There is one more point you might want to mention about modelled rates
in PREDPP. You can send this to nmusers, if you think it is relevant.
The rate R1 computed with a given event record applies to the *preceding*
time period.
Suppose the subject is on drug from time 0 to time 8, and off drug
from time 8 to time 24. The ONDR value must be specified as follows.
(Note that ONDR value when TIME=0 is irrelevant; either 0 or 1 could
be coded.)
First, suppose observations were made at time 8 and 24 (x1 and x2):
#ID TIME AMT RATE ONDR X=DV
1 0 9999 -1 0 .
1 8 . . 1 x1
1 24 . . 0 x2
Second, suppose there was an observation at time 4 (during the "on
drug" period) and at time 24 (the end of the "off drug" period). It is
necessary to use the "other" type event records (EVID) to signal the
change in drug status.
#ID TIME AMT RATE ONDR X=DV EVID
1 0 9999 -1 0 . 1
1 4 . . 1 x1 0
1 8 . . 1 . 2
1 24 . . 0 x2 0
Modeling Rates and Effects of Drug X
I just read the question of the Nonmemuser and Lewis answer and want to add
the following:
You measure concentration of X as response. Since you do not know the volume
of distrubution for that substance you can not estimate the total amount
of the substance in the body and it's change. Therefor you can not estimate
the real production rates.
However you can model concentration of X in response to your drug, if you
assume an arbitary volume of distribution for X. I usually set it just to 1,
because it does not matter and then the "amount" in that compartment has the
same scale as the measured concentrations. The estimated production rates will
not correspond to the real ones, i.e. they will not predict the real amount
produced, but using them you predict concentration of X correctly and that
is probably what you are interested in.
I thought perhaps others might be interested in the following
exchange - I've disguised the drug/firm/person in case there's a
confidentiality issue.
===================================================================
----- Begin Included Message -----
Dear Dr. Sheiner,
Using NONMEM, I am seeking to model the serum concentration of endogenous
substance X in response to multiple doses of a drug that affects X.
The model we're employing is a
one-compartment model with first-order elimination of X and zero-order
(endogenous) input of X. The drug is hypothesized to alter the elimination rate
constant and/or the input rate. So there are four parameters we seek to
estimate:
1) X input rate while on drug
2) X input rate while off drug
3) X elimination rate constant while on drug
4) X elimination rate constant while off drug
My question is this: it seems that in order to estimate the X input rates, we
need to specify the total amount of X the body produces for the duration of
the study. The total amount produced is, of course, unknown: the best we can do
is to make an educated guess as to the total amount. Is there any way around
this problem- namely, is it possible to estimate both the X input rates and
the total amount of X produced? Any guidance you can provide us with respect
to this problem would be gratefully appreciated. Thank you for your attention
to this matter.
Sincerely yours,
A Nonmem User
----- End Included Message -----
I think your problem is very simple and you are not aware of the
capacity to model the rate of an infusion. Taking you literally, you
seem to have only 2 conditions, off drug (ONDR=0) and on drug
(ONDR=1). If so, then you can do your problem with ADVAN1 using
a $PK something like
$PK
KK = THETA(1) ; OFF-DRUG TV ELIM RATE
IF(ONDR.EQ.1) KK = THETA(2) ; ON-DRUG TV ELIM RATE
K = KK*(1+ETA(1)) ; ON-DRUG INDIVIDUAL ELIM RATE
RR = THETA(3) ; OFF-DRUG TV INPUT RATE
IF(ONDR.EQ.1) RR = THETA(4) ; ON-DRUG TV INPUT RATE
R1 = RR*(1+ETA(2)) ; ON-DRUG INDIVIDUAL INPUT RATE
Your input record would be
$INPUT ID TIME AMT RATE ONDR X=DV
and the first record of each subject would specify an infusion of
"infinite" total amount of X which enters at a modeled rate (R1 in $PK).
Your data file might then start (if # was named as IGNORE=# on your
data record), as:
#ID TIME AMT RATE ONDR X=DV
1 0 9999 -1 0 .
etc.
Note:
1.You can see that it would be easy to substitute a "dose-response"
function (e.g., linear) for the in or out-rates if you had a DOSE
(of drug) column in the data.
2. If you wanted to write a PK/PD model with PK of drug and kinetics
of X affected by drug, you could do so using the same input trick,
but of couse, you'd have to write your model using ADVAN6 or higher
(since out rate of X would be the scaled product of drug concentration
times X concentration).
3. ONDR must have the correct value on every record the way I've written
$PK because K and R1 are recomputed at every time depending on its value.
Good luck,
LBS.
And another comment on the "Drug affects X" problem: If X is endogenous,
then it is already at steady-state prior to an intervention (indeed, one
usually measures a "base-line" value), and starting the data off
with a time zero infusion actually starts the system from zero, not
steady-state. If the first observation is sufficiently long after
time zero, the system will get to steady-state and there should be
no problem. If you have a base-line measurement, you will want
to start the sytem off at seady-state, which you can do using,e.g.,
$INPUT ID TIME AMT RATE ONDR X=DV SS
$SUBR ADVAN1
$PK
KK = THETA(1) ; OFF-DRUG TV ELIM RATE
IF(ONDR.EQ.1) KK = THETA(2) ; ON-DRUG TV ELIM RATE
K = KK*(1+ETA(1)) ; ON-DRUG INDIVIDUAL ELIM RATE
RR = THETA(3) ; OFF-DRUG TV INPUT RATE
IF(ONDR.EQ.1) RR = THETA(4) ; ON-DRUG TV INPUT RATE
R1 = RR*(1+ETA(2)) ; ON-DRUG INDIVIDUAL INPUT RATE
etc.
with data file
#ID TIME AMT RATE ONDR X=DV SS
1 0 0 -1 0 . 1
1 0 9999 -1 0 . .
etc.
As an additional comment on this problem:
If you can model the concentrations of the drug then it should be
possible to model the effect of the drug on either input or elimination
of X as a function of the drug concentration. This would require
specifying the problem with a differential equation defined model e.g.
with ADVAN6. The use of drug concentration rather than dose would give
you more information about the relationship of intensity of drug
treatment to the effect on X kinetics and of course the model is more likely
to represent reality by proposing a causal relationship via
concentration rather than dose.
This model has been studied a lot - warfarin pharmacodynamics would be
the most classical case but other examples have been well described e.g.
steroid dynamics (Bill Jusko's group in JPB within the last 12 months).