[R] using metafor for meta-analysis of before-after studies (escalc, SMCC)

[Viechtbauer Wolfgang (STAT)]

The mean percentage change and the raw mean change are not directly comparable, even after standardization based on the SD of the percentage change or raw change values. So, I would not mix those in the same analysis.

Best,
Wolfgang

> -----Original Message-----
> From: Qiang Yue [mailto:qiangmoon at gmail.com]
> Sent: Wednesday, May 22, 2013 20:38
> To: Viechtbauer Wolfgang (STAT); r-help
> Subject: Re: RE: [R] using metafor for meta-analysis of before-after
> studies (escalc, SMCC)
> 
> Dear Dr. Viechtbauer:
> 
> Thank you very much for sparing your precious time to answer my question.
> I still want to make sure for the third question below:  for studies which
> only reported percentage changes (something like: the metabolite
> concentration increased by 20%+/-5% after intervention), we can not use
> the percentage change to calculate SMCC, but have to get the raw change
> first?
> 
> With best wishes.
> ________________________________________
> Qiang Yue
> 
> From: Viechtbauer Wolfgang (STAT)
> Date: 2013-05-21 10:09
> To: Moon Qiang; r-help
> Subject: RE: [R] using metafor for meta-analysis of before-after studies
> (escalc, SMCC)
> Please see my answers below.
> 
> Best,
> Wolfgang
> 
> > -----Original Message-----
> > From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org]
> > On Behalf Of Moon Qiang
> > Sent: Thursday, May 16, 2013 19:12
> > To: r-help
> > Subject: [R] using metafor for meta-analysis of before-after studies
> > (escalc, SMCC)
> >
> > Hello.
> >
> > I am trying to perform meta-analysis on some before-after studies. These
> > studies are designed to clarify if there is any significant metabolic
> > change before and after an intervention. There is only one group in thes
> e
> > studies, i.e., no control group. I followed the e-mail communication of
> > R-help (https://stat.ethz.ch/pipermail/r-help/2012-
> April/308946.html ) and
> > the Metafor Manual (version 1.8-0, released 2013-04-
> 11, relevant contents
> > can be found on pages of 59-61 under 'Outcome Measures for Individual
> > Groups '). I made a trial analysis and attached the output here, I wonde
> r
> > if anyone can look through it and give me some comments.
> >  I have three questions about the analysis:
> >
> > 1) Most studies reported the before-and-after raw change as Mean+/-
> SD, but
> > few of them have reported the values of before-intervention (mean_r and
> > sd_r) and the values of after-
> intervention (mean_s and sd_s), and none of
> > them reported the r value (correlation for the before- and after-
> > intervention measurements). Based on the guideline of the Metafor manual
> ,
> > I
> > set the raw mean change as m1i (i.e., raw mean change=mean_s=m1i), and s
> et
> > the standard deviation of raw change as sd1i (i.e., the standard deviati
> on
> > of raw change =sd_s=sd1i), and set all other arguments including m2i,
> > sd2i,
> > ri as 0, and then calculated the standardized mean change using change
> > score (SMCC). I am not sure if all these settings are correct.
> 
> This is correct. The escalc() function still will compute (m1i-
> m2i)/sqrt(sd1i^2 + sd2i^2 -
>  2*ri*sd1i*sd2i), but since m2i=sd2i=ri=0, this is equivalent to mean_chan
> ge / SD_change, which is what you want.
> 
> Make sure that mean_s is NOT the standard error (SE) of the change scores,
>  but really the SD.
> 
> > 2) A few studies have specified individual values of m1i, m2i, sd1i, sd2
> i
> > ,
> > but did not report the change score or its sd. So can I set r=0 and use
> > these values to calculate SMCC? Since SMCC is not calculated in the same
> > way like 1), will this be a problem?
> 
> Yes, this will be a problem, since you now really assume that r=0, which i
> s not correct. Maybe you can back-
> calculate r from other information (e.g., the p or t value from a t-test -
> - see https://stat.ethz.ch/pipermail/r-help/2012-
> April/308946.html). Or you could try to get r from the authors (then you c
> ould also just directly ask for the change score mean and SD). If that is
> not successful, you will have to impute some kind of reasonable value for
> r and do a sensitivity analysis in the end.
> 
> > 3) some studies reported the percentage mean changes instead of raw mean
> > change (percentage change=(value of after-intervention - value of before
> > intervention) / value of before intervention), I think it may not be the
> > right way to simply substitute the raw mean change with the percentage
> > mean
> > changes. Is there any method to deal with this problem?
> 
> Don't know anything off the top of my head.
> 
> > Any comments are welcome.
> >
> > With best regards.
> >   ------------------------------
> >  Qiang Yue