Comparing models

Post technical questions about making Pmetrics carry out a particular task.
Post Reply
SONIALU
Posts: 10
Joined: Fri Mar 04, 2016 4:26 am

Comparing models

Post by SONIALU » Mon Jan 14, 2019 3:16 am

Dear Michael,

I have a doubt when I try to compare different models.
Can I use the PMcompare when the different models have different pk parameters, for example, when I try to compare a 1 and a 2 -compartment model?

Thank you very much

Sonia

pharmpk
Posts: 34
Joined: Fri Dec 05, 2014 1:35 pm

Re: Comparing models

Post by pharmpk » Fri Feb 08, 2019 1:01 pm

Hi Sonia

Yes, use PMcompare(x,y) where x is the run number for one model and y is the run number for the second model

For example:

PMcompare(19,20,plot=T)
run type nsub nvar par converge -2*LL aic bic popBias popImp popPer_RMSE postBias
1 19 NPAG 35 3 Ke V Ka TRUE 845.0 853.2 865.2 0.8446 14.110 137.6 -0.1934
2 20 NPAG 30 3 Ke V Ka TRUE 802.1 810.4 822.0 -0.1245 8.373 106.5 -0.1587
postImp postPer_RMSE pval
1 0.8962 58.56 NA
2 0.8919 51.17 0.555

With plot=T you get Model plot

Type ?PMcompare in the console for more info.

SONIALU
Posts: 10
Joined: Fri Mar 04, 2016 4:26 am

Re: Comparing models

Post by SONIALU » Mon Feb 18, 2019 9:57 am

Hello!!
Thank you very much!

My doubts are:
- PMcompare can only be used for the comparison of models that have the same number of pk parameters?
- In some PMcompare results, I can see that the -2LL and AIC are lower in one model but the postbias and the postimp are worse. Which is
the variable that is most important in order to select the best model?
-Which is the meaning of the p value?

Thank you very much,

Sonia

Post Reply