group by time interaction in 3dLME and missing data

AFNI Message Board
1

Hi Gang,

I have a study in which I’m comparing activation between patients and controls. Most participants have 2 timepoints, but not all. Is it possible to include all data (including single timepoints) in using 3dLME? If so, how would I model the time between visits?

thanks in advance,
Lara

2

Hi Lara,

Is it possible to include all data (including single timepoints) in using 3dLME?

Yes, you can include all the data you have as input for 3dLME or 3dLMEr.

If so, how would I model the time between visits?

Do you mean how to specify the model? It’s the same as if there were no missing data.

3

Hi Gang,

Thanks for your quick reply!

My setup is below. In addition to group, sex and scan site, I have baseline age (age at time 1), and interval (duration between time 1 and time 2) in my model. My specific question is for participants that only have 1 scan - would I leave “interval” blank?

On another note, I’d appreciate your feedback on whether what I have here is appropriate.

Many thanks again,
Lara

3dLME -prefix /path/to/output/output.nii -jobs 6
-model ‘timegroup+age+sex+interscan+site’
-qVars ‘age,interscan’
-qVarsCenters ‘12.79,2.1’
-ranEff ‘~1’
-SS_type 3
-mask /path/to/mask/mask.nii
-num_glt 1
-gltLabel 1 ‘grpXtime’ -gltCode 1 'group : 1hc -1pt time : 1time1 -1*time2’
-dataTable
Subj time group age sex site interscan InputFile
13649 time1 ctl 13.59 M 1 2.25 13649_GNG_cope2_time3.nii.gz
13649 time2 ctl 13.59 M 1 2.25 13649_GNG_cope2_time4.nii.gz
15388 time1 ctl 11.64 F 2 ??? 15388_GNG_cope2_time3.nii.gz
15341 time1 ctl 12.87 F 3 2.04 15341_GNG_cope2_time3.nii.gz
15341 time2 ctl 12.87 F 3 2.04 15341_GNG_cope2_time4.nii.gz
…
15350 time1 pt 11.69 F 1 2.06 15350_GNG_cope2_time3.nii.gz
15350 time2 pt 11.69 F 1 2.06 15350_GNG_cope2_time4.nii.gz
15398 time1 pt 10.42 F 3 2.07 15398_GNG_cope2_time3.nii.gz
15398 time2 pt 10.42 F 3 2.07 15398_GNG_cope2_time4.nii.gz
15403 time1 pt 12.45 M 2 ??? 15403_GNG_cope2_time3.nii.gz
…

4

Hi Lara,

My specific question is for participants that only have 1 scan - would I leave “interval” blank?

No, remove those rows where you have missing data. 3dLME does not have the imputation mechanism.

On another note, I’d appreciate your feedback on whether what I have here is appropriate.

It looks fine to me.

5

Hi Gang,

Is there another way to construct the model so that all timepoints (including subjects with only 1 timepoint) can be used?

Thanks,
Lara

6

Unfortunately no, unless you’re able to feed in the missing values of the explanatory variable.

7

I see. This may be a silly follow-up question, but for participants with only a single timepoint, would it be acceptable to enter a single uniform value (e.g., 0 or 999) for the “interval” variable?

thanks,
Lara

8

for participants with only a single timepoint, would it be acceptable to enter a single uniform value (e.g., 0 or 999) for the “interval” variable?

What do you mean by ‘a single uniform value’? A fake value? I would simply remove those rows with missing data, and see how the result looks like.

9

Hi Gang,

Getting back to this thread after some time away. I’d like to take a fresh shot at using 3dLME with a structural analysis. I have two groups of subjects with up to 4 timepoints per subject. However, some subjects have as little as one timepoint.

Instead of including a variable that notes the duration between scans (which would cause missing data for subjects with only one scan), could I instead have a variable that denotes time since baseline? This would mean that all subjects could be included in my analysis.

Thanks,
Lara

10

could I instead have a variable that denotes time since baseline?

You could, but it would be a different model with different assumptions. I would pay more attention to the model assumption than the missingness of the data.