NONMEM Users Network Archive

Hosted by Cognigen

Re: mixture model coding

From: Nick Holford <n.holford>
Date: Sat, 05 May 2012 19:14:16 +0200


Some suggestions:

1. I cannot see how you are introducing treatment into the model. You
have something that is a function of time (EFF1) but not of treatment.
I'd suggest including whether treatment is being used or not
(statistical/regulatory view from a dark cave) or a function of dose (a
little bit more pharmacological and enlightened) or even related to drug
concentration (science?).
2. Disease progression in some patients may have a different direction.
Some patients (maybe most) will have an increase in tumour size. Others
may have a decrease without treatment ('spontaneous remission'). I would
not force all patients to go in the same direction by using EXP(eta).
3. Similar remarks relate to treatment effects. The null hypothesis
would assume that individual treatment effects could be both good and
bad. Using EXP(eta) forces the random differences between subjects to be
in the same direction in all subjects.
4. Both the disease progression and treatment random effects would make
fewer assumptions if they were coded as (1+eta) rather than exp(eta).

A mixture model might be a helpful to distinguish between poor
responders and good responders to treatment or to distinguish between
slow progressors or fast progressors without treatment or to distinguish
between patients with smaller or bigger tumour sizes at baseline. This
means you could try a mixture model to help describe any or all of the
parameters that you are suggesting deserve some between subject variability.

But first of all I'd suggest adding treatment to your model....


PS You should consider learning about the difference between sex and gender.
Kim JS, Nafziger AN. Is it sex or is it gender? Clin Pharmacol Ther
2000; 68: 1-3.

On 5/05/2012 6:43 p.m., Gaurav Bajaj wrote:
> Dear All,
> I am trying to model a tumor size data at different time points using
> NONMEM. There is high variability in baseline tumor size and their
> might be sub-populations in the dataset with different distribution
> for size progression. For example, in many cases the baseline tumor
> size is around 100- 200 units, and there are few patients with
> baseline size around 500 - 700 units. Due to this high variability I
> want to try a mixture model - can anybody suggest on how to do it ? I
> have listed the initial code that I used.
> $PROB run# 101
> BASE= THETA(1)*EXP(ETA(1)) ;baseline tumor size
> PR= THETA(2)*EXP(ETA(2)) ;linear tumor progression
> TR=THETA(3)*EXP(ETA(3)) ;treatment effect
> Y=IPRED + ERR(1)
> (0,100) ; baseline
> (0.001,0.5 ) ; progression
> (0.01, 0.06, 1) ; treatment
> 0.5 ; ETA-Baseline
> 0.2 ; ETA-PR
> 0.2 ; ETA-treatment
> 20 ; ERR-add
> ----------
> Thanks,
> Gaurav
> --
> Gaurav Bajaj
> Postdoctoral Fellow, Pharmacometrics
> Laboratory for Applied PK/PD
> Clinical Pharmacology & Therapeutics
> The Children's Hospital of Philadelphia

Nick Holford, Professor Clinical Pharmacology

First World Conference on Pharmacometrics, 5-7 September 2012
Seoul, Korea

Dept Pharmacology& Clinical Pharmacology, Bldg 505 Room 202D
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
email: n.holford

Received on Sat May 05 2012 - 13:14:16 EDT

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: