Hi Samia,
In my humble opinion, there
are two notions in this. First, and most important, the conceptual
notion. This should be determined by the researcher's discretion. I
think the questions that really count rather than just the design will
be: Do the studies measure the same clinical end point? Are the results
overall consistent in regards to the overall effect magnitude and
direction? Are patients characteristics somehow similar across the
studies?
The second notion is the statistical notion of heterogeneity. There is
no arbitrary line or threshold for heterogeneity that you may consider
for such a decision. Further, the presence or absence of heterogeneity
will then dictate which model to use (random versus mixed effects
model). The first assumes similar distribution among all the studies.
The latter assumes that each individual study distribution is part of a
bigger distribution for the combined effect and does the calculation
accordingly.
In an article that I think someone in the list distributed few
weeks ago (unfortunately can't remember who and when) the issue of
dealing with heterogeneity in reviews was discussed. No criticism was
directed to the presence or absence of heterogeneity itself. In fact,
the article criticized usage of an inappropriate model, and emphasized
the importance of addressing and quantifying heterogeneity (Q and I2
statistics); interpreting it if you can and use the proper model. I will
try to find the article and send it to you.
