Ok
many thanks!
ch
Selon Benjamin Kay <[log in to unmask]>:
> I'm sure a more senior person on the mailing list could give you a better
> answer. My impression is that 5-7 patients would be enough for a case study,
> i.e. "What does Network X look like in patients with Disease Y?" Such a low
> number of patients would probably be pushing it for a systematic analysis,
> especially if you don't have control subjects! But again, if the conditition
> you are studying has a dramatic effect on connectivity (i.e. very large
> effect size) then it might be enough.
>
> On Wednesday, August 15, 2012 10:17:46 you wrote:
> > Dear Benjamin,
> >
> > Thank you for this response.
> >
> > I must deal with some rare clinical cases (5 to 7 patients, for instance):
> is it
> > possible to use MELODIC group-analysis for this very small population? or
> is it
> > statistically irrelevant?
> >
> > Thank you.
> > Best,
> > Christophe
> >
> >
> >
> > Selon Benjamin Kay <[log in to unmask]>:
> >
> > > Christophe,
> > >
> > > That will depend entirely on your hypothesis and expected effect size.
> The
> > > final stage of analysis with MELODIC ICA + dual regression typically
> involves
> > > some sort of GLM (often a t-test) to compare connectivity between two or
> more
> > > experimental groups. The standard techniques for estimating effect size
> (e.g.
> > > Cohen's d) and power apply. Since your experimental protocol will
> influence
> > > the standard error, and thus the effect size, of your data, the best way
> to
> > > figure out how many subjects you need is to collect pilot data and
> estimate
> > > your effect size and power from that. As a very rough ballpark figure, I
> have
> > > seen ICA based publications with fewer than 20 total subjects.
> > >
> > > Hope this sets you on the right track,
> > > Benjamin
> > >
> > > On Wednesday, August 15, 2012 09:10:07 you wrote:
> > > > Dear experts,
> > > >
> > > > I would like to know the lower population size required to compute
> MELODIC
> > > ICA
> > > > group-analysis.
> > > >
> > > > Thank you.
> > > > Best,
> > > > Christophe
> > > >
> > >
> >
>
|