[R-meta] Meta-analysis of proportion differences (certain cells frequency)

[Viechtbauer, Wolfgang (NP)]

But what exactly does an author of a study that reports cell counts actually report? Are they are reporting the lymphocyte and total white blood cell count for each participant?

Best,
Wolfgang

> -----Original Message-----
> From: R-sig-meta-analysis <r-sig-meta-analysis-bounces using r-project.org> On Behalf
> Of Jakub Ruszkowski via R-sig-meta-analysis
> Sent: Friday, March 15, 2024 18:28
> To: R Special Interest Group for Meta-Analysis <r-sig-meta-analysis using r-
> project.org>
> Cc: Jakub Ruszkowski <jakub.ruszkowski using gumed.edu.pl>
> Subject: Re: [R-meta] Meta-analysis of proportion differences (certain cells
> frequency)
>
> Dear Wolfgang,
>
> thank you for your answer! Yes, I am aware of the compositional character of
> the data (I wish all authors of primary studies were also) and the huge
> limitations of any attempts to meta-analyze them. Unfortunately, I do not know
> any well-explained method to meta-analyze simultaneously all components of the
> composition properly, that is why I thought about simplification the issue to
> the analysis of differences of main cell types of interest.
>
> Yeah, the authors usually report the mean and SD of the proportions. I forgot
> that sample means even from beta (/Dirichlet) distributions follow the normal
> distribution! Thank you a lot for clarifying that it is ok to use methods for
> mean differences. In case some studies would report cell counts, would you
> rather analyze them together with studies reporting only mean+SD [%] (using
> SMD) or treat them separately?
>
> Best wishes
> Jakub
>
> W dniu 2024-03-15 14:44, Viechtbauer, Wolfgang (NP) napisał(a):
>
> > Dear Jakub,
> >
> > Proportions like you are describing can be thought of as so-called
> 'compositional data' (i.e., data that describe to what extent some whole is
> composed of various subcomponents):
> >
> > https://en.wikipedia.org/wiki/Compositional_data [2]For example, one might
> know that in a given person, 52% of their white blood cells are eutrophils, 36%
> are lymphocytes, 7% are monocytes, and the remaining 5% are other types. But
> without an actual count, these cannot be treated as binomial/multinomial counts
> and are just percentages (or proportions) of the whole.
> >
> > Compositional data analysis is its own subfield in statistics, but whether the
> methods described there are relevant in the present context is not clear to me.
> >
> > Since you mentioned the beta distribution: Yes, one could assume that a
> percentage/proportion like in the case above (i.e., a proportion of 0.36 of the
> white blood cells are lymphocytes) is beta distributed. But note that this is a
> proportion for a single individual. I would assume that there is such a
> proportion for multiple individuals within a group (e.g., patients). Then what
> is it that study authors would report? I would assume that they report the mean
> proportion (with hopefully also the SD of the individual proportions). If so,
> then one could basically just use methods for meta-analyzing means and mean
> differences.
> >
> > Best,
> > Wolfgang