In sequence analysis, my confounders should be measured at the beginning, at the end or during the length of the study time of the sequences? I'm working on a study for analysis patterns of treatment choices via sequence analysis (discrepancy). I have sociodemographic variables (age groups, sex, etc) that I have measured from baseline but have more clinical variables that I would like to introduce as covariates. I understand that the measuring points for covariates (baseline, end of the sequence, other) will be answering different questions.
I would like to know if it is methodologically correct to include a measure of "clinical progression" (i.e. value of endpoint - the value of start point) as a way to measure the worsening of the disease concerning the treatment sequences. My clinical variables are continuous but I can convert them to categorical as they have thresholds (e.g, impaired vs not impaired).
Would this be a correct approach to study the relationship between covariates and the sequences?
Thank you!
 A: Thank you for your comment, Gilbert.
Yes, I would like to evaluate how time-varying covariates (pre-post clinical tests) can explain differences among sequences of a clinical sample.
By sequences, I mean trajectories (longitudinal data) composed of different states ordered chronologically as in a life event (in my study, a state would be a type of medication). Thus, during one sequence, an individual can have several medications during a determined time period. See the picture below as an example. 
In social studies, one normally introduces gender or educational level as covariates for evaluating differences among educational/career sequences (how x factor explains differences among x trajectories). However, I would like to use a time-varying covariate to explain a clinical trajectory, if that makes sense. For example, how much change in physical disability (measured by pre-post clinical test) explains the differences in the medication choices taken throughout a specific period of the disease.
My question was more methodologically oriented, as i am not sure the approach I want to do is valid.
Any comments or suggestions would be of great help as I'm quite a newbie here...
Thank you!!
/Ale
