Hi Sean

An important issue in the LME model concerns the centering of the times of measurement. It changes the interpretation of both the mean response and the variance of the random effects. If t_ijrepresents time since baseline then the subject-specific intercept coefficient ß₁+b_1irepresents the subject-specific mean response at baseline and Var(b_1i) is the between-subject variation in the mean response at baseline. On the other hand, if t_ijis an individual's age at the jth measurement occasion, then ß₁+b_1imight not have a useful interpretation since it represents the subject-specific mean response at age zero. Another choice is to center the times of measurement for all subjects at some common fixed age within the age range of the study participants (i.e. t_ij-a, for some fixed value a). In this case ß₁+b_1irepresents the subject-specific mean response at the common age a and Var(b_1i) is the between-subject variation in the mean response at that age.

You should use real-valued time measurements so the first and second scan times on the same subject are always different.

In your setting you only have two measurement occasions so you can only use a single random effect. If t_ijrepresents time since baseline then you can only use a random effect for the intercept term and add age-at-baseline as an additional time-independent covariate (its value doesn't change through measurement occasions). Otherwise if t_{ij doesn't}represent time since baseline you can test whether a single random effect for the time variable works better_{(i.e see if it produces higher optimal likelihood values). You can specify a random effect for the slope by indicating the column of the time variable:}

[stats, st] = lme_mass_fit_vw(X,3,Y,ni,Mask,[],12);

BTW it is best to order your design matrix in the following way (just for organizational purposes):

X = [Intercept Age(at scan) Group Group*Age Gender] so the time-variying covariates appear first.

You don't need to re-do the design matrix to test the effect of age within a group (group-specific slope different from zero) just use the appropriate contrast.

Hope this helps

-Jorge

El Jueves 12 de diciembre de 2013 14:29, Seán Froudist Walsh <froudiss@tcd.ie> escribió:

Dear Jorge and FreeSurfers,

Would you mind having a look at my set-up and advising on whether it is correct or needs changing?

I have participants with one or two scans, who are divided into two groups. I'm interested in the longitudinal effects of aging, and the difference in the effect of age on the two groups. Not all people were scanned at exactly the same age, and the interscan interval in not exactly the same for each person. Rather, the age at scan looks like a bimodal distribution around the mean 1st and 2nd timepoint.

I believe I read on the list that you should have more timepoints than random effects. I guess I should then only name one random effect (intercept).

At the moment I have my model set up as follows:

X = Intercept Group Age(at scan) Group*Age Gender
[stats, st] = lme_mass_fit_vw(X,1,Y,ni,Mask,[],12);

CM.C = [0 0 0 1 0]

I guess I should then run lme_mass_F(stats,CM);

and lme_mass_FDR2.

Here I am trying to look find the areas that develop differently between the two groups over the timespan we are studying. If you would be able to advise me as to whether I have set up the model wrong it would be greatly appreciated. Age in this case refers to the the age at scan time, regardless of whether it is the first or second scan, so many participants will have two distinct values in the age column.

Also, is it possible within this set-up to test the effect of age within a group, or would that require re-doing the design matrix?

Finally, is it possible to include the by-subject random slopes for the effect of age? In R (lmer) it would be something like (1+age|subject).

Best wishes y gracias,

Seán

_______________________________________________
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.