Dear Anderson and colleagues,
I am following this discussion with interest as I wish (if possible) to set up a longitudinal TBSS analysis on 3 groups with 3 timepoints.
I have a healthy control group scanned at 0 months, 3 months and 12 months (no change expected) and two patients groups who underwent different surgical procedures with scans before, 3 months and 12 months after the procedure (the exact timings would be used in the design below).
From the previous posting, I have come up with the following (simplified to 2 subjects per group) to tackle this to look at linear changes with time. This
will be very complex with around 20 subjects per group. Is this the best way?
(Data) (Grp) (Time,controls) (Time,group1) (Time,group2) (Intercept,subj1) (Intercept,subj2) ... (Intercept,subj6)
(Control1,Visit1) 1 0 0 0 1 0 0 0 0 0
(Control1,Visit2) 1 3 0 0 1 0 0 0 0 0
(Control1,Visit3) 1 12 0 0 1 0 0 0 0 0
(Control2,Visit1) 2 0 0 0 0 1 0 0 0 0
(Control2,Visit2) 2 3 0 0 0 1 0 0 0 0
(Control2,Visit3) 2 12 0 0 0 1 0 0 0 0
(Group1-1,Visit1) 3 0 0 0 0 0 1 0 0 0
(Group1-1,Visit2) 3 0 3 0 0 0 1 0 0 0
(Group1-1,Visit3) 3 0 12 0 0 0 1 0 0 0
(Group1-2,Visit1) 4 0 0 0 0 0 0 1 0 0
(Group1-2,Visit2) 4 0 3 0 0 0 0 1 0 0
(Group1-2,Visit3) 4 0 12 0 0 0 0 1 0 0
(Group2-1,Visit1) 5 0 0 0 0 0 0 0 1 0
(Group2-1,Visit2) 5 0 0 3 0 0 0 0 1 0
(Group2-1,Visit3) 5 0 0 12 0 0 0 0 1 0
(Group2-2,Visit1) 6 0 0 0 0 0 0 0 0 1
(Group2-2,Visit2) 6 0 0 3 0 0 0 0 0 1
(Group2-2,Visit3) 6 0 0 12 0 0 0 0 0 1
Contrasts (using randomise with option --permuteBlocks)
[1 0 0 0 0 0 0 0 0] - where does FA increase linearly in control group over time? (should be none)
[-1 0 0 0 0 0 0 0 0] - where does FA decrease ...
[0 1 0 0 0 0 0 0 0] - where does FA increase following surgical procedure 1?
[0 -1 0 0 0 0 0 0 0] - where does FA decrease ...
[0 0 1 0 0 0 0 0 0] - where does FA increase following surgical procedure 2?
[0 0 -1 0 0 0 0 0 0] - where does FA decrease ...
Another possibility would be simply to do separate pairwise comparisons e.g. 3m vs 0m, 12m vs 0m. What is wrong with the following design very similar to
what was suggested by James?
(First 9 columns for group/timepoint, next 6 columns for intercept for each subject)
con-t1 con-t2 con-t3 grp1-t1 grp1-t2 grp1-t3 grp2-t1 grp2-t2 grp2-t3 con1 con2 grp1-1 grp1-2 grp2-1 grp2-2
1 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 1 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 1 0 0 0 0 0 0 1 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 1 0 0 0 0
0 1 0 0 0 0 0 0 0 0 1 0 0 0 0
0 0 1 0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 1 0 0 0
0 0 0 0 1 0 0 0 0 0 0 1 0 0 0
0 0 0 0 0 1 0 0 0 0 0 1 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 1 0 0
0 0 0 0 1 0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 1 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 1 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 1 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 1 0 0 0 0 1 0
0 0 0 0 0 0 1 0 0 0 0 0 0 0 1
0 0 0 0 0 0 0 1 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 1 0 0 0 0 0 1
Example contrasts
[-1 1 0 0 0 0 0 0 0 ...] - where is FA higher in control group at 3m compared to 0m?
[0 0 0 1 -1 0 0 0 0 ...] - where does FA fall following surgical procedure 1 by 3m?
Many thanks for advice.
Gavin
|