[R-meta] OR using metafor: Standard group with many zeros

[Nelson Ndegwa]

Dear Wolfgang,

Thank you!

nelson

On Sat, 18 Apr 2020 at 18:36, Viechtbauer, Wolfgang (SP) <
wolfgang.viechtbauer using maastrichtuniversity.nl> wrote:

> Dear Nelson,
>
> 1) You mean that there are studies where there were 0 patients (not just 0
> events, but actually no participants) in one of the two groups? If so, I
> don't see how such studies could be included in a meta-analysis that aims
> to estimate ORs. Those are essentially one-arm studies, so they don't
> provide any information about group differences.
>
> 2) help(rma.glmm) explains what nAGQ represents. When using glmer() for
> fitting the 'UM.RS' model, one has to use the Laplacian approximation
> (i.e., nAGQ=1). You could try using:
>
> control=list(package="GLMMadaptive")
>
> which then uses the GLMMadaptive package, which does allow for using
> proper quadrature when fitting the model.
>
> As for the convergence warning - that comes from glmer(). It is quite
> quick to flag potential non-convergence, but this may be a false positive.
> Again, you could compare results when using GLMMadaptive.
>
> Best,
> Wolfgang
>
> >-----Original Message-----
> >From: R-sig-meta-analysis [mailto:
> r-sig-meta-analysis-bounces using r-project.org]
> >On Behalf Of Nelson Ndegwa
> >Sent: Saturday, 18 April, 2020 17:09
> >To: r-sig-meta-analysis using r-project.org
> >Cc: wvb using wvbauer.com
> >Subject: [R-meta] OR using metafor: Standard group with many zeros
> >
> >Hi,
> >
> >I am interested in calculating an OR for a developing some outcome after
> >new-treatment compared to standard treatment. I have proportions data,
> with
> >19 studies.
> >
> >I ran a Random-Effects Model (code below, data attached) with the model
> >type: Unconditional Model with Random Study Effects as follows, then
> >exponentiated the coefficient for an OR:
> >
> >Two questions:
> >
> >1. I noticed there were quite a few zeros in the standard group (about
> 8/19
> >studies) . This is sometimes because there were no standard patients in
> that
> >study. 0 patients and 0 events. Should it be a zero or a . for missing
> >instead? what is the best way to handle this situation? I tested running
> >above code by using both TRUE or FALSE in the drop00 option, but the
> results
> >didn't change.
> >2. I got results but also this message which am not sure what it means for
> >the results I have got, are they still valid? "Currently not possible to
> fit
> >RE/ME model='UM.RS' with nAGQ > 1. nAGQ automatically set to 1.Model
> failed
> >to converge with max|grad| = 0.00207718 (tol = 0.001, component 1)"
> >I shall greatly appreciate your comments/advice.
> >
> >=========
> >CODE RUN
> >=========
> >res1 <- rma.glmm(measure="OR", ai=treatment_events, n1i=treatment_total,
> >ci=control_events, n2i=control_total, data=repdat, model="UM.RS")
> >
> >print(res1, digits=3)
> >
> >predict(res1, transf=exp, digits=2)
>

	[[alternative HTML version deleted]]