[R-meta] Network meta-analysis: including a within-person (split-leg) study + pooling absolute vs relative changeNetwork meta-analysis: including a within-person (split-leg) study + pooling absolute vs relative change

[Jorge Teixeira]

Thanks Wolfgang.

I did that to highlight that the means were not independent.

I also assumed that the data had to be in long format to estimate the
V matrix. Is that not the case? :)

On Wed, 28 Oct 2026 at 10:45, Viechtbauer, Wolfgang (NP)
<wolfgang.viechtbauer using maastrichtuniversity.nl> wrote:
>
> Hi Jorge,
>
> Much more readable!
>
> I see no inherent problem with including study IV, but not sure why you structured it as you did. In essence, it is like this:
>
>            Obs     Tx A    Tx B    Tx C
> Study I      1       100     150       -
> Study II     1       200       -     200
> Study III    1       200     200       -
> Study IV     1         -     170     180
>
> But since study IV is a within-subjects design, the two means are not independent. So you would have to compute the covariance between them and include this in the V matrix. The covariance between the two means is a function of the correlation of the raw measurements for the two legs. If this is not reported, then you could use a guestimate and do a sensitivity analysis.
>
> Best,
> Wolfgang
>
> > -----Original Message-----
> > From: R-sig-meta-analysis <r-sig-meta-analysis-bounces using r-project.org> On Behalf
> > Of Jorge Teixeira via R-sig-meta-analysis
> > Sent: Wednesday, October 28, 2026 10:57
> > To: Michael Dewey <lists using dewey.myzen.co.uk>; R meta <r-sig-meta-analysis using r-
> > project.org>
> > Cc: Jorge Teixeira <jorgemmtteixeira using gmail.com>
> > Subject: Re: [R-meta] Network meta-analysis: including a within-person (split-
> > leg) study + pooling absolute vs relative changeNetwork meta-analysis: including
> > a within-person (split-leg) study + pooling absolute vs relative change
> >
> > Thanks, Michael. And now?!
> >
> > I'm planning a small network meta-analysis and would value your
> > input on two design questions.
> >
> > SETUP
> >
> > I have three parallel-group RCTs comparing treatments A, B and C
> > (mean value at post per arm):
> >
> >               Tx A    Tx B    Tx C
> >   Study I      100     150       -
> >   Study II     200    200      -
> >   Study III      -     200     200
> >
> > I would like to add a fourth study (Study IV). It is a within-person
> > design: each participant's legs were randomized to different
> > treatments, so comparisons are within-group rather than
> > between-group. I have the post-treatment means + SD for each
> > condition.
> >
> > QUESTION
> >
> > Can Study IV be meta-analyzed together with the three parallel RCTs?
> >
> >             Obs     Tx A    Tx B    Tx C
> >   Study I      1       100     150       -
> >   Study II     1       200       -     200
> >   Study III    1       200     200       -
> >   Study IV     1         -       -     180
> >   Study IV     2         -     170       -
> >
> > Thanks a lot for your help.
> >
> > Best,
> > Jorge
> >
> > On Wed, 28 Oct 2026 at 09:12, Michael Dewey <lists using dewey.myzen.co.uk> wrote:
> > >
> > > Dear Jorge
> > >
> > > You have set your mailer to post in HTML which mean that we are not
> > > seeing what you posted. Not sure how you get plain text.
> > >
> > > I think the answer to question 2 is No. It is fairly easy to create
> > > examples with the same absolute change and different relative change.
> > >
> > > Michael