of SMD age on variable X. What does it mean exactly?  
rma(yimeta1, vimeta1, mods = cbind(yimeta2), data = df)

                estimate      se    zval    pval    ci.lb   ci.ub     
intrcpt      xxx  xxx  xxx2  xxx   xxx  xxx  ***
SMDage     xxx  xxxx xxx  xxx xxx  xxx

Hope my explanation is sufficiently clear, 
thank you very much and best wishes, 
Matilde
  

> On 1 December 2017 at 17:49, Viechtbauer Wolfgang (SP) <wolfgang.viechtbauer at maastrichtuniversity.nl> wrote:
> Indeed, this might be one approach.
> 
> But what is it that you want to examine, Matilde?
> 
> Whether the difference between the two groups (with respect to variable y) is associated with the average age of the total sample? Then computing the mean age (but maybe computing a weighted average of the mean age of the two groups in case the groups are not of equal size) and using that as a moderator is fine.
> 
> But it sounds to me like you are wondering if the difference between the two groups (with respect to variable y) might be a confound of the two groups having different mean ages. In that case, you could use the *difference* in the mean age of the two groups as a moderator.
> 
> Best,
> Wolfgang
> 
> -----Original Message-----
> From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On Behalf Of Michael Dewey
> Sent: Friday, 01 December, 2017 17:20
> To: Matilde Vaghi; r-sig-meta-analysis at r-project.org
> Subject: Re: [R-meta] metafor-mota-analysis with between group moderator
> 
> Dear Matilde
> 
> I think you may need to use the so-called bivariate approach
> 
> http://www.metafor-project.org/doku.php/analyses:vanhouwelingen2002
> 
> The example there is not identical to your problem but I think it is
> close enough. If Wolfgang does not post saying I am talking rubbish then
> try it and post again with any more queries.
> 
> Michael
> 
> On 01/12/2017 11:43, Matilde Vaghi wrote:
> > Dear all,
> >
> > I am using metafor to conduct a between group meta-analysis to test if
> > there is a difference between groups (controls vs. patients) on the
> > continuous independent variable y.
> >
> > I need to check the influence of relevant moderator, in particular I obtain
> > a between group difference on the variable y and I want to check if the
> > finding remains significant after account for the effect of age (if the
> > difference in the independent variable y is still difference once adjusted
> > for age).
> >
> > On the metafor manual, a clear example shows how to assess the effect of a
> > moderator (such as latitude) on the variable of interest. However, such
> > example refers to the presence of a study-level moderator (each study was
> > conducted at a different latitude, therefore one value is associated to
> > each study). In my case, each study have two values (i.e, mean age controls
> > and mean age patients). Therefore, it is not immediately clear how to adapt
> > the following
> >
> > Rma(yi, vi, mods=cbind(Latitude), data=df)
> >
> > for situations in which there is one moderator variable but two values
> > associated to it for each study. I tentatively computed a mean age per
> > study ((mean age controls for study 1+ mean age patients for study1)/2;
> > ((mean age controls for study 2+ mean age patients for study2)/2 etc) to
> > have one value per study but I am not sure this is the best way forward.
> >
> > Any advice would be highly appreciated. Many thanks and best wishes,
> >
> > Matilde


	[[alternative HTML version deleted]]