makeNPDE and VPC plotting error

For problems related to Pmetrics execution, not for modeling or dose optimization issues.
Post Reply
njrhodes
Posts: 12
Joined: Fri Sep 12, 2014 1:03 pm

makeNPDE and VPC plotting error

Post by njrhodes » Fri Oct 23, 2015 8:36 am

Hi all,

I am trying to make an NPDE object using the following command based on the manual:
makeNPDE(34, limits=c(0,3), nsim=1000)
vpc <- plot(sim.34, log=TRUE)

When I use the fitted pop model to make an NPDE, I get the following at the end of the simulation syntax:

Automatic detection of variables is ON. The program will attempt to detect both mandatory variables (ID, X, Y) and optional variables (IPRED, MDV, CENS) when they are not specifically given or when the user-specified names are not found in the dataset, by looking in the names of the columns (to override this behaviour, please use argument detect=FALSE in the call to npdeData().
Using the object called thisobs in this R session as the data.
Can't find a column named mdv in the dataset containing missing data status, will attempt automatic detection.


The following NpdeData object was successfully created:

Object of class NpdeData
longitudinal data
Dataset thisobs
Structured data: out ~ time | id
predictor: time ()
Using the object called thissim in this R session as the data.
Computing the npde for subject 1
Computing the npde for subject 2
Computing the npde for subject 3
Computing the npde for subject 4
Error in chol.default(x) :
the leading minor of order 1 is not positive definite
[[1]]
<simpleError in decorr.chol(varsim): object 'ymat' not found>

attr(,"class")
[1] "PMnpde" "list"

I'm not sure what went wrong but it looks the the object wasn't created. Any advice regarding the sim command and ways to avoid this are appreciated!

Best,

Jim

User avatar
mneely
Site Admin
Posts: 209
Joined: Tue Nov 08, 2011 11:08 am
Contact:

Re: makeNPDE and VPC plotting error

Post by mneely » Mon Oct 26, 2015 1:41 pm

Pmetrics uses the npde package and this is an error arising from that package. I suspect that there is something about the model though. Not very helpful, I'm afraid. Is there something strange about subject 4 or 5?

njrhodes
Posts: 12
Joined: Fri Sep 12, 2014 1:03 pm

Re: makeNPDE and VPC plotting error

Post by njrhodes » Mon Oct 26, 2015 3:24 pm

Not that I can think of, though subj 2 and 3 seem to be outliers in some regards. I wonder if the limits on the pop model are an issue. Do you think I should go back and reset the limits to the exact limits of the model file?

Appreciate the insight!

Thanks!

Jim

njrhodes
Posts: 12
Joined: Fri Sep 12, 2014 1:03 pm

Re: makeNPDE and VPC plotting error

Post by njrhodes » Wed Oct 28, 2015 6:27 am

So, I tried setting the limits to the model file, but the NPDE seems to sill have some trouble parsing the sims. I wonder if this is because some of the predictions from the pop model are negative?
njrhodes wrote:Not that I can think of, though subj 2 and 3 seem to be outliers in some regards. I wonder if the limits on the pop model are an issue. Do you think I should go back and reset the limits to the exact limits of the model file?

Appreciate the insight!

Thanks!

Jim

User avatar
mneely
Site Admin
Posts: 209
Joined: Tue Nov 08, 2011 11:08 am
Contact:

Re: makeNPDE and VPC plotting error

Post by mneely » Mon Nov 02, 2015 5:39 am

Hi Jim,

It is possible that the negative predictions are causing an error. I don't know specifically yet how the npde package handles those. In my experience models which make negative predictions, when they should not be negative have some problems. Do you have many?

Laure
Posts: 7
Joined: Mon Jun 13, 2016 1:49 am

Re: makeNPDE and VPC plotting error

Post by Laure » Tue Dec 06, 2016 2:21 am

Dear gentlemen,

I got the same message when running NPDE


Automatic detection of variables is ON. The program will attempt to detect both mandatory variables (ID, X, Y) and optional variables (IPRED, MDV, CENS) when they are not specifically given or when the user-specified names are not found in the dataset, by looking in the names of the columns (to override this behaviour, please use argument detect=FALSE in the call to npdeData().
Using the object called thisobs in this R session as the data.
Can't find a column named mdv in the dataset containing missing data status, will attempt automatic detection.


Have you solved the problem?

Regards

Laure

njrhodes
Posts: 12
Joined: Fri Sep 12, 2014 1:03 pm

Re: makeNPDE and VPC plotting error

Post by njrhodes » Mon Dec 12, 2016 2:03 pm

I've been able to get NPDE to work well after I updated to the most recent version of Pmetrics and stopped using tdm gfortran.

As Michael notes above, truncating the simulated distribution by setting limits may be one solution.

Code: Select all

makeNPDE(XX, limits=c(0,3), nsim=1000)
where XX is the run of interest and the parameter limits are 0 x the lower limit and 3 x the upper limit of the model file parameter ranges, respectively.

If setting limits still results in an error, would consider looking at the documentation for NPDE and see if you can uncover where the problem is coming from that way: http://www.npde.biostat.fr/userguide_npde2.0.pdf.

Best,

Jim

Post Reply