Mixed Model Analysis with covariates

[Chris Hunt]

Hi,

 

I have a randomised parallel group trial.  Subjects were randomly selected for one of three treatment groups.  They were seen for Baseline, 1-week and 2-week visits.   We would like to look for differences between both visits and groups.  The dependent variable is continuous and follows a close to normal distribution.  The data was analysed using a mixed model analysis with visit, group, and visit*group interaction as fixed factors and subject as a random factor. 

 

However, it was found that there were differences between the groups at baseline.  I have been asked to repeat the analyses using differences from baseline, i.e. using a model as above (except visit would only have two levels not three). 

 

I was thinking that a better way would be to add the baseline score as a covariate in the mixed model analysis.

 

Could anybody help me with which method to follow and why or point me to a reference to help?

 

Regards,

Chris

[Peter Flom]

Chris Hunt wrote

 

I think you are right.  I think this is really the point of mixed models.

 

See, e.g., Hedeker and Gibbons, Longitudinal Data Analysis

 

HTH

 

Peter

 

[Munyaradzi Dimairo]

Hi Chris

The conventional mixed model in clinical trial setting assumes negligible baseline imbalance (randomisation being effective). in this way the estimated effects with be in the order (0 (baseline), xx (time 1) xxx (time 2)) or (0 (baseline), xx (time averaged)). In case of baseline imbalances, you just need to extend the ANCOVA model in longitudinal framework by including baseline as a covariate. reference by Twisk explains this in brief (Aplied longitudinal data in epidemiology).

You actually brought forward an issue that i came across recently!!!!


Best wishes

Munya

--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .
MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules

[mdim...@gmail.com]

Sorry was a typo there; estimated effects will instead of with! :-((

Sent using BlackBerry® from Orange

---

From: Munyaradzi Dimairo <mdim...@gmail.com>

Date: Wed, 10 Nov 2010 16:46:18 +0000

To: <meds...@googlegroups.com>

Subject: Re: {MEDSTATS} Mixed Model Analysis with covariates

[Chris Hunt]

Peter and Munya - Thank you very much for your replies.