[Freesurfer] lme model for within-subjects repeated measures design

[martina.papmeyer]

Dear FreeSurfer experts



I have one question regarding my data analysis and would be extremely thankful 
for any advice!



My data-set is as follows: I have repeated measures (time point 0 (T0), time 
point 1 (T1)) of several subjects. All individuals underwent an intervention at 
one of the time points and a placebo condition at the other time point in a 
fully randomized fashion. Thus, half of the subjects received treatment at T0 
and half of them at T1. I am interested in the putative effect of the 
intervention on cortical thickness in a ROI. A major challenge is that the time 
between T0 and T1 varies between individuals and that I expect the time to 
impact on my dependent variable and to likely interact with the condition 
(treatment versus placebo).



I thought about conducting a simple repeated-measures ANOVA. However, as 
stated, I want to take the time between the two sessions into account. I also 
thought about an analysis of rates or percent changes. However, this approach 
does not model the correlation among the repeated measures and is thus 
associated with a reduction in power.



Accordingly, I am trying to use lme models to analyse my data. Since I have no 
between-group variable but a within-subjects design, I am concerned if my 
thoughts are correct and would be grateful for feedback.



I ran the longitudinal FS stream and followed the longitudinal lme model 
tutorial. I propose the following lme model with one random factor: thickness = 
intercept (random factor) + time since baseline + ICV + condition (placebo or 
treatment) + timeXcondition + Age (does not change across time interval) + 
gender



The analysis finishes with 0% non-covergence. Can you tell me if my model is 
suitable given the fact that it is a within-subjects design? I also started 
wondering if it was possible to model time as a random factor but I think that 
I read that this is not suitable if you only have two groups (in my case: 
conditions).



Thank you very much for help and advice!



All best wishes, Martina









Universitäre Psychiatrische Dienste Bern (UPD)
Universitätsklinik für Psychiatrie und Psychotherapie
Systemische Neurowissenschaften der Psychopathologie
Zentrum für Translationale Forschung
Dr. phil. Martina Papmeyer, Wissenschaftliche Mitarbeiterin
Bolligenstrasse 111, CH-3000 Bern 60
Tel: ++41 0(31) 930 9599, Fax: ++41 0(31) 930 9961
Mail: martina.papme...@puk.unibe.ch
www.puk.unibe.ch




___
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.

Re: [Freesurfer] lme model for within-subjects repeated measures design

[Martin Reuter]

Hi Martina,

so you don't have a baseline (no treatment) measurement? If you have a 
treatment at T0, you mean during an interval before T0, right? But since 
you did not scan before that treatment, you cannot quantify that change? 
The design is not clear to me.


About the random effect (with only two time points and two groups) I 
think having the intercept is enough.


Best, Martin


On 04/18/2016 11:51 AM, martina.papme...@puk.unibe.ch wrote:


Dear FreeSurfer experts

I have one question regarding my data analysis and would be extremely 
thankful for any advice!


My data-set is as follows: I have repeated measures (time point 0 
(T0), time point 1 (T1)) of several subjects. All individuals 
underwent an intervention at one of the time points and a placebo 
condition at the other time point in a fully randomized fashion. Thus, 
half of the subjects received treatment at T0 and half of them at T1. 
I am interested in the putative effect of the intervention on cortical 
thickness in a ROI. A major challenge is that the time between T0 and 
T1 varies between individuals and that I expect the time to impact on 
my dependent variable and to likely interact with the condition 
(treatment versus placebo).


I thought about conducting a simple repeated-measures ANOVA. However, 
as stated, I want to take the time between the two sessions into 
account. I also thought about an analysis of rates or percent changes. 
However, this approach does not model the correlation among the 
repeated measures and is thus associated with a reduction in power.


Accordingly, I am trying to use lme models to analyse my data. Since I 
have no between-group variable but a within-subjects design, I am 
concerned if my thoughts are correct and would be grateful for feedback.


I ran the longitudinal FS stream and followed the longitudinal lme 
model tutorial. I propose the following lme model with one random 
factor: thickness = intercept (random factor) + time since baseline + 
ICV + condition (placebo or treatment) + timeXcondition + Age (does 
not change across time interval) + gender


The analysis finishes with 0% non-covergence. Can you tell me if my 
model is suitable given the fact that it is a within-subjects design? 
I also started wondering if it was possible to model time as a random 
factor but I think that I read that this is not suitable if you only 
have two groups (in my case: conditions).


Thank you very much for help and advice!

All best wishes, Martina

Universitäre Psychiatrische Dienste Bern (UPD)
*Universitätsklinik für Psychiatrie und Psychotherapie*
Systemische Neurowissenschaften der Psychopathologie
Zentrum für Translationale Forschung
Dr. phil. Martina Papmeyer, Wissenschaftliche Mitarbeiterin
Bolligenstrasse 111, CH-3000 Bern 60
Tel: ++41 0(31) 930 9599, Fax: ++41 0(31) 930 9961
Mail: martina.papme...@puk.unibe.ch
www.puk.unibe.ch



___
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


--
Martin Reuter, PhD
Assistant Professor of Radiology, Harvard Medical School
Assistant Professor of Neurology, Harvard Medical School
A.A.Martinos Center for Biomedical Imaging
Massachusetts General Hospital
Research Affiliate, CSAIL, MIT
Phone: +1-617-724-5652
Web  : http://reuter.mit.edu

___
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.

Re: [Freesurfer] lme model for within-subjects repeated measures design

[Martina Papmeyer]

Hi Martin 

Thank you very much for your response!  To clarify the design: There are 44 
subjects, all have been scanned twice and thus have repeated measures of 
cortical thickness. 22 subjects were first (T0) scanned 2 hours after a placebo 
treatment. Some days later, the identical subjects were scanned again (T1) but 
this time 2 hours after a "real" treatment. The other 22 subjects were scanned 
first (T0) after "real" treatment and then some days later after placebo 
treatment. The interval between T0 and T1 varies between subjects which I would 
like to take into account into my analyses. The time between receiving 
placebo/real treatment and MRI acquisition is identical among all subjects 
(2hours) and thus not of concern. 

Thank you very much for your help! Best, Martina 

Sent from my iPad

> On 21 Apr 2016, at 22:09, Martin Reuter  wrote:
> 
> Hi Martina,
> 
> so you don't have a baseline (no treatment) measurement? If you have a 
> treatment at T0, you mean during an interval before T0, right? But since you 
> did not scan before that treatment, you cannot quantify that change? The 
> design is not clear to me.
> 
> About the random effect (with only two time points and two groups) I think 
> having the intercept is enough.
> 
> Best, Martin
> 
> 
>> On 04/18/2016 11:51 AM, martina.papme...@puk.unibe.ch wrote:
>> Dear FreeSurfer experts
>>  
>> I have one question regarding my data analysis and would be extremely 
>> thankful for any advice!
>>  
>> My data-set is as follows: I have repeated measures (time point 0 (T0), time 
>> point 1 (T1)) of several subjects. All individuals underwent an intervention 
>> at one of the time points and a placebo condition at the other time point in 
>> a fully randomized fashion. Thus, half of the subjects received treatment at 
>> T0 and half of them at T1. I am interested in the putative effect of the 
>> intervention on cortical thickness in a ROI. A major challenge is that the 
>> time between T0 and T1 varies between individuals and that I expect the time 
>> to impact on my dependent variable and to likely interact with the condition 
>> (treatment versus placebo).
>>  
>> I thought about conducting a simple repeated-measures ANOVA. However, as 
>> stated, I want to take the time between the two sessions into account. I 
>> also thought about an analysis of rates or percent changes. However, this 
>> approach does not model the correlation among the repeated measures and is 
>> thus associated with a reduction in power.
>>  
>> Accordingly, I am trying to use lme models to analyse my data. Since I have 
>> no between-group variable but a within-subjects design, I am concerned if my 
>> thoughts are correct and would be grateful for feedback.
>>  
>> I ran the longitudinal FS stream and followed the longitudinal lme model 
>> tutorial. I propose the following lme model with one random factor: 
>> thickness = intercept (random factor) + time since baseline + ICV + 
>> condition (placebo or treatment) + timeXcondition + Age (does not change 
>> across time interval) + gender
>>  
>> The analysis finishes with 0% non-covergence. Can you tell me if my model is 
>> suitable given the fact that it is a within-subjects design? I also started 
>> wondering if it was possible to model time as a random factor but I think 
>> that I read that this is not suitable if you only have two groups (in my 
>> case: conditions).
>>  
>> Thank you very much for help and advice!
>>  
>> All best wishes, Martina
>>  
>>  
>>  
>>  
>> Universitäre Psychiatrische Dienste Bern (UPD)
>> Universitätsklinik für Psychiatrie und Psychotherapie
>> Systemische Neurowissenschaften der Psychopathologie
>> Zentrum für Translationale Forschung
>> Dr. phil. Martina Papmeyer, Wissenschaftliche Mitarbeiterin
>> Bolligenstrasse 111, CH-3000 Bern 60
>> Tel: ++41 0(31) 930 9599, Fax: ++41 0(31) 930 9961
>> Mail: martina.papme...@puk.unibe.ch 
>> www.puk.unibe.ch
>>  
>>  
>> 
>> 
>> ___
>> Freesurfer mailing list
>> Freesurfer@nmr.mgh.harvard.edu
>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
> 
> -- 
> Martin Reuter, PhD
> Assistant Professor of Radiology, Harvard Medical School
> Assistant Professor of Neurology, Harvard Medical School
> A.A.Martinos Center for Biomedical Imaging
> Massachusetts General Hospital
> Research Affiliate, CSAIL, MIT
> Phone: +1-617-724-5652
> Web  : http://reuter.mit.edu 
> ___
> 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 inf

Re: [Freesurfer] lme model for within-subjects repeated measures design

[Martin Reuter]

Thanks,

what I mean is in order to quantify a treatment effect for the subjects 
that received treatment at T0, you would need another scan before that 
(Tminus1). You would need three time points:


group1: baseline, placebo, treatment
group2: baseline, treatment, placebo

you could then test if treatment or placebo differs from baseline.

If you assume that the ordering is not important, you could use a binary 
variable (drug / placebo) instead of time in LME and control for the 
time delta as a co-variate. This may make sense, as you are not really 
interested in change / time unit.


Another model would be to use time as usual in LME and add drug as a 
time varying covariate. I am not a statistician, so I don't really know 
what would be best in your case.


Cheers, Martin



On 04/21/2016 06:08 PM, Martina Papmeyer wrote:

Hi Martin

Thank you very much for your response!  To clarify the design: There 
are 44 subjects, all have been scanned twice and thus have repeated 
measures of cortical thickness. 22 subjects were first (T0) scanned 2 
hours after a placebo treatment. Some days later, the identical 
subjects were scanned again (T1) but this time 2 hours after a "real" 
treatment. The other 22 subjects were scanned first (T0) after "real" 
treatment and then some days later after placebo treatment. The 
interval between T0 and T1 varies between subjects which I would like 
to take into account into my analyses. The time between receiving 
placebo/real treatment and MRI acquisition is identical among all 
subjects (2hours) and thus not of concern.


Thank you very much for your help! Best, Martina

Sent from my iPad

On 21 Apr 2016, at 22:09, Martin Reuter > wrote:



Hi Martina,

so you don't have a baseline (no treatment) measurement? If you have 
a treatment at T0, you mean during an interval before T0, right? But 
since you did not scan before that treatment, you cannot quantify 
that change? The design is not clear to me.


About the random effect (with only two time points and two groups) I 
think having the intercept is enough.


Best, Martin


On 04/18/2016 11:51 AM, martina.papme...@puk.unibe.ch wrote:


Dear FreeSurfer experts

I have one question regarding my data analysis and would be 
extremely thankful for any advice!


My data-set is as follows: I have repeated measures (time point 0 
(T0), time point 1 (T1)) of several subjects. All individuals 
underwent an intervention at one of the time points and a placebo 
condition at the other time point in a fully randomized fashion. 
Thus, half of the subjects received treatment at T0 and half of them 
at T1. I am interested in the putative effect of the intervention on 
cortical thickness in a ROI. A major challenge is that the time 
between T0 and T1 varies between individuals and that I expect the 
time to impact on my dependent variable and to likely interact with 
the condition (treatment versus placebo).


I thought about conducting a simple repeated-measures ANOVA. 
However, as stated, I want to take the time between the two sessions 
into account. I also thought about an analysis of rates or percent 
changes. However, this approach does not model the correlation among 
the repeated measures and is thus associated with a reduction in power.


Accordingly, I am trying to use lme models to analyse my data. Since 
I have no between-group variable but a within-subjects design, I am 
concerned if my thoughts are correct and would be grateful for feedback.


I ran the longitudinal FS stream and followed the longitudinal lme 
model tutorial. I propose the following lme model with one random 
factor: thickness = intercept (random factor) + time since baseline 
+ ICV + condition (placebo or treatment) + timeXcondition + Age 
(does not change across time interval) + gender


The analysis finishes with 0% non-covergence. Can you tell me if my 
model is suitable given the fact that it is a within-subjects 
design? I also started wondering if it was possible to model time as 
a random factor but I think that I read that this is not suitable if 
you only have two groups (in my case: conditions).


Thank you very much for help and advice!

All best wishes, Martina

Universitäre Psychiatrische Dienste Bern (UPD)
*Universitätsklinik für Psychiatrie und Psychotherapie*
Systemische Neurowissenschaften der Psychopathologie
Zentrum für Translationale Forschung
Dr. phil. Martina Papmeyer, Wissenschaftliche Mitarbeiterin
Bolligenstrasse 111, CH-3000 Bern 60
Tel: ++41 0(31) 930 9599, Fax: ++41 0(31) 930 9961
Mail: martina.papme...@puk.unibe.ch
www.puk.unibe.ch



___
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


--
Martin Reuter, PhD
Assistant Professor of Radiology, Harvard Medical School
Assistant Professor of Neurology, Harvard Medical School
A.A.Martinos Center for Biomedical Imaging
Massachusetts Gener