As Babak Saravi and Shuaibu Suleiman have already pointed pout: Usually Q and I2 are used when assessing the heterogeneity. However, one has to be cautious when intepreting them. I'll address the problems below.

The problem with Q is that it depends on the number of studies so it's hard to interpret it without having additional information. Researchers often report a p-value for the Q statistic. A small p-value could mean that there is heterogeneity. However, it is well known that p-values depend on the sample size (in this case number of studies). Thus, when analyzing many studies one can get a small p-value even if there is only marginal heterogeneity. In contrast, when only a few studies are available one could get a large p-value even if there is a considerable amount of heterogeneity.

The problem with I2 is that it only indicates the RELATIVE amount of true heterogeneity (as compared to the error variance). Babak Saravi cited the threshold of 75%, which is often interpreted as high heterogeneity. However, it is not recommended to evaluate the heterogeneity solely using the I2 value. Why? I'll illustrate it with a short example.

• Consider a meta-analysis where the true heterogeneity (tau) equals .01 and the error heterogeneity equals .005. In this case I2 = 80%. [.012/(.012+.0052)]
• Now consider a meta-analysis with tau = .10 and error = .05. We get the same I value I2 = 80%. [.102/(.102+.052)]
• Can we really say that we have a high amount of true heterogeneity in both scenarios? No. We have a high RELATIVE amount (relative to the error heterogeneity). In order to assess the absolute amount of heterogeneity it's good to interpret the tau value (true heterogeneity) on it's own.
• You're estimating the prevalence in your meta-analysis, so a tau value of .01 would mean that it's normal that the prevalence estimates vary by 1% between studies or within studies. In many cases it's not a big deal. A prevalence of 74%, 75% or 76% is huge and a difference of 1% doesn't seem to be noteworthy so many people would argue that there is no considerable heterogeneity (it depends on the context!). However, consider the 2nd scenario. If the prevalence estimates vary by 10% (e.g., 75%, 65%, 85%) then most people would agree that there is a lot of heterogeneity.
• Some statistical packages report the true heterogeneity as variance (tau2). In this case it is recommended to take the square root in order to interpret the heterogeneity on the scale of the chosen effect size (e.g., prevalence).
• One of the researchers that popularized I2 (Julian Higgins) has coauthored an article dealing with the wrong interpretation of I2: DOI: 10.1002/jrsm.1230

Summing up, please do not rely solely on Q and I2 when assessing heterogeneity. It can lead to wrong conclusions. Consider interpreting the value of true heterogeneity (tau) too. In addition, you can assess the range of the effect sizes (e.g., "the prevalence estimates vary from 24% to 37% between the studies").

Babak Saravi, thank you for the information :) The studies you suggested were helpful, and we were able to confirm what methods we will use in our research.

Cochran’s Q statistic with df and R2 analog

1.. ((PDF) Random effects meta-analysis of COVID-19/S. aureus partnership in co-infection

https://www.researchgate.net/publication/345260193_Random_effects_meta-analysis_of_COVID-19S_aureus_partnership_in_co-infection)

Very good Mr.Babak Saravi