help
help > RE: How to use "age" as covariates?
Jun 13, 2014 02:06 AM | Alfonso Nieto-Castanon - Boston University
RE: How to use "age" as covariates?
Hi Yifei,
I do not see anything out of order in your conn_*.mat file (perhaps only that you seemed to have removed the global signal as part of the preprocessing step, and that you are using an explicit mask for voxel-level analyses, which I assume may be a gray-matter mask, although since you seem to be mostly performing ROI-to-ROI analyses this would not come into play anyway).
Just to make sure everything looks correct I would only check a couple of additional potential issues:
1) Given that you are skipping the standard temporal preprocessing options (confound removal), since you have already-preprocessed BOLD timeseries, the main thing I would check is to make sure that the histogram displays shown in the Preprocessing step look reasonably centered (mean zero) and scaled (not too large variance) for each subject. If that is not the case that would signal that your preprocessing steps have not sufficiently removed potential biases in the BOLD signal and I would consider perhaps adding motion regressors or other potential confounding effects in the preprocessing step.
and 2) Since you are using non-standard ROIs I would also check that all of your functional and ROI files are correctly coregistered (use the corresponding 'check coregistration' buttons to check this)
If the voxel-to-voxel histogram displays in the Preprocessing step look correct, and the functional/ROI files seem to be correctly coregistered, then I do not see anything else to worry about regarding the original Setup/Preprocessing steps.
Regarding your second-level analyses, I believe they are correct but there might be some issues there depending on whether I am interpreting the order of your contrast vectors correctly, so let me rephrase the proper contrast vectors for your different analyses just to make sure (I am listing below the subject effects sorted in the order I believe they will appear in your gui, if they appear in different order you need to resort the contrast vector values accordingly):
1) to compare between the two groups (HC and AD) the association between FC and performance:
Select: HC, AD, performance_HC, performance_HD, age, gender
Enter contrast: [0 0 1 -1 0 0];
2) to compare the association between FC and performance within one group (e.g. HC), but control for age and gender:
Select: HC, performance_HC, age_HC, gender_HC
Enter contrast: [0 1 0 0];
3) to look at the connectivity differences between HC and AD, correcting for potential age and gender differences:
Select: HC, AD, age, gender
Enter contrast: [1 -1 0 0]
4) to look at the connectivity within one group (e.g. HC), correcting for potential age and gender differences across groups:
Select: HC, age_HC_centered, gender_HC_centered
Enter contrast: [1 0 0]
note: you should first create the new variables age_centered_HC and gender_centered_HC (age_centered_HC represents the age values centered -subtracting the mean across all subjects including both groups- and then selecting only the HC subjects; setting the AD subjects to 0)
Last, in all of these analyses, FC is not the predictor but the outcome variable. If, instead, you want to perform regression analyses where FC acts as the predictor (e.g. predicting performance using age, gender, and FC as predictors), you would need to first use 'import values' (right click on the second-level ROI-to-ROI results table) to import your relevant FC measures as second-level covariates, then go to Tools->Calculator, and there:
Select in the 'between-subject effects' list: AllSubjects, age, gender, and one of your new FC measures (e.g. 'conn between ROI01 and ROI02')
Enter the between-subjects contrast: [0 0 0 1]
Select in the 'measures' list: performance
Enter the between-measures contrast: 1
Hope this helps
Alfonso
Originally posted by Yifei Zhang:
I do not see anything out of order in your conn_*.mat file (perhaps only that you seemed to have removed the global signal as part of the preprocessing step, and that you are using an explicit mask for voxel-level analyses, which I assume may be a gray-matter mask, although since you seem to be mostly performing ROI-to-ROI analyses this would not come into play anyway).
Just to make sure everything looks correct I would only check a couple of additional potential issues:
1) Given that you are skipping the standard temporal preprocessing options (confound removal), since you have already-preprocessed BOLD timeseries, the main thing I would check is to make sure that the histogram displays shown in the Preprocessing step look reasonably centered (mean zero) and scaled (not too large variance) for each subject. If that is not the case that would signal that your preprocessing steps have not sufficiently removed potential biases in the BOLD signal and I would consider perhaps adding motion regressors or other potential confounding effects in the preprocessing step.
and 2) Since you are using non-standard ROIs I would also check that all of your functional and ROI files are correctly coregistered (use the corresponding 'check coregistration' buttons to check this)
If the voxel-to-voxel histogram displays in the Preprocessing step look correct, and the functional/ROI files seem to be correctly coregistered, then I do not see anything else to worry about regarding the original Setup/Preprocessing steps.
Regarding your second-level analyses, I believe they are correct but there might be some issues there depending on whether I am interpreting the order of your contrast vectors correctly, so let me rephrase the proper contrast vectors for your different analyses just to make sure (I am listing below the subject effects sorted in the order I believe they will appear in your gui, if they appear in different order you need to resort the contrast vector values accordingly):
1) to compare between the two groups (HC and AD) the association between FC and performance:
Select: HC, AD, performance_HC, performance_HD, age, gender
Enter contrast: [0 0 1 -1 0 0];
2) to compare the association between FC and performance within one group (e.g. HC), but control for age and gender:
Select: HC, performance_HC, age_HC, gender_HC
Enter contrast: [0 1 0 0];
3) to look at the connectivity differences between HC and AD, correcting for potential age and gender differences:
Select: HC, AD, age, gender
Enter contrast: [1 -1 0 0]
4) to look at the connectivity within one group (e.g. HC), correcting for potential age and gender differences across groups:
Select: HC, age_HC_centered, gender_HC_centered
Enter contrast: [1 0 0]
note: you should first create the new variables age_centered_HC and gender_centered_HC (age_centered_HC represents the age values centered -subtracting the mean across all subjects including both groups- and then selecting only the HC subjects; setting the AD subjects to 0)
Last, in all of these analyses, FC is not the predictor but the outcome variable. If, instead, you want to perform regression analyses where FC acts as the predictor (e.g. predicting performance using age, gender, and FC as predictors), you would need to first use 'import values' (right click on the second-level ROI-to-ROI results table) to import your relevant FC measures as second-level covariates, then go to Tools->Calculator, and there:
Select in the 'between-subject effects' list: AllSubjects, age, gender, and one of your new FC measures (e.g. 'conn between ROI01 and ROI02')
Enter the between-subjects contrast: [0 0 0 1]
Select in the 'measures' list: performance
Enter the between-measures contrast: 1
Hope this helps
Alfonso
Originally posted by Yifei Zhang:
Hi Alfonso,
Thank you very much for the meaningful advices before and your excellent work on updating the toolbox and its manual.
I've trying to use the toolbox for some time. I did some tests to see whether I could manage it properly, I used ADNI data to run some ROI-to-ROI resting state functional connectivity analysis with 90 AAL atlas, DMN, Salience and memory networks to see the group difference between HC and AD. I don't think I got results that make sense, either it was hard to get any significant result or the significant connections were much more in AD patients than in healthy controls.
So, I am wondering if I could trouble you to have a look at the .mat file I attached to check if I did something wrong for setting up the toolbox. The data were already preprocessed without the toolbox: corrected for head motion dependent image distortions, normalised to MNI space, smoothed with 5mm FWHM, detrended, filtered(0.08-0.01) and removed white matter, CSF and head motion covariates (without removing global signal).
In addition, I would like to run the regression like this: Performance~functional connectivity+age+gender, here FC is the predictor. Do I properly define the contrasts as follows:
1) to compare between the two groups (HC and AD) the association between FC and performance, select the covariates of performance_HC(values for HC, 0s for AD), performance_AD(values for AD, 0s for HC), HC, AD, age, gender and define the contrast of [0 0 1 -1 0 0].
2) to compare the association between FC and performance within one group(e.g. HC group), but control for age and gender, select the covariates of performance_HC, HC, age_HC, gender_HC (using 0 and 1 to stand for male and female, values for HC and 0s for AD) and define the contrast of [0 1 0 0]. Is there a mistake to define the covariate of gender_HC because both male and AD subjects are defined by 0?
Thank you in advance for any suggestion!
Best regards,
Yifei
Thank you very much for the meaningful advices before and your excellent work on updating the toolbox and its manual.
I've trying to use the toolbox for some time. I did some tests to see whether I could manage it properly, I used ADNI data to run some ROI-to-ROI resting state functional connectivity analysis with 90 AAL atlas, DMN, Salience and memory networks to see the group difference between HC and AD. I don't think I got results that make sense, either it was hard to get any significant result or the significant connections were much more in AD patients than in healthy controls.
So, I am wondering if I could trouble you to have a look at the .mat file I attached to check if I did something wrong for setting up the toolbox. The data were already preprocessed without the toolbox: corrected for head motion dependent image distortions, normalised to MNI space, smoothed with 5mm FWHM, detrended, filtered(0.08-0.01) and removed white matter, CSF and head motion covariates (without removing global signal).
In addition, I would like to run the regression like this: Performance~functional connectivity+age+gender, here FC is the predictor. Do I properly define the contrasts as follows:
1) to compare between the two groups (HC and AD) the association between FC and performance, select the covariates of performance_HC(values for HC, 0s for AD), performance_AD(values for AD, 0s for HC), HC, AD, age, gender and define the contrast of [0 0 1 -1 0 0].
2) to compare the association between FC and performance within one group(e.g. HC group), but control for age and gender, select the covariates of performance_HC, HC, age_HC, gender_HC (using 0 and 1 to stand for male and female, values for HC and 0s for AD) and define the contrast of [0 1 0 0]. Is there a mistake to define the covariate of gender_HC because both male and AD subjects are defined by 0?
Thank you in advance for any suggestion!
Best regards,
Yifei