3 replies to this topic

[jpenzens]

Hi All 

I am modeling a large crossover study with many dosing occasions per subject.  I have IOV on several parameters. There are covariates which change between occassions, e.g. DOSE or ROUTE.    

 

I would like to visualize DOSE vs the IOV, to perhaps screen for covariate effects.  The issue is, EtaCovariate, and the corresponding standard phoenix plot outputs, only have the covariate for the first occasion, so observing COV-ETA for all occasions is not possible.  

 

I could do a work around IF there was a pattern between occasion and the covariate - but it is a random crossover study, so that isn't feasible.  

 

I think your current implementation is the best in general, but doesn't work for my specific use case.  Any idea for a work around? 

Thanks,
Justin 

[smouksassi1]

Dear Justin,

The software does not support it out of the box please file a case with support so they escalate this feature up in the list of features to add.

What you are facing is common to all cases when you have a time varying covariate where the software only keep the first encountered value for reasons not obvious to me.

your best bet is to construct the data you need before being able to plot it for example you need something like this:
 

instead of this

 

ID etaiov1 etaiov2  firstcovariatevalue

1     -0.2         0.4             0

 

 

ID  etaiov  occ    covariatethatchangewithtimeorocc

1         -0.2      1                          0

1         -0.4       2                          1

 

 

you might want to use excel, R or phoenix data manipulation features before being able to do the boxplot or the plot you need.

 

 

 

[Ana Henry]

It is my understanding that one is not able to obtain covariate box plots for a covariate that changes with time within a subject because only ONE eta per subject is indicated in the model.  However, one can add a time-varying covariate to the model and see the effect on model parameter(s).   If you want to visualize the effect then you can request a table of structural parameters and add the time-changing covariate to the 'When Covar' line.  Then you can create boxplots of the structural parameter vs the time-varying covariate (note that you might need to take the log of the structural parameter)

 

The answer above is for a model using FOCE-ELS engine that assumes that eta is the same for a subject over time (e.g. a condition like FED FASTED). 

stparm(ka=tvka*exp(dkadFED1*(FED==1)*exp(nka)

 

If you don't want this, you can modify the model to fit 2 separate etas for each condition for example

fcovariate (FED())

ka=(FED==0? ka1:ka2)

stparm(ka1=tvka1 *exp(nka1))

stparm(ka2=tvka2 *exp(nka2))

[jpenzens]

Regarding your first point 

>It is my understanding that one is not able to obtain covariate box plots for a covariate that changes with time within a subject because only ONE eta per subject is indicated in the model. 

 

I agree that is a proper approach.  

Thank you for your suggestion to make a table - it did the trick.  
 

 

It is my understanding that one is not able to obtain covariate box plots for a covariate that changes with time within a subject because only ONE eta per subject is indicated in the model.  However, one can add a time-varying covariate to the model and see the effect on model parameter(s).   If you want to visualize the effect then you can request a table of structural parameters and add the time-changing covariate to the 'When Covar' line.  Then you can create boxplots of the structural parameter vs the time-varying covariate (note that you might need to take the log of the structural parameter)

 

The answer above is for a model using FOCE-ELS engine that assumes that eta is the same for a subject over time (e.g. a condition like FED FASTED). 

stparm(ka=tvka*exp(dkadFED1*(FED==1)*exp(nka)

 

If you don't want this, you can modify the model to fit 2 separate etas for each condition for example

fcovariate (FED())

ka=(FED==0? ka1:ka2)

stparm(ka1=tvka1 *exp(nka1))

stparm(ka2=tvka2 *exp(nka2))