From Thu Aug 15 08:55:11 1996
Subject: Sample a Weighted combination of 2 compartments

In NONMEM, I can see how to specify which compartment corresponds to a particular sample but I cannot see how to specify that a sample is a weighted combination of two compartments. For example, suppose that my sample comes half from the central compartment and have from the peripheral compartment. Is there a section of thier users guide that discusses this ?


From alison Thu Aug 15 09:53:58 1996
Subject: Re: sample a weighted combination of 2 compartments

This raises an interesting question about modelling possiblities in PREDPP and NM-TRAN.

In Guide IV, we say that the $ERROR block may include on the right reserved variables A(i):

The array elements A(1), ..., A(10) symbolize special right-hand quantities, the amounts in compartments 1 through 10 (including the output compartment).

This gives you the opportunity to model Y in terms of more than one compartment simultaneously; e.g.,

Y = some function of A(i) and A(j).

You could include code along the following lines in the $ERROR block.

1) Fully general model

Suppose compartment 1 is the central compartment and compartment 2 is the peripheral comparment, and that there are observations of each compartment separately as well as combined. Define TYPE as follows:

TYPE=1 when the sample (by which you mean the observed DV, I suppose) is from compartment 1
TYPE=2 when the sample is from compartment 2
TYPE=3 when the sample is the mean of the two compartments.

$INPUT .... TYPE ....
C1 = A(1) / S1 ; concentration in compt 1
C2 = A(2) / S2 ; concentration in compt 2
C3 = (C1 + C2)/2 ; mean concentration
Y1= C1 + ERR(1)
Y2= C2 + ERR(2)
Y3= C3 + ERR(3)
IF (TYPE.EQ.1) Q1=1
IF (TYPE.EQ.2) Q2=1
IF (TYPE.EQ.3) Q3=1
Y= Q1*Y1 + Q2*Y2 + Q3*Y3

Note that the value of CMT is irrelevant, because I don't allow PREDPP use the default compartment for observations; instead, I control this by means of TYPE.

Note also that amounts A(1) and A(2) might be what you want in the model for Y3; i.e., you might prefer to deal with amounts.

Also, I don't know what the right way is to model the ERR terms. Clearly, ERR(1) and ERR(3) are highly correlated, as are ERR(2) and ERR(3). You'd need full covariance: e.g., if the data are population,


2) Less general model

All this can be done more simply if *ALL* observations are the same combination of the concentrations in compartments 1 and 2. Let CMT = 1 be the default compartment for observations, so that F is A(1)/S1 by default. Then TYPE is not needed:

C2= A(2)/S2
Y= (F+C2)/2 + ERR(1)