Adjustment for correlated data in meta-anlaysis

[Ainslie]

Hi,
I'm completing a systematic review and performing meta-anlaysis forest plots. The research question related to the finding of risk factors for a complication of diabetes. There are therefore many different risk factors and often there are only three papers with similar methods(observational, longitudinal) and effect measure e.g. odds ratio or risk ratio. There are several occasions where the population study, being longitudinal has reported findings several years apart, this raises the problem of similar study design for a similare risk factor, however there is correlated or clustered data- based on a simlar population sample.
How can I adjust the meta-analysis to accommodate correlated population data. Should the meta-analysis not be performed at all?

 

[hlsmith]

Hmm, good question. So is it the exact some study design, but for different time periods?


P.S., As you mentioned, meta-analyses based on observational data are difficult.

 

[hlsmith]

Well, I was just thinking about your question. I have not done anything analyses with the same issues. The MA process with control for heterogeneity between studies, though you have the exact opposite, though only one cluster, so most studies aren't in a cluster. MAYBE, this is just a maybe, an addition dummy term can be added to the model which is 1 if they are in that cluster and 0 if not. You may also be able to fiddle around with sensitivity analysis of leaving it out and including it to see how sensitive results are.


I would image you may be able to do some literature review and see how others address this concern in their MAs. Since this has to come up in MAs.


Let us know what you find out or do, since I would be interested to know!