I suggest you perfrom sensitivity analysis at first to determine whether certain studies cause the heterogeneity and when excluded the pooled analysis is homogenous. If this is the case, you can exclude them and highlight their methodological differences in the discussion. Regrading the regression; given that the moderator has a categorical binary nature (children/adult), it would be more appropriate to use the subgroup analysis - i.e subdivide your forest plot into two groups using "adult" and "children" as a covariate factor. When it comes to meta-regression, it is better -but not essential- to use continuous data like Age "in years". For instance, the interpretation in your example (if age was in years) would be that: for each increase in age by 1 year, the prevalence rates increase by 0.5. However, age can only explain 7% of the difference in prevalence rates between groups, making it a weak predictor. Accordingly, you will need to employ multiple regression models using different moderators like geography/race/BMI..etc till you reach a significant R-squared that can explain and predict the changes in prevalence rates. Hope this makes sense to you. If you need any further help, don't hesitate to tell.

Mohamed Abdel-Maboud - That is really helpful, thank you for taking the time to answer. I started with the sensitivity analysis, and no study appears to be having undue influence on the heterogeneity. All of my moderators are categorical, meaning I could simple conduct subgroup analysis rather than a meta-regression - is that correct? So, if then, do I simply look at the heterogeneity for each? For example, I have the following results from a subgroup analysis based on age:

Children - pooled prevalence=40%, I2=99%

Adults- pooled prevalence=52%, I2=98%

So although there is a difference in the pooled prevalence rates, there is not much difference in heterogeneity between the two groups. Do I conclude that age of participants does not impact on the heterogeneity between the studies?

How would you suggest writing this up in the report?

Thank you again for your help!

Grace Trundle - In that case, I would suggest you employ the random effect model in your analysis and create a separate forest plot for adults and a separate forest plot for children. Further, you can subgroup each one of them based on a certain factor like geography or residency, for instance. Regarding the interpretation, the concept of heterogeneity is totally subjective. We can think of it like this: when I combine 4 studies of oranges and 1 study of apples, the pooled analysis is heterogeneous. Accordingly, I exclude that study, and the analysis becomes homogenous. What does that mean? Well, it does not mean that "apples" affected the heterogeneity. It means that this "study" affected the heterogeneity because it included different subjects (which is "apples"). In contrast, when I exclude that study and the heterogeneity still significant, what does that mean? Well, it means that I have deeper variations within studies that I need to account for -using different moderators in subgroup analysis and meta-regressions (if the data permit). However, if the heterogeneity (either statistical or methodological) still there, then I have to highlight in my interpretations that the analysis of the current data does not give a conclusive answer. And so, further primary studies are needed with explicit criteria and methodology.

Mohamed Abdel-Maboud - That example was extremely helpful. Thank you very much for all your help! Hopefully others will also find answers in this thread :)