question about longitudinal analyses in two treatment arms

AFNI Message Board
1

Hello All,

I have the following data –

I have 205 subjects - each subject was imaged at 3 timepoints (baseline, followup 1 and followup 2)
The 205 subjects are split in two treatment arms with 100 subjects in the first one and 105 subjects in the second one.

I have used freesurfer to parcellate each subject’s image into 187 ROIs.
For each ROI, I have the average cortical thickness for that ROI.
For each timepoint, I have saved the results as a matrix (cortical thickness in 187 ROIs x 205 subjects) in a text file.
So I have three files for the three timepoints.

I have the age and cognitive score for each subject at each timepoint. And their gender.

I want to answer the following questions –

Do the baseline cortical thickness correlate with the corresponding cognitive score at baseline ?
I am interested in testing whether the cortical thickness changed significantly across the different timepoints and does that relate to the change in the cognitive score
Is there an effect of treatment on this change in cortical thickness across time ?

Can you please suggest what programs in afni will allow me to test these questions ?

Thank you very much,

Cheers

Rito

2

Rito,

The 205 subjects are split in two treatment arms with 100 subjects in the first one and 105 subjects in the second one.

What is the difference between these two arms?

Do the baseline cortical thickness correlate with the corresponding cognitive score at baseline ?

This can be addressed through a simple regression model using, for example, 3dttest++.

I am interested in testing whether the cortical thickness changed significantly across the different timepoints and does that
relate to the change in the cognitive score. Is there an effect of treatment on this change in cortical thickness across time ?

I assume that both age and cognitive score vary across the three time points. You can use 3dLMEr to address these two questions.

The above approaches are all under the conventional mass univariate framework, so you would have to deal with the multiple testing issue. It is possible to adopt the Bayesian multilevel modeling framework so that the multiplicity issue would be resolved, but the associated program RBA would have to be modified for your situation.

3

Hi Rito,

To correlate cortical thickness with cognitive score at baseline, that would presumably be done were the correlation is across subjects for each ROI, does that seem correct? So using 3dttest++ as Gang suggets, the input files would be one list of ROI values (baseline) per subject. Then the cognitive scores would be applied as a covariate.

<Gang, please correct me if this does not seem correct.>

Comparison of the two groups would then be a 2-sample t-test. So A command would look similar to the class demo with a covariate file, AFNI_data6/group_results/s6.ttest.covary .

In this case it might read something like:

3dttest++ -prefix cognitive_100_vs_105    \
          -AminusB                              \
          -setA group_100                       \
               s0000 s0000_ROIs.1D              \
               s0001 s0001_ROIs.1D              \
               ...                              \
               s0099 s0099_ROIs.1D              \
          -setB group_105                       \
               s1000 s1000_ROIs.1D              \
               s1001 s1001_ROIs.1D              \
               ...                              \
               s1104 s1104_ROIs.1D              \
          -covariates cognitive_scores_covary.txt

Where s0000_ROIs.1D would have the 187 ROI averages, one value per row.
And where cognitive_scores_covary.txt would be a text file with one row per subject, and might look like:

subject cog_score
s0000  43.7
s0001  86.0
...
s1104  56.2

Does that seem reasonable?

  • rick