Dear experts,
my questions concerns VBM analyses between patients and controls with additional covariates. Specifically, I would like to perform a multiple regression on grey matter concentrations and clinical scores.
I analyzed structural T1-images of patients and controls by using the vbm5-toolbox and spm5 (Matlab R2010a). I observed differences between both groups by means of a two-sample t-test adjusted for TIV.
In general, the main goal of my study is to get information about patients’ co-morbidity and its relationship to grey matter values.
Now, if I want to compute a multiple regression with clinical scores, there seem to be several ways to do this. Here are a few of them which I have found in various recent publications:
1) Performing a 2-sample t-Test of patients vs. controls and then selecting those clusters that show a significant effect in the between-group comparison; extracting the values from theses clusters in every single patient and performing an external multiple regression
2) Firstly, performing a 2-sample t-Test for patients vs. controls; secondly, performing a multiple regression of grey matter concentrations and clinical scores only for the patients in SPM5
I already tried this and found very interesting results for the multiple regression, however in areas which have not become significant in the 2-sample t-Test. How can I interpret these findings with respect to the between-group comparison?
3) Including multiple regressors in the 2nd level 2-sample t-Test in SPM5 (one value for each score and each subject, i.e. patients and controls) and assessing their respective effects in the whole group.
I would be very grateful to know which of these approaches you would recommend for analyzing associations between grey matter values and clinical scores.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dr. Dipl.-Psych. Matthias Wittfoth
Department of Neurology
Hannover Medical School
Carl-Neuberg-Str. 1
30625 Hannover, Germany
Tel.: ++49-(0)511-532-2439
[log in to unmask]
member of
NeuroImaging and Clinical Applications (www.nica-hannover.net)
|