Dear Thiago Ramos Grigio,

You can perform subgroup analysis on dosage, however with meta-analysis it should be conducted based on high heterogeneity to find if the statistical heterogeneity is due to different doses used.

Hi Thiago,

If both the 20mg and 40mg effect sizes are measured with respect to the same control group, and thus originate from the same study, I would not perform a subgroup analysis in a standard meta-analysis framework. I'll refer to Chapter 10 of the following book:

https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/multilevel-ma.html

"Statistical independence is one of the core assumptions when we pool effect sizes in a meta-analysis. If there is a dependency between effect sizes (i.e. effect sizes are correlated), this can artificially reduce heterogeneity and thus lead to false-positive results."

If the estimates for 20mg and 40mg come from the same study, the estimates are likely correlated, which would lead to a violation of the independence assumption.

If you are still interested in performing a subgroup analysis, you could incorporate it into a multilevel meta-analysis with medication dose (20 mg vs 40mg) as one of the explanatory variables. If you look at equation (10.8) in the book chapter I referred to earlier, you see that a multilevel meta-analysis incorporates a second random-effect: one to account for clustering of effect sizes within the same study. Statistically, this would be the correct way to pool your correlated effect sizes (if you insist to combine them).

Hope this helps!

Kind regards,

Michael

Thank you very much Michael! So clear!!!!