[R-meta] High heterogeneity and publication bias in mean difference meta-analysis

[Guido Schwarzer]

Dear Martin,

First some comments on posts to r-sig-meta-analysis. Please only send 
emails to the address used in my reply (see 
https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis). Furthermore, 
as Gerta, Matthias and Wolfgang are all subscribers of this list, there 
is no need to send the email twice to them.

As I do not know the specifics of your data I can only make some general 
comments.

You have considerable between-study heterogeneity with I2 values above 
90%. The first step is to check whether your study data is correct. For 
example, taking a standard error as a standard deviation can result in a 
very narrow confidence interval in a study leading to excessive 
between-study heterogeneity.

The funnel plot asymmetry in the third group is due to smaller / 
imprecise studies having different results than larger / more precise 
studies. This could either be due to clinical heterogeneity or bias. You 
could conduct meta-regression if you have data for potential effect 
modifiers. A contour-enhanced funnel plot could be used to evaluate 
whether the funnel plot asymmetry is due to publication bias, i.e., 
small studies are only published if they are statistically significant.

There are several other possibilities to evaluate small study effects 
using R (see sections "Investigating small study bias" and "Unobserved 
studies" on https://cran.r-project.org/web/views/MetaAnalysis.html).

Best,

Guido

-- 
Dr. Guido Schwarzer
Institute of Medical Biometry and Statistics,
Faculty of Medicine and Medical Center - University of Freiburg

Postal address: Zinkmattenstr. 6a, D-79108 Freiburg

Mail: sc using imbi.uni-freiburg.de
Homepage: https://www.imbi.uni-freiburg.de/

ORCID iD: https://orcid.org/0000-0001-6214-9087
R-book: https://www.springer.com/gp/book/9783319214153