Hi Timo thanks for your answer. do you know where I can look up prerequisites or rules that apply for meta-analysis?

Hi Glenn!

Thank you for your comprehensive answer. I want to ask my question to you more specifically. When one performes a metanalysis to find out if plasma cholesterol is associated with cardiovascular diseases this can be done easily, because measuring cholesterol is standardised in all labs around the world. But when we do Meta-analysis in trails evaluating plant cholesterol (phytosterols) with cardiovascular diseases then it is more tricky, because the measurement of phytosterols in different labs is different and in contrast to serum cholesterol is not standardised. It is well known that dependent on internal standards used and different techniques serum phytosterol levels differ from lab to lab up to 200-300 %. Given that there are these big differences I believe that a Meta-Analysis cannot be done. Do you agree? And if yes where can I find an appropriate citation ?

Dear Oliver

What I can recommend is highly dependent on the numbet of included studies

1- if the number of studies are high enough you can perform subgroup analyses or metaregression to explore the effect of different measurement methods

2- otherwise just stick to systematic review and drop meta-analysis.

A good source for you is the following book

http://eu.wiley.com/WileyCDA/WileyTitle/productCd-EHEP002313.html

Ramin

Thank you for your valuable suggestions !

Dear Oliver,

Have a look at the Cochrane web site http://www.cochrane.org/ and i recomand you also the Cochrane hand book that is a reference for systematic review and meta analysis.

You can also find a statement for meta analysis that is call Prisma http://www.prisma-statement.org/

Best.

Seb.

Dear Sebastien, thank you !

I have used as reference the book of Diane B. Petitti: Meta-analysis, decision analysis and cost-effective analysis(Oxford University Press).

My comments: you can "standardize" the procedures but I wonder about reliability/reprucibility across different labs participating in different researches: I do not remember of what should be required, i.e. the "techical error", a measure of variability between for example two samples of the same specimen. But this can be said of any measure, of outcome too.

But I feel that selection bias can be relevant, i.e. inclusion and exclusion criteria, and so on, all things that you cannot control for nor quantify.

For these reason I advise to be cautious about results.

Dear Vincento, thank you for your answer. I agree with you that a standardized measurement is a prerequisite for performing a meta-analysis. As far as I understand you you agree on this. I am looking for a citation that confirms this opinion. Do you know of one ?

We agree, Oliver, but I wonder about the possibility "in real world" to have different investigations in different countries that exchanged some samples of their subjects to explore variation among different labs. What we can pretend is only a full, in-depth description of the procedure.

But remember, at least for clinical research: this does not apply only for labs but also for definitions of diseased and non diseased or for any clinical outcome (except, hopefully, dead versus survived in prognosis studies): think, for example to the definition of dyspepsia and contrasting results of the studies on results of the studies.

About labs I had a paper of the American Academy of Laboratory Medicine, that unfortunately I borrowed to my colleagues and I cannot retrieve.

About the issue of definitions ("which is which") I think there is not a paper but chapters of books of Epidemiology or of many medical matters, as it is rather obvious.

In multicentric studies I suggest to look for Cohen k (inter.rater agreement beyond chance for categorical classifications, and Intraclass Correlation Coeffcient for continuous variables, commonly performed by professional softwares).

I hope this was useful to all of you.

The cochrane site has a handbook -- http://handbook.cochrane.org/ -- which is a good review.

Could you have a review article that was part meta-analysis examining standardized tests and part examining non-standardized tests?