Hi Jodie,
are you familiar with mixed effects models? If not I would recommend you include a biostatistician. What I mean (and you can do this in the matlab code, no need to include it in the qdec file) is to use a piecewise linear model.
So for a subject with visits at 0, 0.4 and 0.9 years you would have this design matrix:
1 0 0 1 0.4 0 1 0.9 0.5
your model could be like this: Y_ij = beta0 + b_i + beta1 x_ij + beta2 z_ij +e_ij
(i is subject index, j is time point: 1,2,3)
where you have a random effect b_i for a subject specific intercept, x_ij is years and z_ij is the third column in your matrix: z_i1 = 0 z_i2 = 0 z_i3 = x_i3 - x_i2
for hypothesis testing contrast, see my email below.
Best, Martin
On 05/12/2014 01:14 PM, Jodie Davies-Thompson wrote:
Hi Martin,
Thanks for your reply. I'm not entirely sure what you mean (sorry - I'm new to this!). Do you mean something like this?
For examining an increase between timepoint 1 and 2, but the same between timepoint 2 and 3?
fsid fsid-base years years-to-years Day1 base 0 0.4 Day2 base 0.4 0.4 Day3 base 0.8 0
All the best, Jodie
Message: 6 Date: Fri, 09 May 2014 16:52:25 -0400 From: Martin Reuter mreuter@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] nonlinear models in longitudinal analysis To: freesurfer@nmr.mgh.harvard.edu Message-ID: 536D4009.7030907@nmr.mgh.harvard.edu Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Hi Jodie,
'years' should contain the real time between time points (usually, unless you have test-retest where time is irrelevant, or different task that can be ordered as you like). For what you are trying to do you should use linear mixed effects models with a piecewise linear setup. This will allow you to estimate a slope before and after timepoint 2 and compare those slopes. Your design matrix would probably have the column of 1, years and a column with years-years(at tp2 of that subject) if that value is positive 0 otherwise
Testing if the coef of the 3rd column is > 0 is asking if the slope increases in the second interval. Not exactly what you want, but close and of course you can test if the slope in interval 1 is different from zero (second column) and if the slope in interval 2 is different from zero ( 0 1 1, ie the sum of the last two columns).
Best, Martin
On 05/08/2014 03:48 PM, Jodie Davies-Thompson wrote:
Dear Freesurfer experts,
I am attempting to do a longitudinal analysis of cortical thickness with 3 timepoints. So far, I have successfully completed the analysis for long_mris_slopes comparing timepoint 1 to timepoint 2, timepoint 2 to timepoint 3, and a linear increase for timepoint 1 to 2 to 3.
However, I am primarily interested in finding regions which show a) an increase between timepoint 1 and 2, but no difference between 2 and 3, and b) no difference between timepoint 1 and 2, but an increase between 2 and 3.
Would the correct way of 'modelling' these two in the qdec file be:
a) fsid fsid-base years day1 base 0 day2 base 3 day3 base 3
and b) fsid fsid-base years day1 base 0 day2 base 0 day3 base 3
All the best, Jodie
Jodie Davies-Thompson, Postdoctoral Fellow Department of Ophthalmology & Visual Sciences UBC/VGH Eye Care Centre 2550 Willow Street Vancouver, BC, V5Z 3N9 Canada Tel: 604-875-4111 ext 69003
Jodie Davies-Thompson, Postdoctoral Fellow Department of Ophthalmology & Visual Sciences UBC/VGH Eye Care Centre 2550 Willow Street Vancouver, BC, V5Z 3N9 Canada Tel: 604-875-4111 ext 69003
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer