Dear all,
I am currently working with panel data with two time points - baseline
and follow-up. The problem I have is non-independent multiple outcome
variables.
When analysing the baseline data, I used XTGEE in STATA to carry out
Huber logistic regression, to analyse non-independent multiple
outcomes.
To carry out my analyses the first thing I did was to re-arrange the data,
so that I had 4 records per person, corresponding to the 4 symptom
variables and create a ‘new’ outcome variable (symptom – yes/no –
1/0), indicating the presence or absence of a symptom. There were then
covariates that indicated the symptom type. The remaining covariates
(e.g. gender) remained the same for all 4 records of a particular person.
(i.e. similar to 'repeated measures' structure):
I then fitted a logistic regression model, using STATA XTGEE, as
follows:
logit (p) = b0 + b1*symp2 + b2*symp3 + b3*symp4 + b4*sex +
b5*(symp2*sex) + b6*(symp3*sex) + b7*(symp4*sex)
The outcome is presence of symptom (yes/no), where p is the probability
of having a symptom.
symp2, symp3 and symp4 are the ‘new’ dummy variables for symptom
2, symptom 3 and symptom 4 respectively (symp1 – symptom 1 is the
reference symptom).
The correlation between the observations on each person was taken into
account through the ‘working’ correlation matrix.
The problem I face now is I have the same data structure at follow-up.
That is, I have "repeated repeated" measures.
Is it possible to model the data in such a way that I could account for the
non-independent multiple outcomes (symptoms) at follow-up and also
adjust for the non-independent variables measured at baseline and other
covariates?
Can I repeat the same analysis as above, but using the follow-up non-
independent variables and add the baseline variables as standard
covariates?
I would appreciate any guidance on this.
Many thanks in advance,
Zoe
Zoe Morgan
------------------------
Medical Statistician
University of Leicester
Department of Psychiatry
Brandon Mental Health Unit
Leicester General Hospital
Gwendolen Road, Leicester LE5 4PW, UK.
e-mail: <[log in to unmask]>
Voice: + 44 116 225 6185 / + 44 116 225 6295
Fax: + 44 116 225 6312
|