Hi Khalid,

To add from what Abdelrhaman described.

I am not sure whether you are dealing with a continuous or a categorical outcome.   For studies with cluster RCTs and Cross over trials, you can account for design effect using DE= 1 + (m-1) ICC. Then you can adjust the SE using this formula SEadj = sqrt(DE)*SE. I had similar question few months ago, this paper is very helpful resource https://www.ncbi.nlm.nih.gov/pubmedhealth/PMH0077775/pdf/PubMedHealth_PMH0077775.pdf

Dear Dr. Abdelrahman and Dr. Mihiretu, I really appreciate your help. Thank you

Hi Khaled,

This is a commonly encountered dilemma when conducting a meta-analysis. There are a few ways of going about it, some of which have already been neatly outlined above. Just wanted to add that you can choose to analyse the studies as whole and then conduct subgroup analyses to see if the conclusions drawn differ - and the reasons for these differences if any. There is always the theoretical risk that the beneficial effects of the first treatment might carry over into the second treatment period and thereby confound the detection of treatment effects. You can also examine the overall heterogeneity of the studies using the Q statistic or I-squared index and hope to mitigate the high heterogeneity (often arbitrarily defined as I-squared >75%) by using a random-effects model for your meta-analysis.

Additionally, you can also examine the overall heterogeneity of the pooled studies using the Q statistic or I-squared index, and hope to mitigate the high heterogeneity (often arbitrarily defined as I-squared >75%) by using a random-effects model for your meta-analysis.

Some researchers have also proposed the use of a multilevel model and a Bayesian hierarchical model for combined trial design meta-analysis. I have attached a recently published paper on this topic for your reference.

Hope it helps. Good luck with your research!

Best wishes,

Qin Xiang 

http://onlinelibrary.wiley.com/doi/10.1002/jrsm.1236/abstract