Dear Dominic,
1. If you fit trial as a fixed effect ( as in a conventional meta-analysis), then the main effect difference between populations is automatically adjusted for.
2. The further issue is one of population by treatment interaction. This is a difficult point and it hard to see how judgement can be avoided. For example, from one point of view every meta-analysis is carried out in population X but the results are used in population Y. This is because we studied patients yesterday (X) in order to decide how to treat a different set of patients tomorrow (Y).
See
1. Senn, SJ. The many modes of meta, Drug Information Journal 2000; 34: 535-549.
2. Senn, SJ. Added Values: Controversies concerning randomization and additivity in clinical trials, Statistics in Medicine 2004; 23: 3729-3753.
3. Senn, S. Hans van Houwelingen and the Art of Summing up, Biometrical Journal 2010; 52: 1-10.
and also
4. Yates, F, Cochran, WG. The analysis of groups of experiments, Journal of Agricultural Science 1938; 28: 556-580.
for an early discussion!
Regards
Stephen
-----Original Message-----
From: Evidence based health (EBH) [mailto:[log in to unmask]] On Behalf Of Dominic Hurst
Sent: 13 January 2011 15:32
To: [log in to unmask]
Subject: Combining results for meta-analysis
Hi, I'd appreciate some help on the following:
A systematic review looks to compare the effectiveness of two interventions, A and B, in a particular population, X.
The interventions, though, are commonly used in a discrete population Y also.
Some of the studies retrieved compare A and B just in the desired population X, but others compare the interventions in a mix of populations X and Y.
In the latter there may not have been block randomisation so the proportions of X and Y receiving A or B may be unbalanced.
In doing a meta-analysis of these studies, should one be cautious in looking to combine the results from the X-only studies with those extracted from the X-Y mixed studies? Does it matter that in removing the subgroup X from the mixed study the original randomisation has been disrupted and does it matter that the A and B intervention groups may be then be unbalanced?
Would it be reasonable to test for the significance of this with sensitivity analysis by removing the results from the mixed studies after the meta-analysis?
Thanks,
Dominic
|