Thanks Martin. I assume in GLM approach I should calculate change for  session 1 and 2 and then 3 and 4 and then run difference of difference.

We actually randomized the order so half day1 is plcebo and half drug. So I just need to be caretabout the design for LME.

BR

On Mon, 30 Jan 2023 at 17:09, Reuter, Martin,Ph.D. <MREUTER@mgh.harvard.edu> wrote:

Hi Amirhossein,

So you have session 1 , placebo , session 2

Another day session 3,  drug, session 4 ?

Again if this is for all subjects, easiest is to subtract session 2 from 1, and 4 from 3, to get thickness/volume differences for each condition. Then compute the difference of the differences and run a GLM testing for difference from zero.

An LME approach could be:

Column of 1

Column of time (zero for session 1 and 3, one for session 2 and 4)

Column of day  (zero for session 1 and 2, one for session 3 and 4)

Column of drug (zero for session 1,2,3 and one for session 4)

The last one is the interesting one. But I would discuss this with a bio-statistician. I develop methods for image analysis and this could be wrong (or sub-optimal). 

Best, Martin 

On 30. Jan 2023, at 16:40, amirhossein manzouri <a.h.manzouri@gmail.com> wrote:

Thanks a lot Martin for the information.

We have actually 2 sessions of placebo for each subject. How do you suggest to do the analysis including that data?

BR

On Mon, 30 Jan 2023 at 16:30, Reuter, Martin,Ph.D. <MREUTER@mgh.harvard.edu> wrote:

Hi Amirhossein, 

- If you have two time points for all participants, 

- and the time difference is the same for all

you can simply subtract the thickness (or volume) values per participant and run a regular GLM. LME is a little overkill here. 

In LME, you have one column of ones, and one of the time (which is 0 and t alternating ) , this is not the time difference!  The first time point is at time 0 and the second at time t (in hours or days whatever). If the time really does not matter, you can also put 0 and 1. 

You can run the model with no random effect or with one random effect. The wiki MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be https://surfer.nmr.mgh.harvard.edu/fswiki/LinearMixedEffectsModels describes how to compare those models, also how to compute significance. 

You probably have more columns (also if you do the GLM) e.g. the amount of drug that was given, or who got the drug and who got placebo. Otherwise you cannot check for a drug effect. That column would be the one you are interested in. 

Without a placebo group, you will find difference across time, but you will not know if they are from the drug or from the fact that people are familiar with the scanner (less head motion) or more annoyed by the scanner (more head motion) or more tired, or more dehydrated, or rehydrated if you give the drug with water, or simply a time-of-the-day effect, or scanner heats up etc. 

Some of these confounders are problematic anyway, as the drug can have a sedative effect (less motion?) or was given with water (re-hydration). The second can be controlled by giving placebo with the same amount of water. Disentangling motion from drug effect (due to the possible correlation) would only be possible if you separately measure motion or take fMRI or diffusion motion estimates as a proxy for motion during T1. 

Head motion reduces grey matter estimates:

MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be doi.org

Dehydration effects: 

Responses of the Human Brain to Mild Dehydration and Rehydration Explored In Vivo by 1H-MR Imaging and Spectroscopy

MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be ajnr.org

 

Best, Martin 

On 30. Jan 2023, at 15:18, amirhossein manzouri <a.h.manzouri@gmail.com> wrote:

Hi,

I have 2 sessions of data acquired in the same day for each participant before and after the drug intake. I wonder how to analyse this with LME tool. I create design matrix X in 2 columns, first all ones and second the time differences(which are the same) and wonder if I need to only run the model with one random effect like

lhTh0_1RF = lme_mass_fit_EMinit(X,[1],Y,ni,lhcortex,3);

And what would be the next steps to get the stats and sig.mgh

Best regards,
Amirhossein Manzouri 

 

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 Mass General Brigham Compliance HelpLine at MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be https://www.massgeneralbrigham.org/complianceline .

--

Best regards,
Amirhossein Manzouri