Hi Bronwyn, 

to shorten equations, lets set t = years_form_baseline
a = age
g = group
s = sex

so your model is
Y_ij = b0 + b1 t_ij + b2 a_i + b3 g_i + b4 s_i + b5 t_ij a_i + b6 t_ij g_i + b7 a_i g_i + b8 t_ij a_i g_i
(as a fist step, I would consider simplifying it, by dropping the age interactions). 

Anyway

for male (s_i =0) and controls (g_i = 0) this reduces to
Y_ij = b0 + b1 t_ij + b2 a_i + b5 t_ij a_i 
so b1 is the slope for male controls, controlling for age and the slope age interaction. (0 1 0….)

Now for female (s=1) patients (g=1) we get this:

Y_ij = b0 + b1 t_ij + b2 a_i + b3  + b4 + b5 t_ij a_i + b6 t_ij + b7 a_i + b8 t_ij
       = (b0+b3+b4) + (b1 + b6 + b8 ) t_ij + (b2+b7) a_i + b5 t_ij a_i 
So the slope for female patients (controlling for age and age time interaction) would be
(b1 + b6 + b8)
0 1 0 0 0 0 1 0 1

The difference in slope between female patients and male controls would be
0 0 0 0 0 0 1 0 1 (or the negative of that depending which way you subtract). Similarly you can look at group differences (controlling for age gender and interactions). 

Always write out the full model to make sure you understand what you are doing. 

To complete the picture, here is the contrast for the slope of male patients
0 1 0 0 0 0 1 0 1 (it is the same as for female patients, because you don’t have a timeXgender interaction. So that is your patient slope )
Therefore the 
0 0 0 0 0 0 1 0 1  is the slope difference between the groups. 

I would recommend you talk to a local biostatistician, to make sure you are actually modelling what you want to model. And that you are interpreting the results correctly.  

Grüße, Martin

On 06 Mar 2017, at 18:56, Bronwyn Overs <b.overs@neura.edu.au> wrote:

Hi Martin,

Thank you for your response, that is much clearer. 

I am also a little confused about how to specify the exact contrasts we wish to test and was hoping to get some advice. My design matrix X includes the following columns:
1. Intercept
2. Years from baseline
3. Age at baseline
4. Group (patients labelled 1, controls 0)
5. Gender (females labelled 1, males 0)
6. Col 2 (years) * Col 3 (age)
7. Col 2 (years) * Col 4 (group)
8. Col 3 (age) * Col 4 (group)
9. Col 2 (years) * Col 3 (age) * Col 4 (group)

If I test the following contrast, is it giving me the effect of years across all groups and genders, or just years for male controls:
CM.C = [0 1 0 0 0 0 0 0 0]

Also, what contrast should I use to examine the effect of years in my patient group irrespective of gender?

Kind regards,

Bronwyn Overs
Research Assistant

Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
M 0411 308 769 T +61 2 9399 1883 F +61 2 9399 1265


On 4 Mar 2017, at 12:43 am, Martin Reuter <mreuter@nmr.mgh.harvard.edu> wrote:

Hi Bronwyn, 

I think years-between-scans should be years-from-baseline-scans . You may need to compute that if what you have is really years between neighbouring scans.

1. Usually people use intercept and maybe years-from-baseline as random effects. I would not include too many random effects, as it each adds a lot of free parameters and you need a lot of data to fit all that in a meaningful way. Which of your columns are random effects can be passed lme_fit_FS(X,[1 2],Y(:,1)+Y(:,2),ni);
for example has column 1 and 2 as random effects. 

2. You can do a model comparison as described on our wiki 

You run the more complex model first (do the EM init and maybe RgGrow and RgW fit) and then the simple one (only the EMinit and  RgW fit) and do a likelihodd ratio test.  An example is on the above wiki.

Best ,Martin 





On 27 Feb 2017, at 04:16, Bronwyn Overs <b.overs@neura.edu.au> wrote:

Dear mailing list,

I am trying to run a LME model using the matlab tools, but I’m unsure how to specify the model we wish to run. We have a qdec file that contains the following columns:
fsid, fsid-abse, years between scans, age at baseline, gender, group

We want to specify a model where we can examine four interaction terms (years*age, years*group, age*group, years*age*group), as well as random effects for the intercept, years and age. My questions are:
1. How do we specify a model that will include the random effects we want?
2. How do we compare our full model (3 random effects) with a model excluding the random effect for age?

Kind regards,

Bronwyn Overs
Research Assistant

Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
M 0411 308 769 T +61 2 9399 1883 F +61 2 9399 1265


_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.

_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer