- On 23 Jul 2013 at 08:21:12, Xinting Wang sent the message

Back to the Top

Dear all,

I am trying to build a population PK model with 2 compartment, first order absorption. The raw data

has four doses, and I included a fractional absorption compared with the lowest dose in the control

file. The base model has already had 13 THETA values. So, I am wondering, if after base model and I

get an estimation of the Fractional Absorptions, could I set them (Estimation of Fractional

Absorption) to fixed values, so that adding more covariate might work out faster and avoid the error

message in the covariance step? Many thanks.

Below is the PK part of the control value.

FA1=0

FA2=0

FA3=0

FA4=0

IF(DOSE.EQ.250) THEN

FA1=1

ENDIF

IF(DOSE.EQ.500) THEN

FA2=1

ENDIF

IF(DOSE.EQ.850) THEN

FA3=1

ENDIF

IF(DOSE.EQ.1000) THEN

FA4=1

ENDIF

F1=FA1+FA2*THETA(10)+FA3*THETA(11)+FA4*THETA(12)

BWNORM=BW/66

TVCL=THETA(1)

CL=TVCL*BWNORM**THETA(6)*EXP(ETA(1))

TVV2=THETA(2)

V2=TVV2*BWNORM**THETA(7)*EXP(ETA(2))

TVKA=THETA(3)

KA=TVKA*BWNORM**THETA(13)

TVQ=THETA(4)

Q=TVQ*BWNORM**THETA(8)*EXP(ETA(3))

TVV3=THETA(5)

V3=TVV3*BWNORM**THETA(9)*EXP(ETA(4))

S2=V2/1000

S3=V3/1000

--

Xinting - On 23 Jul 2013 at 11:05:02, Nick Holford sent the message

Back to the Top

Xinting,

This kind of question is probably better sent to nmusers rather than pharmpk. The moderator of

pharmpk censors previous postings which makes it hard to interpret responses because the background

information is not included.

In this particular case you can reduce the number of THETAS by 4 by not attempting something with

little expectation of value by trying to estimate allometric exponents for CL, V2, Q and V3. The

allometric exponent is known so why waste your time trying to estimate it? If you have a small data

set with a narrow distribution of weights e.g. 100 subjects with a CV of weight of 20% then you

cannot expect to get a precise estimate of the exponent (see Anderson & Holford 2008).

It makes more sense to standardize parameters to 70 kg rather than an arbitrary value of 66 kg. See

Holford, Yeo, Anderson 2013).

You should NOT fix the estimates of F at different doses when investigating other covariates. If

other covariates are explanatory then clearly your base model estimates indicate some model

misspecification and so your estimate of F will be wrong. If there are really differences in F with

dose then these should be estimated as part of your final model.

It is a common mistake to assume that failure of convergence in NONMEM is an indication of a bad

fit. This is not true. Many people have looked at this and found no evidence for an association

between convergence and goodness of fit. Go and read the nmusers archives

(http://www.mail-archive.com/nmusers.-a-.globomaxnm.com/msg01087.html). You should evaluate your model

based on parameter plausibility and VPCs (don't waste your time with residual plots and so called

diagnostic plots of PRED vs DV, IPRED vs DV).

Best wishes,

Nick - On 24 Jul 2013 at 09:02:50, Xinting Wang sent the message

Back to the Top

Dear Nick,

Thanks very much for your kind reply.

I understand that in general the rule of allometric scaling for CL follows

the 3/4 rule. As you and B.J. Anderson pointed out in your paper (B.J.

Anderson, N.H.G. Holford, 2008, Annu. Rev. Pharmacol. Toxicol, 48:303-32),

no evidence has been provided to reject this rule. However, as you also

mentioned in this paper, "experimental designs for estimating clearance are

much less robust and the error in estimation of the allometric coefficient

can be expected to be larger...".

Theoretically, we do not have proof to reject this, but in the particular

case, using 3/4 might actually increase OFV and thus reduce the

goodness-of-fit. When I replaced the 0.75 with a THETA on CL, the OFV value

decreased by 15, which I think should be a significant impact on

estimation. However, replace them all with THETAs could on the other hand

decrease the ability of successful estimation. How should balance this

problem?

Regards - On 24 Jul 2013 at 11:12:47, Nick Holford sent the message

Back to the Top

Xinting,

You wrote:

"Theoretically, we do not have proof to reject this, but in the particular case, using 3/4 might

actually increase OFV and thus reduce the goodness-of-fit. When I replaced the 0.75 with a THETA on

CL, the OFV value decreased by 15, which I think should be a significant impact on estimation.

However, replace them all with THETAs could on the other hand decrease the ability of successful

estimation. How should balance this problem? "

As I pointed out in my previous email ("Go and read the nmusers archives

(http://www.mail-archive.com/nmusers.aaa.globomaxnm.com/msg01087.html))" NONMEM models should not be

judged by "successful termination". Did you read and understand that?

In general fixing parameters for all those parameters which theory provides a value will stabilise

your estimation.

Theory is always preferable to empirical results. So in fact there is no problem.

Best wishes,

Nick - On 24 Jul 2013 at 13:11:39, Stephen Duffull sent the message

Back to the Top

Xinting

I think something of more general interest is considering whether you wish to use "a priori"

covariate models or "a posteriori" covariate models.

A priori covariate models are derived based on previous theoretical work or empirical work by others

and have been tried in other circumstances (e.g. allometric scaling, Cockroft and Gault for eGFR

etc).

A posteriori covariate models are derived based on your data only to provide the best fit for your

current data set.

This is a conceptual choice and defines the way you build models. There is no right and wrong about

what to do it - it's just up to you.

Note, however, that a priori covariate models have generally stood the test of time and shown to be

useful in a wide spectrum of the population (e.g. eGFR models) whereas an posteriori model has only

been tested in your specific population.

The question, from an epidemiology perspective is: are you looking for internal or external

validity. The question from a pharmacometrics perspective is are you interested in describing your

data or predicting into new settings.

On the whole I feel more comfortable with using tried and tested models ...

Regards

Steve

--

Professor Stephen Duffull

Chair of Clinical Pharmacy

School of Pharmacy

University of Otago

PO Box 56 Dunedin

New Zealand

E: stephen.duffull.-a-.otago.ac.nz

www.pharmacometrics.co.nz

Want to post a follow-up message on this topic?
If this link does not work with your browser send a follow-up message to PharmPK@lists.ucdenver.edu with "Is it reasonable to set a few THETA at fixed values" as the subject |

Copyright 1995-2014 David W. A. Bourne (david@boomer.org)