Longitudinal VBM with covariates

[Susanne Stickel]

  Dear Experts,

this is the first time I am using the SwE toolbox and I have a really basic question as I am not quite sure if I am even using the toolbox correctly.
I have 20 postpartum women with 4 sMRI measurements (3 weeks apart) and we got an infant attachment score at all time points. The goal is to study the relationship between attachment score and brain structure.

I did two different analyses, one with only the attachment scores as the variable of interest and one where I also have total intracranial volume and age as covariates. When I have only the attachment score in the model, the results look very nice and the patterns are also consistent with the literature on the maternal brain and attachment. However, when I have TIV and age as covariates, there are no supratresholds anymore, which I find quite strange. In "normal" VBM analyses the covariates do not have such an extremely large impact on the results, so I am afraid that my models do not make sense at all and that I am already making a big mistake with the design matrix....
I have attached two screenshots to the design matrices and hope it is enough to tell me what mistake I am making here....




Regards, 

Susanne

[Grant Tays]

Hi Susanne,

I'm not particularly an expert on all of this yet, but I've been working with SwE on a number of data sets and have gotten a lot of feedback from Dr. Nichols here, so I hope he can step in and answer more thoroughly.  We have had similar styles of analysis, and were told that with a group of this size, SwE can be limited. Dr. Nichols advice to my team members and I was basically to collapse the multiple timepoints into one column that is weighted with however you expect the changes to occur (linearly increasing, stable, etc). I think this would apply to your covariate columns in particular to maximize degrees of freedom.  Additionally, some of the SwE options are better for the smaller sample size.  Hope it helps a little!