sdm-help-list > Heterogeneity test for peaks from meta-regression
Showing 1-3 of 3 posts
Display:
Results per page:
Jun 15, 2020  05:06 PM | Yen-Wen Chen
Heterogeneity test for peaks from meta-regression
Dear experts, 

Is is appropriate to do heterogeneity test (like I-square, Bias Test) for result peaks from meta-regression analysis? 

From our meta-analysis with structural MRI studies, we didn't find any peaks at uncorrected p < .005. (we only got 1 peak at right parahippocampal gyrus, at a liberal threshold p < .05, k > 10)

We would like to know if the mean age of the samples could contribute to the null result, and we did it with two approaches, 

(1) First, we included mean age as covariate in Mean analysis step to control for age: the result was similar to what we got when no covariate was controlled (although this time, we got two peaks (right parahippocampal and right putamen), at liberal threshold p < .05, k > 10)

(2) Next, we conducted a meta-regression to see if mean age of the samples yielded any effects: we got 1 peak at stringent threshold p < .0005 (k > 10). This peak was located at right putamen, very close (almost the same) to the putamen cluster from meta-analysis controlling for mean age (from our first step above)

Our rationale to use the above approaches is that the heterogeneity can contribute to positive result of meta-analysis (i.e., there is significant peak across studies), and we thought the heterogeneity could also contribute to the null meta-analysis results (please correct me, if the rationale doesn't make sense). 

From the previous meta-analysis studies I had read, I didn't see any study talked about testing heterogeneity for meta-regression results. 

I'm wondering if it makes sense to do heterogeneity tests for results from meta-regression? (in our case, the peak from mean age)

I'd appreciate any opinion or suggestions!

Thank you so much.

Sincerely, 
Samantha
Jun 16, 2020  01:06 PM | Lydia Fortea - Instituto de Investigaciones Biom├ędicas August Pi i Sunyer (IDIBAPS)
RE: Heterogeneity test for peaks from meta-regression
Dear Samantha, 

It is true that papers usually doesn't report the heterogeneity test in the meta-regressions, as it is not the "main analysis"; but it can be done and those tests can prove the strenght and interpretation of your analysis. To summarize is more a personal decision, but is always good to perform statistical analysis that support your hypotheses.

Kind Regards, 
Lydia
Jun 16, 2020  03:06 PM | Yen-Wen Chen
RE: Heterogeneity test for peaks from meta-regression
Hi Lydia, 

Thanks for the clarification and suggestion!

Best wishes, 
Sam.