Interpretation of subgroup meta-analysis:
Only one subgroup is detected hetegogeneity of i = 70% but in the other i = 0%, and the studies are clinical homogeneous. the best analysis would be to assign this heterogeneity to a false-positive and is possible use the fixed model instead of the random model ?
Overall result (but not subgroup) of the analysis is similar, independently of the model (fixed or random). the test for difference between the subgroups did not show difference.
The purpose of a subgroup meta-analysis is to determine if the moderator variable (i.e., a dichotomous one in your case) can explain any heterogeneity in all of the studies. You did not state what that test's p value is. If it is not significant, I do not care about the heterogeneity within the subgroups. If that p value is significant, then I will inspect the heterogeneity of the subgroups. In your example, the I2 of 70% for one subgroup suggests to me that there may be at least one more moderator variable accounting for some heterogeneity.
I do not understand your comments about a false positive, studies being clinically homogeneous when your results suggest otherwise, and your desire to use a fixed-effect model to calculate an overall effect size. For the meta-analysis data sets I have seen over the last 10 years, rarely have I seen one that is appropriate to use a fixed-effect model for calculating an overall effect size.
I did the subgroup analysis because I wanted to compare if the effect of the treatment would differ between the subgroups (it was my research question). Results showed: subgroup 1 (chi²=6.76 df=2 (p=0.03) i²=70%) test of overall effect, subgroup 2 (chi²=0.08 df=2 (p=0.96) i²=0%).
In total, the overall treatment effect had non-significant heterogeneity (chi² = 7.43 df = 5 (P = 0.19) i = 33%) and test of overall effect was significant p = 0.0006. there was no difference between the groups (p =0.44).The test of overall effect is similar if I use fixed or random model.
The question is: should I present the graph in the random model only because heterogeneity appeared in a subgroup?
Best regards,
You can perform meta-regression, and code the subgroup study as categorical variable, and see how change the overall effect in the different subgroup studies.
Heterogeneity is the measure of inconsistency. In your primary analysis you did not find any, but you found it in the subgroup. The absence of heterogeneity means that more or less all included studies showed similar effects. Therefore, a subgroup analysis is not required. If I understood correctly, the subgroup analysis for the given moderator was prespesified in your protocol and you keep on that without modifications
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis