Re: [Freesurfer] FSGD design for multi-centre study

List overview All Threads
Download

newer

older

QDEC vs. Command Line Stream

Template recon-all register...

C.P.E. Rollins

9 Apr 2018 9 Apr '18

9:50 a.m.

Hi Douglas,

Thanks for your response. I just wanted to confirm how the FSGD and contrast files might look for the multicentre study. If I'm to create a separate class for each combination of diagnosis, sex, and center, would there then be 24 classes? (as there are 2 diagnoses (patient vs. control), 2 sex, and 6 centres). For instance: GroupDescriptorFile 1 Group control_Male_centre1 Group control_Female_centre1 Group control_Male_centre2 Group control_Female_centre2 Group control_Male_centre3 Group control_Female_centre3 Group control_Male_centre4 Group control_Female_centre4 Group control_Male_centre5 Group control_Female_centre5 Group control_Male_centre6 Group control_Female_centre6 Group patient_Male_centre1 Group patient_Female_centre1 Group patient_Male_centre2 Group patient_Female_centre2 ...

Then, my contrast between patients and controls would look like 1 1 1 1 1 1 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 0 for a doss model, if I'm additionally controlling for age, or 24 zeros (instead of 1) for a dods model? Moreover, should I expect any issues with using such a high number of classes?

Thanks again, Colleen

Show replies by date

Douglas N. Greve

9 Apr 9 Apr

10:58 a.m.

New subject: FSGD design for multi-centre study

That looks right to me. If you use a DOSS model and have a lot of subjects, I don't think there should be a problem with so many classes.

On 04/09/2018 09:50 AM, C.P.E. Rollins wrote:

...

Hi Douglas,

Thanks for your response. I just wanted to confirm how the FSGD and contrast files might look for the multicentre study. If I'm to create a separate class for each combination of diagnosis, sex, and center, would there then be 24 classes? (as there are 2 diagnoses (patient vs. control), 2 sex, and 6 centres). For instance: GroupDescriptorFile 1 Group control_Male_centre1 Group control_Female_centre1 Group control_Male_centre2 Group control_Female_centre2 Group control_Male_centre3 Group control_Female_centre3 Group control_Male_centre4 Group control_Female_centre4 Group control_Male_centre5 Group control_Female_centre5 Group control_Male_centre6 Group control_Female_centre6 Group patient_Male_centre1 Group patient_Female_centre1 Group patient_Male_centre2 Group patient_Female_centre2 ...

Then, my contrast between patients and controls would look like 1 1 1 1 1 1 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 0 for a doss model, if I'm additionally controlling for age, or 24 zeros (instead of 1) for a dods model? Moreover, should I expect any issues with using such a high number of classes?

Thanks again, Colleen

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

C.P.E. Rollins

10 Apr 10 Apr

6:55 a.m.

New subject: FSGD design for multi-centre study

Thanks for your help. I've noticed, however, that 2 of the centres contribute a smaller number of subjects and don't in fact have any female controls. Should I remove these as classes (ie. take out control_Female_centre1 and control_Female_centre2) and then change the subsequent contrast file to look like: 1 1 1 1 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 0 Or would this be a problem since now the number of classes that are controls or patients are uneven (10 vs. 12)?

Thanks again, Colleen

--- Colleen Rollins, PhD Candidate Department of Psychiatry University of Cambridge Herchel Smith Building, Robinson Way, Cambridge CB2 0SZ

On 2018-04-09 14:50, C.P.E. Rollins wrote:

...

Hi Douglas,

Thanks for your response. I just wanted to confirm how the FSGD and contrast files might look for the multicentre study. If I'm to create a separate class for each combination of diagnosis, sex, and center, would there then be 24 classes? (as there are 2 diagnoses (patient vs. control), 2 sex, and 6 centres). For instance: GroupDescriptorFile 1 Group control_Male_centre1 Group control_Female_centre1 Group control_Male_centre2 Group control_Female_centre2 Group control_Male_centre3 Group control_Female_centre3 Group control_Male_centre4 Group control_Female_centre4 Group control_Male_centre5 Group control_Female_centre5 Group control_Male_centre6 Group control_Female_centre6 Group patient_Male_centre1 Group patient_Female_centre1 Group patient_Male_centre2 Group patient_Female_centre2 ...

Then, my contrast between patients and controls would look like 1 1 1 1 1 1 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 0 for a doss model, if I'm additionally controlling for age, or 24 zeros (instead of 1) for a dods model? Moreover, should I expect any issues with using such a high number of classes?

Thanks again, Colleen

C.P.E. Rollins

16 Apr 16 Apr

12:58 p.m.

New subject: FSGD design for multi-centre study

Hi,

I just wanted to ask again regarding my question below.

I've noticed that 2 of the centres contribute a smaller number of subjects and don't in fact have any female controls. Should I remove these as classes (ie. take out control_Female_centre1 and control_Female_centre2) and then change the subsequent contrast file to look like: 1 1 1 1 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 0 Or would this be a problem since now the number of classes that are controls or patients are uneven (10 vs. 12)?

Thanks again,

Colleen

Douglas Greve

17 Apr 17 Apr

12:24 p.m.

New subject: FSGD design for multi-centre study

sorry for the delay ...

You will definitely need to change the contrast values so that the positives and negatives both add to the same number (with opposite sign). It is a bit of a judgement call in terms of removing them entirely. With an unbalanced design, you cannot remove the fixed effect of that particular site, and reviewer might object to that. I would probably run it both with and without to see if it really makes a difference. You could also test the significance of the other site effects and, if they are not sig, then argue that the site effect is small.

On 4/16/18 12:58 PM, C.P.E. Rollins wrote:

...

Hi,

I just wanted to ask again regarding my question below.

I've noticed that 2 of the centres contribute a smaller number of subjects and don't in fact have any female controls. Should I remove these as classes (ie. take out control_Female_centre1 and control_Female_centre2) and then change the subsequent contrast file to look like: 1 1 1 1 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 0 Or would this be a problem since now the number of classes that are controls or patients are uneven (10 vs. 12)?

Thanks again,

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

C.P.E. Rollins

20 Apr 20 Apr

6:19 a.m.

New subject: FSGD design for multi-centre study

External Email - Use Caution

Thanks a lot for the explanation. The issue is that I don't think Freesurfer will run a design for which there are no subjects for a given class. So if I keep the classes (24 classes since 6 centres x 2 gender x 2 groups (patient vs. control), and have the contrast with 12 (-1) and 12 (1), -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 1 1 1 1 1 1 1 1 1 1 1 1 0 I get the error: -------------------------------- ERROR: matrix is ill-conditioned or badly scaled, condno = 1e+08 -------------------------------- Possible problem with experimental design: Check for duplicate entries and/or lack of range of continuous variables within a class.

I'm assuming this is because there are no participants in, for example, patient_centre2_female Is there any way to get around this issue, or should I remove those centres from my analysis (since I can't only remove "patient_centre2_female", as this would make the contrast unbalanced (positives and negatives would not add to the same number). I hope this was clear but please let me know if it was not.

Thanks again, Colleen

Douglas Greve

24 Apr 24 Apr

1:10 p.m.

New subject: FSGD design for multi-centre study

Just remove the class.

ps. Please remember to include previous correspondences so we have context

On 4/20/18 6:19 AM, C.P.E. Rollins wrote:

...

     External Email - Use Caution
Thanks a lot for the explanation. The issue is that I don't think Freesurfer will run a design for which there are no subjects for a given class. So if I keep the classes (24 classes since 6 centres x 2 gender x 2 groups (patient vs. control), and have the contrast with 12 (-1) and 12 (1), -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 1 1 1 1 1 1 1 1 1 1 1 1 0 I get the error:

ERROR: matrix is ill-conditioned or badly scaled, condno = 1e+08

Possible problem with experimental design: Check for duplicate entries and/or lack of range of continuous variables within a class.

I'm assuming this is because there are no participants in, for example, patient_centre2_female Is there any way to get around this issue, or should I remove those centres from my analysis (since I can't only remove "patient_centre2_female", as this would make the contrast unbalanced (positives and negatives would not add to the same number). I hope this was clear but please let me know if it was not.

Thanks again, Colleen

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

2963

Age (days ago)

2978

Last active (days ago)

freesurfer@nmr.mgh.harvard.edu

6 comments

3 participants

tags (0)

participants (3)

C.P.E. Rollins
Douglas Greve
Douglas N. Greve