My colleagues and I are currently undertaking an Umbrella review, for which we have extracted some SMDs from the included reviews but we are not totally sure about the most suitable statistical methods to generate an overall effect estimate.
I'd really appreciate your feedback.
Sincerely, Jose.
Hi,
So the methodology around umbrella reviews is still in its infancy and there are many unknowns.
The Cochrane handbook has advice on this for their overviews (another term for umbrella reviews) and suggests that Umbrella reviews are not designed to perform meta-analyses of pooled estimates from different reviews. The risk being that we dilute the effect estimate by doing so.
It may be possible to combine the individual studies (that are homogenous and relevant to the question being asked) from different reviews into one overall meta-analysis, but again the risk of bias in this approach is high as you are relying on reviews search and inclusion/exclusion etc. Some would suggest just doing a much bigger systematic review in this case.
This is something I have been trying to tackle with an overview I am performing and I think the option I will go for (to minimise risk of bias) is to just present the SMDs for similar/same outcomes side by side to allow for a comparison between reviews.
These, and other issues, need thoughtful consideration to help establish a sound methodology for this type of meta-review.
David.
in addition to Davids (excellent) contribution, have a look at 'summary of findings tables' as a way of presenting the data. The risks and confounders associated with attempting statistical combination of effects from across reviews is likely to negate any benefit the umbrella review would otherwise achieve!
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis