Hi FSL experts,
I have a question regarding an event related phMRI with covariate. We have a group of subjects that received IV drug on one occasion and placebo on another. We created individual regressors using their individual blood levels of drug (interpolated curve across all volumes and z transformed).
We have run the higher level(s) a number of ways (with arguments ensuing regarding the most appropriate and interpretable method). One issue of note is that this is a very small number of subjects (5).
A number of things are of interest. The simple dependent t (activation of drug greater than placebo). The dependent t with a covariate. Here we have a choice of two highly correlated covariates - one is the peak drug value (demeaned?), the other is a measure of drug exposure history (demeaned?).
One method we have tried is a two level higher level analysis. First using lower level feat directories to get the paired copes for input to the next level.
group EV1 s1 s2 etc with contrast for EV1 = +1 all other EVs =0
1 1 1.0 0 using a fixed effects model
1 -1.0 1.0 0
1 1 0 1.0
1 -1.0 0 1.0
These copes are then used at the next level with group = 1 for all and EV1 = 1 for all. This will tell us how much these subjects activated to the drug on average (having already accounted for the difference between drug and placebo at the previous level).
This can then be run by itself or with one of the covariates - demeaned or not???
One question is whether this level should be a mixed (flame 1) or fixed design, given that there are only 5 subjects AND it may not make sense to generalize beyond this group of subjects.
ALTERNATIVELY
We have run these subjects with just one higher level
group EV1 s1 s2 etc with contrast for EV1 = +1 (and -1 if desired) all other EVs =0
1 1 1.0 0 using a mixed effects model (again question re: fixed or not given above mentioned issues)
1 1 0 1.0
1 -1.0 1.0 0
1 -1.0 0 1.0
This will tell us how much activation for drug>placebo (or vice versa)
However, adding in the covariate is a little more confusing. We don't have a peak drug measure during the placebo scan (subjects have a baseline only and were abstinent before the scan), so this can be 0. With both active and placebo in the same design, do we demean the covariate as one group, ie include the zeros, or separate the runs, demeaning just the active and zeros for the placebos? If we use the exposure history as a covariate, the measure is the same for both runs.
Is there some way to make sense of all of this?
It is interesting to see the activation in response to the drug. But, it is also interesting to note that the peak blood measures of drug were very different in these subjects (given the same dose/body weight). And that this was highly correlated with their exposure history. This is a drug that definitely crosses the BBB very rapidly. It might be expected that subjects' activation would be affected by their individual exposure/blood levels.
Thank you so much.
Regards,
Rinah
|