# Heterogeneity evaluation in meta-analysis

I ran meta-analysis to combine findings of 15 studies. I used Comprehensive meta analysis software to generate the results. The purpose of the analysis is to see if the results are consistent and can support hypothesis in biology regarding one of the metabolites in the blood.

I calculated Q and I2 statistics and I found high heterogeneity between the studies (I2=86%). I did funnel plot and I found outliers. Removing two outliers from funnel plot and repeating I2 statistics yielded that I2 dropped to 0%

Is this the optimal way to evaluate heterogeneity in meta-analysis? If not, what is the correct and more robust method?

• Please edit your question I2 as I - squared and add tag - heterogeneity. – Subhash C. Davar Dec 14 '17 at 7:24

If you delete the outliers without any scientific justification but just because they are inconvenient for your model you risk coming to a misleading conclusion. You are presumably doing your modelling in order to get a summary value so if deleting the outliers affects the summary you need to explain that. Concentrating purely on the heterogeneity may not be the best way forward. You need to examine the characteristics of the studies for an explanation. Describing the pattern of heterogeneity as well as reporting a summary statistic like $I^2$ is also called for. With this number of studies (15) the funnel plot may be less informative than you hope too.