Missing covariate values

Post technical questions about making Pmetrics carry out a particular task.
Post Reply
pharmpk
Posts: 41
Joined: Fri Dec 05, 2014 1:35 pm

Missing covariate values

Post by pharmpk » Fri May 17, 2019 9:22 am

I'm trying to use covariates as a way of including drug interactions on the primary drug parameters. However, not all subjects receive the same potentially interfering drugs. Thus, there are many missing entries for the secondary drug exposure. Entering 0.0 allows the analysis to proceed as does -99. Entering a . doesn't work. 0.0 or -99.0 being included as values doesn't really help. Any suggestions.

My plan for now is to leave out these secondary drug covariates and look at them after the Pmetrics analysis.

Thanks, David

User avatar
mneely
Site Admin
Posts: 233
Joined: Tue Nov 08, 2011 11:08 am
Contact:

Re: Missing covariate values

Post by mneely » Fri May 17, 2019 10:29 am

You could do something like INT = 0 or 1 as a covariate for receipt of an interacting drug and then have your model do this:

#PRI
Ke1,0,5
Ke2,0,5
...

#COV
INT

#SEC
Ke = Ke1*(1-INT) + Ke2*INT

This will set Ke = Ke1 when INT=0 (no interacting drug) and Ke = Ke2 when INT=1 (interacting drug present).

Does that help?

User avatar
mneely
Site Admin
Posts: 233
Joined: Tue Nov 08, 2011 11:08 am
Contact:

Re: Missing covariate values

Post by mneely » Fri May 17, 2019 10:35 am

If you have multiple interacting drugs possible and you want to model all of them, then set a covariate to 0 or 1 for each drug and use an exponential model.

#PRI
Ke0,0,5
mod1,-1,1
mod2,-1,1
mod3,-1,1
mod4,-1,1
...

#COV
INT1
INT2
INT3
INT4
...

#SEC
Ke = Ke0 * exp(mod1*INT1 + mod2*INT2 + mod3*INT3 + mod4*INT4)

Of course, if you know a drug is an inhibitor, then you can have more informed ranges on the mod (modify) variable for that drug, like [-1,0] or [-2,0] so that if present, it can only reduce Ke. Likewise, for a known inducer, the range could be [0,1] or some other positive range so that it could only increase Ke if present.

pharmpk
Posts: 41
Joined: Fri Dec 05, 2014 1:35 pm

Re: Missing covariate values

Post by pharmpk » Mon May 20, 2019 10:17 am

Thanks for the suggestions. However, I already have two metabolites in the model and 10-12 covariates with only 1- 4 entries per subject. I think after considering wt, age, type covariates for inclusion in the model I'll look at the DDI 'covariates' post Pmetric analysis.

User avatar
mneely
Site Admin
Posts: 233
Joined: Tue Nov 08, 2011 11:08 am
Contact:

Re: Missing covariate values

Post by mneely » Wed May 22, 2019 9:11 am

PMstep() will do a linear regression backwards elimination (could do forward/backward) of all covariates vs. all parameters and report the P values for regression coefficients based on retention in the linear model. It uses AIC to discard or retain covariates. PMstep is based on the R "step" function. You can look at the help on either for more info.

You can also plot any covariate against any other covariate or parameter. See ?plot.PMcov for help. This might suggest a non-linear regression to try, or for binary covariates, will generate a box plot.

Post Reply