I appreciate your providing the pros and cons associated with both options.
Lets say for now we want to pursue option 2.
My understating is that since you recommend not to include ‘visit’ and 'time_from_baseline’ in the same model, one of the steps in creating MLE model for option 2 would be to not include ‘time_from_baseline’ as a random variable.
As for any other considerations to be kept in mind while creating the LME model for option 2, kindly provide your thoughts on the matter.
Mayank
On Apr 10, 2018, at 8:56 AM, Diers, Kersten /DZNE Kersten.Diers@dzne.de wrote:
ATTENTION: This email originated from a sender outside of MCW. Use caution when clicking on links or opening attachments. ________________________________
On Mo, 2018-04-09 at 19:53 +0200, Kaushal, Mayank wrote:
I am indeed considering LME model since I have multiple visits (4) for each subject and multiple groups (3).
Based on your response, I am inclined to include the following.
Intercept - random variable Slope (time_base_scan) - random variable
As I am including time_base_scan as the random variable, this should take care of the timing information. Am I correct in making this assumption?
Moving on to the categorical variables - group and visit. I want to evaluate the following effects
- group
- visit
- group and visit
My concern here is how do I design the contrast matrix for this?
Let's consider the model first:
I would not recommend to include both 'visit' and 'time_from_baseline' in the same model. This is because these two variables should be highly correlated, and such a redundancy may make the estimation and interpretation of the model difficult. Therefore it is better to include just one of them, but not both.
So the basic decision would be whether to treat time 1) as continuous (then use 'time_from_baseline') or 2) as categorical (then use 'visit'):
In my eyes, there are advantages and disadvantages to both, see below. Regardless of that, in both cases it will be possible to assess the effects of group, time/visit, and their combination (e.g. different slopes between groups, or presence of group differences only at some visits but not all).
Option 1)
Using 'time_from_baseline', and hence treating time as continuous, gives the LME model that we have been discussing so far. So there is no need to create a new model, and as far as I can see, your research questions can already be answered with the current model.
Although it will not be possible to directly contrast, say, visit 1 and 2 with this approach, you will still be able to make statements like 'Per 1 year (or 2 years, or 1 month, or whatever unit of time, depending on your observation period), we expect changes of magnitude X (or 2*X, or X/12, ...) , according to this model', and this may even be more accurate than expressing change in units of visits. Note, though, that these changes will be modeled as linear changes across time (unless you include additional e.g. quadratic terms in the model).
Option 2)
Using 'visit', and hence treating time as categorical, would result in a model like a repeated-measures ANOVA. The main advantages, in my eyes, would be that you could explicitly compare between, say, visit 1 and 2, or 1 and 4, etc. So, the categorical model may be a little bit easier to interpret, which is possibly another advantage.
A disadvantage of this option is that you'd need to specify a new model. As far as I can see, the LME framework can also cover the categorical scenario, but I'd need to check how. Otherwise, one could use a classical repeated-measures ANOVA (https://urldefense.proofpoint.com/v2/url?u=https-3A__surfer.nmr.mgh.harvard&... .edu/fswiki/RepeatedMeasuresAnova). Another disadvantage is that the categorical approach discards information about the precise timing, and that it makes the assumption that the visits occurred at comparable time intervals across subjects.
I thought it would be good to lay out the two main modeling options, as they appear to me now, before proceeding.
Best,
Kersten
I am aware that I have asked more than my fair share of questions and very grateful to you for being exceedingly patient in answering them.
Mayank
On Apr 9, 2018, at 12:15 PM, Diers, Kersten /DZNE <Kersten.Diers@dz ne.de> wrote:
ATTENTION: This email originated from a sender outside of MCW. Use caution when clicking on links or opening attachments. ________________________________
Hello,
this is somewhat difficult to answer, so the following is a personal opinion.
I think the first question is whether your would like to go for a) a cross-sectional design or b) a longitudinal design.
If a), you'd have a single scan per subject (e.g., baseline) and could discard the temporal information. So if you do not have a longitudinal research question and your goal is to model group differences at a single time-point, it is not necessary (and overly complicated, actually) to use a LME model.
If b), you'd have multiple scans per subject, and would therefore use the LME model. I would not recommend to exclude the timing information from the model / design matrix in this case, no matter how close the temporal spacing may be. It is still possible to test for effects other than time, e.g. for group differences at baseline.
Best regards,
Kersten
On Mi, 2018-04-04 at 23:37 +0200, Kaushal, Mayank wrote:
The continuous variable in my analysis is time_base_scan with the time points evaluated by me being of acute and subacute nature. However, due to the small differences in time elapsed between successive visits (duration between visits is a couple of weeks), is it possible to simply model LME analysis around categorical variables without including continuous variables?
More specifically, I want to see the effect of group. Further, I don’t intend to include time_base_scan as a continuous variable. What would the contrasts used by me to evaluate a categorial variable?
Further, my analysis includes three groups. So, my understanding is that evaluating group as a categorical variable would’ve no bearing on the column composition of X and they would remain the same as mentioned in my previous mail.
The columns in X are as follows: Column 1: intercept Column 2: time_base_scan (This signifies the time elapsed from the base scan in days. Each subject had upto 4 visits with scans undertaken on each visit. For the scan taken on the first visit 0 was entered in the qdec.dat.table file) Column 3: age1stscan (This is the age of the subject at first scan) Column 4: group 1 (Value of 1 entered in this column if the subject belonged to group 1 and value of 0 entered in column 5 that signifies group 2) Column 5: group 2 (Value of 1 entered in this column if the subject belonged to group 2 and value of 0 entered in column 4 that signifies group 1)
Mayank On Mar 30, 2018, at 6:30 AM, Diers, Kersten /DZNE <Kersten.Diers@ dzne .demailto:Kersten.Diers@dzne.de> wrote:
ATTENTION: This email originated from a sender outside of MCW. Use caution when clicking on links or opening attachments. ________________________________ Hello,
please find my responses below.
Best regards,
Kersten
On Do, 2018-03-29 at 00:11 +0200, Kaushal, Mayank wrote:
Hi Kersten,
Apologies for the delay. I am still in the process of trying to figure out the gaps in my understanding and would appreciate your inputs.
I created the design X from M using: X = [ones(length(M),1) M M(:,1).*M(:,3) M(:,1).*M(:,4)];
The columns in X are as follows: Column 1: intercept Column 2: time_base_scan (This signifies the time elapsed from the base scan in days. Each subject had upto 4 visits with scans undertaken on each visit. For the scan taken on the first visit 0 was entered in the qdec.dat.table file) Column 3: age1stscan (This is the age of the subject at first scan) Column 4: group 1 (Value of 1 entered in this column if the subject belonged to group 1 and value of 0 entered in column 5 that signifies group 2) Column 5: group 2 (Value of 1 entered in this column if the subject belonged to group 2 and value of 0 entered in column 4 that signifies group 1) Column 6: Column 2 (time_base_scan) x Column 4 (group 1) Column 7: Column 2 (time_base_scan) x Column 4 (group 2)
So my understanding is that subjects belonging to group 3 have 0 in both columns 4 and 5 and consequently, columns 6 and 7 would also be 0 for them. This would be translated by matlab lme toolbox while doing spatiotemporal analysis. Is my understanding correct?
Yes, this is correct, apart probably from a little typo:
Column 7 should read: " Column 2 (time_base_scan) x Column 5 (group 2)" (group2 is column 5, not 4, in X; please adapt your design matrix if necessary)
Group 3 will be your reference group then, and will implicitly modeled by this design matrix. I speculate this is what you mean by 'translated'.
I have attached X as well as M as separate excel files for your consideration. In addition, I have attached qdec.dat.table file as well a word doc labeled “workflow” that detail the order of matlab functions I have used to perform the analysis.
My objective is to highlight clusters that are significantly different between the groups over time using spatiotemporal analysis.
The contrasts used by me for the analysis: CM.C = [0 0 0 0 0 1 0; 0 0 0 0 0 -1 1]
Kindly comment on my choice of contrasts. Is this the correct choice for contrasts if I want to find significant clusters based on cortical thickness between all three groups over time?
Yes, the contrast is also correct for your purpose. It will identify regions in which cortical thickness changes across time differ between groups.
Mayank
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto:Freesurfer@nmr.mgh.harvard. edu https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.nmr.mgh .har vard.edu_mailman_listinfo_freesurfer&d=DwIBAg&c=aFamLAsxMIDYjNglY HTMV 0iqFn3z4pVFYPQkjgspw4Y&r=RqvEwdmcEg_aWsE7PBL80w&m=jz6TYzDc4SmT6ar 1_JE w5rF4JCOlAxM5KmH1GK_SDNc&s=9INo1YCHATevCPV_UsV08YkkiXAOEcUSw7nvJL vyBa g&e=
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 https://urldefense.proofpoint.com/v2/url?u=http-3A__www.partners. org_ complianceline&d=DwIBAg&c=aFamLAsxMIDYjNglYHTMV0iqFn3z4pVFYPQkjgs pw4Y &r=RqvEwdmcEg_aWsE7PBL80w&m=jz6TYzDc4SmT6ar1_JEw5rF4JCOlAxM5KmH1G K_SD Nc&s=guYHSBVwZaPkOne0blcYjP_gI5nw3ZybG4nnnHojJKk&e= . 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.
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.nmr.mgh .harvard.edu_mailman_listinfo_freesurfer&d=DwIGaQ&c=aFamLAsxMIDYj NglYHTMV0iqFn3z4pVFYPQkjgspw4Y&r=RqvEwdmcEg_aWsE7PBL80w&m=oogcc8F CLuvkkltLtZP70UyiCeWTP_CLDrmHvxx_yj4&s=0jhKXk4zw3- V7Oms3DoEZzo8oSM9ney-DW5Wj3QtvDI&e=
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 https://urldefense.proofpoint.com/v2/url?u=http-3A__www.partners. org_complianceline&d=DwIGaQ&c=aFamLAsxMIDYjNglYHTMV0iqFn3z4pVFYPQ kjgspw4Y&r=RqvEwdmcEg_aWsE7PBL80w&m=oogcc8FCLuvkkltLtZP70UyiCeWTP _CLDrmHvxx_yj4&s=I3HTqixlvSUyAOJrD5diLImtXBNte1CuOp0YWct5roU&e= . 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.
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.nmr.mgh.h arvard.edu_mailman_listinfo_freesurfer&d=DwIGaQ&c=aFamLAsxMIDYjNglY HTMV0iqFn3z4pVFYPQkjgspw4Y&r=RqvEwdmcEg_aWsE7PBL80w&m=oogcc8FCLuvkk ltLtZP70UyiCeWTP_CLDrmHvxx_yj4&s=0jhKXk4zw3-V7Oms3DoEZzo8oSM9ney- DW5Wj3QtvDI&e=
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.nmr.mgh.harvard.ed...
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 https://urldefense.proofpoint.com/v2/url?u=http-3A__www.partners.org_complia... . 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.
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.nmr.mgh.harvard.ed...