[R-meta] Sample sized weighted studies for metamean function

[Massimo Baudo]

Thank you all for the precious feedback.

Best,

Massimo

> Il giorno 17 ott 2024, alle ore 07:02, Dr. Gerta Rücker via R-sig-meta-analysis <r-sig-meta-analysis using r-project.org> ha scritto:
> 
> Hi Massimo and others,
> 
> In addition to what Guido and Wolfgang already said (and I agree with them), I have a question: Why is it a problem if larger studies are downweighted because of larger variances? Note that it is a characteristic of the sampling variance that it does not only reflect the size of the study, but also its precision. In a large study with widely varying times the mean time is estimated with less precision. The inverse variance method penalizes this by downweighting - which is correct.
> 
> Another thought: If I had to compare times between two groups, I would not use the mean difference (MD), but the ratio of means (ROM). The analogon for one-group meta-analysis would mean taking log times and their variances.
> 
> Best,
> Gerta
> 
> 
> UNIVERSITÄTSKLINIKUM FREIBURG
> Institute for Medical Biometry and Statistics
> 
> Dr. Gerta Rücker
> Guest Scientist
> 
> Stefan-Meier-Straße 26 · 79104 Freiburg
> gerta.ruecker using uniklinik-freiburg.de
> 
> https://www.uniklinik-freiburg.de/imbi-en/employees.html?imbiuser=ruecker
> 
> 
> -----Ursprüngliche Nachricht-----
> Von: R-sig-meta-analysis <r-sig-meta-analysis-bounces using r-project.org> Im Auftrag von Viechtbauer, Wolfgang (NP) via R-sig-meta-analysis
> Gesendet: Donnerstag, 17. Oktober 2024 12:38
> An: R Special Interest Group for Meta-Analysis <r-sig-meta-analysis using r-project.org>
> Cc: Viechtbauer, Wolfgang (NP) <wolfgang.viechtbauer using maastrichtuniversity.nl>
> Betreff: Re: [R-meta] Sample sized weighted studies for metamean function
> 
> Hi all,
> 
> I haven't checked what specific methods are implemented in metamean() for converting five-number summary statistics to means/SDs, but there is a function for this as well in metafor:
> 
> https://wviechtb.github.io/metafor/reference/conv.fivenum.html
> 
> And Guido's 'P.S.' is very important, as it sounds like you may be feeding incorrect input to the rma() function.
> 
> Best,
> Wolfgang
> 
>> -----Original Message-----
>> From: R-sig-meta-analysis <r-sig-meta-analysis-bounces using r-project.org> On Behalf
>> Of Dr. Guido Schwarzer via R-sig-meta-analysis
>> Sent: Thursday, October 17, 2024 12:26
>> To: R Special Interest Group for Meta-Analysis <r-sig-meta-analysis using r-
>> project.org>
>> Cc: Dr. Guido Schwarzer <guido.schwarzer using uniklinik-freiburg.de>
>> Subject: Re: [R-meta] Sample sized weighted studies for metamean function
>> 
>> Massimo,
>> 
>> R package meta does not have an argument like 'weights' to provide user-defined
>> weights.
>> 
>> You could use metamean() from R package meta to approximate the standard error
>> from quantiles and use the calculated values as input to rma() from R package
>> metafor. List elements 'TE', 'seTE', and 'n' from a metamean object contain the
>> necessary input to rma().
>> 
>> Best,
>> Guido
>> 
>> P.S. I notice in your rma command the assignment 'vi = MVR_data$time_SD' which
>> is probably wrong. Argument 'vi' expects a variance (you could use argument
>> 'sei' to provide a standard error). Furthermore, time_SD doesn't sound like a
>> standard error which you could provide in argument 'sei'. It is possible to
>> provide the information for studies with single means using arguments 'mi',
>> 'sdi' and 'ni'; see the help page of escalc().
>> 
>> -----Ursprüngliche Nachricht-----
>> Von: R-sig-meta-analysis <r-sig-meta-analysis-bounces using r-project.org <mailto:r-
>> sig-meta-analysis-bounces using r-project.org>> im Auftrag von R-meta <r-sig-meta-
>> analysis using r-project.org <mailto:r-sig-meta-analysis using r-project.org>>
>> Antworten an: R-meta <r-sig-meta-analysis using r-project.org <mailto:r-sig-meta-
>> analysis using r-project.org>>
>> Datum: Mittwoch, 16. Oktober 2024 um 19:22
>> An: R-meta <r-sig-meta-analysis using r-project.org <mailto:r-sig-meta-analysis using r-
>> project.org>>
>> Cc: Massimo Baudo <massimo.baudo using icloud.com <mailto:massimo.baudo using icloud.com>>
>> Betreff: [R-meta] Sample sized weighted studies for metamean function
>> 
>> Dear Community,
>> 
>> I have the following question regarding the metamean function in meta, and
>> couldn’t find any previous threads on this regard.
>> 
>> I would like to calculate the pooled estimated mean of a follow-up time
>> parameter: time from the first surgery to the second surgery. Studies report
>> either mean (SD), or median (q1-q3). From my understanding (let me know if I am
>> wrong), only the inverse of the variance is used to estimate the studies in
>> metamean. This would be incorrect from my point of view to be used in this
>> specific case, as the assumption is that larger studies have lower variance,
>> thus weighted higher. However, for “follow-up” parameters larger studies could
>> still have a large variance as some patients may have the whole follow-up
>> length, while others would have a very short one. In this case, these larger
>> studies would be weighted less (as if they were smaller studies), while smaller
>> studies, which may have a smaller variance, would be weighted very high.
>> Therefore, I would prefer to weight this parameter by the sample size. I could
>> do it with metafor as follows:
>> 
>> MVR_data$weights = MVR_data$Samplesize / sum(MVR_data$Samplesize)
>> meta_analysis <- rma(yi = MVR_data$time_mean, vi = MVR_data$time_SD, weights =
>> MVR_data$weights, data = MVR_data, method = “DL")
>> summary(meta_analysis)
>> 
>> This way though, I “loose” the very useful function of mean/SD approximation of
>> the metamean function when dealing with median/q1/q3 or min/max.
>> Is there a possible solution to my problem with metamean?
>> 
>> Thank you for your support,
>> 
>> Massimo
> _______________________________________________
> R-sig-meta-analysis mailing list @ R-sig-meta-analysis using r-project.org
> To manage your subscription to this mailing list, go to:
> https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis
> _______________________________________________
> R-sig-meta-analysis mailing list @ R-sig-meta-analysis using r-project.org
> To manage your subscription to this mailing list, go to:
> https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis