Dear developers, 

We highly appreciate the further development of the SDM toolbox - we switched to SDM-PSI v6.21 in our recent VBM meta-analysis and included for the first time raw data maps in addition to reported findings. Briefly, we aim at determining robust GMV alterations for a mental disorder and include data from 21 studies and 2 raw maps from original studies reporting patient vs control comparisons, threshold for the meta-analysis is p < 0.0025, uncorrected. Pooling only the 23 studies without the raw maps yields decreased volume in the anterior insula. After including the two raw maps results however change drastically in terms of direction and location (increased volume in bilateral lingual gyrus).

Our questions:
(1) Is it possible that raw maps from only 2 studies have such a strong influence? Most importantly, one of the raw maps is from a large mega study with >170 subjects and the original study reports Null findings under the threshold of TFCE p < 0.05. Could this be the reason for the strong influence? Does SDM in general support inclusion of raw maps from Null findings studies?
(2) Is SDM-PSI already implements the jackknife sensitivity analysis in the main meta-analysis to generate the significant cluster? How should we testify that the identified clusters are robust across the studies (e.g., in AES-SDM there is a separate jackknife analysis in the software)?
Thanks in advance and best wishes,

Lucy

Hi Lucy, 

I'm not an expert but I've been using this software for a few months now. I believe the jackknife sensitivity analysis has to be completed manually (i.e. repeating the analysis leaving one study out each time) with the new SDM-PSI version you're using, as there is no inbuilt function for this as there was for the previous AES-SDM. If I remember correctly, there are other forum posts here discussing this, where the developers provide more in-depth responses/refer to literature where you can read more about conducting jackknife sensitivity analyses manually. 

Can I ask why you chose the sig threshold of p .025? I've been struggling with what stat threshold to use for my analysis, and so I'm curious to hear reasonings for particular threshold choices, e.g. why did you opt for the uncorrected over the tfce corrected? 

Best, 
Molly

Hi Molly,

    Thanks for the reply about the jackknife. 
    Regarding the threshold: actually I choose both tfce corrected and p <0.0025 uncorrected. The uncorrected p < 0.0025 was chosen following this study:

Chavanne, A. V., & Robinson, O. J. (2021). The overlapping neurobiology of induced and pathological anxiety: a meta-analysis of functional neural activation. American Journal of Psychiatry, 178(2), 156-164.

Best,
Lucy
     
Originally posted by Molly Rowlands: