NONMEM Users Network Archive

Hosted by Cognigen

RE: Weight based dosing

From: Stefanie Hennig <s.hennig>
Date: Wed, 9 Dec 2015 11:26:33 +0000

Dear Nicolas,

In the study we tested several different degrees of correlation between cov=
ariates. On purpose, we did not attribute a specific clinical meaning to an=
y covariates. However, you could if you wanted to.

We found that once a covariate was included in the model, adding a second c=
ovariate that was correlated with the first covariate, the reduction in the=
 unexplained variability was not as large compared to when we added the fir=
st covariate. This happened even with both covariates being true covariates=
 and both had the same size of relationship with the parameter.
Overall, we found that the change in OFV is a very good indicator for findi=
ng the true covariate. We also found that small reductions in unexplained v=
ariability might actually be connected with larger changes in explained var=
iability. The latter is not something we look at readily. Specifically in =
NONMEM, it is not something that can be read off from the standard output.=
 We have made a distinction in the article between model fit and clinical i=
nfluence of a covariate. These are two different things. Just like Mike Fo=
ssler mentioned earlier in his reply.

Best wishes

-----Original Message-----
From: SIMON Nicolas [mailto:Nicolas.SIMON
Sent: Wednesday, 9 December 2015 7:01 PM
To: Stefanie Hennig; Sultan,Abdullah S; nmusers
Subject: RE : Weight based dosing

Hi Stefanie,
in your article did you take into account that covariates may have clinical=
Best regards

Pr Nicolas SIMON MD PhD
Clinical Department
Medical School of Marseille

De : owner-nmusers
 de Stefanie Hennig [s.hennig
re 2015 02:59 : Sultan,Abdullah S; nmusers
ers] RE: Weight based dosing

Dear Abdullah,

I assume that you have only looked at the decrease in the variance around C=
L comparing the base model and the covariate model, when you state : "only =
explained 9% of the variability".
We have shown that total parameter variability is changing throughout model=
 building and a decrease in unexplained parameter variability is not equal =
to an increase in explained parameter variability when adding covariates in=
 your model. So the explained parameter variability might have increased b=
y more than 9%.
 Please see the reference below. This methodology is now also implemented i=
n PsN and you can perform it alongside your covariate model building. You m=
ight want to try this. The manuscript below also discusses the difference b=
etween improved model fit and clinical significance of a covariate.
(Hennig S, Karlsson MO. Concordance between criteria for covariate model bu=
ilding. J. Pharmacokinet. Pharmacodyn. 2014;41:109-125.)

Further, I would like to highlight to you that others have previously discu=
ssed on NMusers and in the literature that for an easier comparison of stud=
y results it is preferred to use a standard weight of 70kg.
Also, if you did use an allometric scaling model on CL/F, you would have us=
ed an power exponent of or estimated this exponent, but not a slope. =
So I am unsure about the slope effect that you are talking about and cannot=
 comment further on this.
Best wishes and a Happy holiday season

Dr Stefanie Hennig

Lecturer | Pharmacometrics
School of Pharmacy| Pharmacy Australia Centre of Excellence (PACE) |The Uni=
versity of Queensland, QLD 4072, Australia
Phone: +61 7 334 61970, Fax: +61 7 334 61999, Email: s.hennig
Please note my working days are Monday to Thursday.

"You can't fix by analysis what you bungled by design." Light, Singer and W=

The World Conference of Pharmacometrics in Brisbane 2016<h=


From: owner-nmusers
 Behalf Of Sultan,Abdullah S
Sent: Wednesday, 9 December 2015 10:40 AM
To: nmusers
Subject: [NMusers] Weight based dosing

Hi everyone,

I am developing a POP PK model for an anti-infective drug, I am trying to d=
etermine if dosing should be weight based or not. The range of weight in th=
e study was 40-100 kg.

Weight was statistically significant for Cl/F but only explained 9% of the =
variability observed for Cl.

I used allometric scaling to describe weights effect on Cl/F and slope effe=
ct of weight was 0.58, and scaled to 60 kg (the median).

Based on the slope effect estimated, AUC is predicted to decrease by 15% fo=
r an 80 kg individual, and increase by 25% for an individual that weights 4=
0 kg compared to a 60 kg individual.

How much should I trust the slope effect determined by my study? and should=
 I rely on it to develop the dosing regimen?

if weight only explained 9% of variability observed with Cl/F, could that i=
ndicate that it is not clinically significant and weight based dosing is no=
t required?


Abdullah Sultan, PhD candidate

University of Florida
Received on Wed Dec 09 2015 - 06:26:33 EST

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to:

Once subscribed, you may contribute to the discussion by emailing: