Dear experts,

I’m observing – what I think – are implausible results from a simple correlational analysis between FA maps and body mass index, and was hoping if someone could shed some insight into what I’m doing wrong.

Briefly:

I have two groups of subjects, patients vs. controls (n = 21 in each group). I ran the entire preprocessing + TBSS pipeline as recommended. I then ran an unpaired t-test comparison of FA maps (while controlling for body mass index, age, and head motion) . This showed no significant between-group differences using randomize (TFCE, p < .05 corrected).

As secondary analyses, I wanted to perform exploratory correlational analyses between FA maps and body mass index (within each group separately). I separated the all_FA_skeletonised 4D file into 2 separate 4D files, one for patients and one for controls. Then I generated a GLM for each group separately, with 3 regressors: body mass index, age, and head motion (the two latter are covariates of no interest). No demeaning was performed. I created the following t-contrasts: 1 0 0 and -1 0 0. This should give me the positive and negative correlations between FA maps and body mass index, while controlling for age and head motion.

I then ran: randomise –i FA_Patients_skeletonised –o BMI –m mean_FA_skeleton_mask –d BMI.mat –t BMI.con –n 5000 –T2 –uncorrp -V

I did this for the controls as well, using their FA maps.

When looking at the BMI_tfce_corrp_tstat1 files (positive correlations) for each of the two groups, with a threshold of min 0.95 and max 1, there are significant results throughout all WM fibers. Indeed, the images look more like a WM skeleton file, and not a result image (see attachment). This is the case for both the patient and the control group. This is NOT the case for the BMI_tfce_corrp_tstat2 files though, which showed no results.

I also performed similar correlational analyses using the same input files and basic GLM design, but with another regressor of interest (a self-report measure). This analysis showed no significant correlations.

See attachments for my design and results (thresholded at p < .05 corrected using tbss_fill).

Am I doing something wrong here; and why do these “implausible” results occur only with positive correlations between BMI and FA maps?

All help is highly appreciated!

Best,

-Lasse

Lasse Bang, Ph.D

Postdoctoral Researcher

Regional Department for Eating Disorders (RASP)

Oslo University Hospital, Ullevål HF

Oslo, Norway

E-mail: [log in to unmask] / [log in to unmask]

Phone: +47 23 02 73 71 /+47 41 42 97 04

IKKE SENSITIVT INNHOLD