sdm-help-list > Significance thresholds, anisotropy, probability, modality and mask
Showing 1-5 of 5 posts
Display:
Results per page:
Jul 24, 2020  07:07 AM | tmp
Significance thresholds, anisotropy, probability, modality and mask
Dear experts,

I am almost ready to conduct a meta-analysis using SDM, however, I have some questions before I can start the final analysis.

In their paper: Radua, J., & Mataix-Cols, D. (2009). Voxel-wise meta-analysis of grey matter changes in obsessive–compulsive disorder. British Journal of Psychiatry, 195(5), 393–402. https://doi.org/10.1192/bjp.bp.108.05504... the authors state: "We focus on results with uncorrected P<0.001 because we found in the previous simulations that uncorrected P<0.001 or even 0.002 was, in our method, empirically equivalent to corrected P<0.05."
On the other hand I have read that there should not be any intra-study differences in significance thresholds. However, if I admit results obtained with an uncorrected threshold of <.001 could I not also admit results from the same study with a more conservative threshold, say a corrected p<.05?
How about inter-study differences? Since I don't input the significance threshold or the p-values of the results anywhere in SDM (or am I wrong?) I am thinking about how to deal with different significance thresholds between studies.
The quote above is why I'm leaning to include studies with an uncorrected threshold of p<0.001 and above, meaning I would also include results from analyses with a more conservative threshold. Do you think that would be okay?

Concerning the Anisotropy, would you recommend to use full anisotropy for all meta-analyses of neuroimaging studies? Do I understand correctly that if full anisotropy is used the value of the isotropic FWHM does not have much influence on the results?

Lastly, could you recommend a resource to determine which probability I should use in the "Threshold" tab? And do you have any recommendations for which modality and mask to use in the preprocessing when analyzing cortical thickness data and gray matter density data respectively?

I would greatly appreciate your help!

Thank you in advance!
Aug 3, 2020  08:08 AM | Lydia Fortea - Instituto de Investigaciones Biomédicas August Pi i Sunyer (IDIBAPS)
RE: Significance thresholds, anisotropy, probability, modality and mask
Hello, 

In the tutorial of the main sdmproject webpage you have the recommendations of how to include the peaks from each study depending of the different threshold used (section 4. Collection data, 3). Moreover in the sdm table you should specify the threshold used for each study.

Considering the Anisotropy or FWHM, I would recommend you to use the default values. 

The parameters to use in the threshold tab are personal choice, depending if you want results more conservative and robust (pv voxel_corrected < 0.01 and cluster extent of 100) or more permisives (pv_uncorrected < 0.05 and cluster extent of 1)

The modality when handling gray matter (for both, cortical thickness and density) is the VBM-gray matter, and the mask would be freesurfer for cortical thickness and grey matter for density

Kind regards,
Lydia
Aug 5, 2020  02:08 AM | tmp
RE: Significance thresholds, anisotropy, probability, modality and mask
Dear Lydia,

thank you for your reply!

I wonder if you could help me with some other questions I have:

I have read that "FreeSurfer does not report true "Talairach" coordinates. The coordinates listed under "Talairach" are actually based on Matthew Brett's 10/8/98 non-linear transform from MNI305 space." (Source: https://surfer.nmr.mgh.harvard.edu/fswik...) Would you recommend to code all results from FreeSurfer analyses that are reported in papers as "Talairach", as "Brett" for the purpose of a meta-analysis? Further, since FreeSurfer is not listed in the list of software packages I assume this would be coded as "other" in the naming of the txt files?

About possible covariates, so far I haven't found any information about how to include dimensional covariates. Is this possible or would I have to convert them into categorical variables?

Concerning the filter in the mean analysis, currently I can only choose one of my variables as a filter and unfortunately I can't find an explanation of what exactly filtering means in this context. Could you point me in the direction of where to find a detailed explanation of the filtering?

Speaking of t_thr, I understand that I should code the critical t-value that was used for the analyses in the studies I'm including. Is that correct? Since many studies don't directly report this I'm wondering if I can use the tool https://www.sdmproject.com/utilities/?sh... to convert significance thresholds to t-thresholds?

About the tool: I have noticed that it gives me different results depending on which "decimal indicator" I use (a comma or a period). Could you tell me which one is the correct one?

Thanks again, I really appreciate your help!

Tania
Aug 7, 2020  11:08 AM | Lydia Fortea - Instituto de Investigaciones Biomédicas August Pi i Sunyer (IDIBAPS)
RE: Significance thresholds, anisotropy, probability, modality and mask
Hello Tania, 

It seems it is true what you said about the freesurfer reported coordinates, so maybe you should check properly if they didn't convert the Brett coordinates into Talairach before reporting the results; and if they didn't report it change the name to Brett. Indeed, you should put the code "other" in the text files.

For the covariables, you can use numerical variables (continuous, ordinal) or categorical.

The filter is used if you want to perform a meta-analysis for a subgroup. The variable here needs to be categorical and binomial (so SDM can differentiate two groups). For example, only take those papers that study the effect in adult samples.

Speaking of t_thr, you are right and yes, you can convert the reported values into the ones you need, with the conversion tool provided by SDM. You should use period to write down your values.

Kind regards,
Lydia
Aug 8, 2020  06:08 AM | tmp
RE: Significance thresholds, anisotropy, probability, modality and mask
Thanks so much! :)
Originally posted by Lydia Fortea:
Hello Tania, 

It seems it is true what you said about the freesurfer reported coordinates, so maybe you should check properly if they didn't convert the Brett coordinates into Talairach before reporting the results; and if they didn't report it change the name to Brett. Indeed, you should put the code "other" in the text files.

For the covariables, you can use numerical variables (continuous, ordinal) or categorical.

The filter is used if you want to perform a meta-analysis for a subgroup. The variable here needs to be categorical and binomial (so SDM can differentiate two groups). For example, only take those papers that study the effect in adult samples.

Speaking of t_thr, you are right and yes, you can convert the reported values into the ones you need, with the conversion tool provided by SDM. You should use period to write down your values.

Kind regards,
Lydia