From: Nick Holford <*n.holford*>

Date: Fri, 17 Oct 2008 09:27:35 +1300

Ayyappa,

It is possible to do some testing of demographic covariates using ANOVA

-- but this requires the two stage population approach i.e. obtain

individual parameter estimates first then apply ANOVA as though the

parameters were observations. The standard two stage method is known to

produce biased estimates of the variances because the true between

subject variability is confounded with the individual parameter

estimation error (Sheiner 1984). There are fancier two stage methods

that can account for this but they are somewhat complicated. All two

stage methods require that there is sufficient data per individual to

estimate all parameters of interest. While this is desirable from a

design viewpoint, even for a full population analysis, the reality is

that PKPD studies are usually sub-optimally designed and individuals may

not have enough observations. The two stage approach cannot deal with

this but the full population approach can.

The full population approach allows you to simultaneously estimate the

relationship between the parameter of interest eg EC50 and the covariate

e.g. age on a continuous scale. This means you get a more realistic

estimate fo parameter uncertainty because you are not making the

assumption that the EC50 values are estimated without error e.g. bias

arising from not understanding the covariate relationship. In addition,

covariate relationships can be non-linear. ANOVA cannot handle

non-linear covariate relationships as far as I know. Thus the population

approach is more honest and more flexible than ANOVA.

Finally as Sir Michael Rawlins (Chairman of the NICE in the UK) pointed

out yesterday the traditional statistical approach to clinical trials

does not adequately describe the clinical pharmacology and benefits of

medicines. The flexibility of the population approach allows it to used

for 'learning' as well as 'confirming' (Sheiner 1997). This combination

of approaches is in keeping with the broader philosophy posed by Rawlins.

http://www.politics.co.uk/opinion-formers/press-releases/royal-college-physicians-sir-michael-rawlins-attacks-traditional-ways-assessing-evidence-$1245035$365674.htm

Sheiner LB. The population approach to pharmacokinetic data analysis:

rationale and standard data analysis methods. Drug Metab Rev.

1984;15(1-2):153-71.

Sheiner LB. Learning versus confirming in clinical drug development.

Clinical Pharmacology & Therapeutics. 1997;61(3):275-91.

ayyappa.5.chaturvedula

*>
*

*> Nick,
*

*>
*

*> May be I did not put forth the question right. Let me try again, I
*

*> want to know the advantage of analyzing the pooled data from different
*

*> clinical studies to understand the demographic differences by ANOVA vs
*

*> finding a demographic covariate tested by NONMEM.
*

*>
*

*> Regards,
*

*> Ayyappa Chaturvedula
*

*> GlaxoSmithKline
*

*> 1500 Littleton Road,
*

*> Parsippany, NJ 07054
*

*> Ph:9738892200
*

*>
*

*>
*

*> *"Nick Holford" <n.holford *

*> Sent by: owner-nmusers *

*>
*

*> 16-Oct-2008 14:38
*

*>
*

*>
*

*> To
*

*> nmusers *

*> cc
*

*>
*

*> Subject
*

*> Re: [NMusers] Why population PK approach is good for pooling data
*

*> compared to Classical statistical analysis?
*

*>
*

*>
*

*>
*

*>
*

*>
*

*>
*

*>
*

*>
*

*>
*

*> Ayyappa,
*

*>
*

*> As far as I know any 'classical statistical analysis' that one can do
*

*> using regression can be done with NONMEM. It is also possible to do
*

*> hypothesis testing on means (t-test, ANOVA), logistic regression and
*

*> survival analysis.
*

*> What kinds of 'classical statistical analysis' do you want to do that
*

*> you cannot do with NONMEM?
*

*>
*

*> Nick
*

*>
*

*>
*

*> ayyappa.5.chaturvedula *

*> >
*

*> > Dear Group,
*

*> >
*

*> > Could anybody explain or direct me to some literature why population
*

*> > PK approach allows us to pool data from different studies but no
*

*> > classical statistical analysis?
*

*> >
*

*> > Regards,
*

*> > Ayyappa Chaturvedula
*

*> > GlaxoSmithKline
*

*> > 1500 Littleton Road,
*

*> > Parsippany, NJ 07054
*

*> > Ph:9738892200
*

*>
*

*> --
*

*> Nick Holford, Dept Pharmacology & Clinical Pharmacology
*

*> University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
*

*> Zealand
*

*> n.holford *

*> http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
*

*>
*

*>
*

*>
*

--

Nick Holford, Dept Pharmacology & Clinical Pharmacology

University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand

n.holford

http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford

Received on Thu Oct 16 2008 - 16:27:35 EDT

Date: Fri, 17 Oct 2008 09:27:35 +1300

Ayyappa,

It is possible to do some testing of demographic covariates using ANOVA

-- but this requires the two stage population approach i.e. obtain

individual parameter estimates first then apply ANOVA as though the

parameters were observations. The standard two stage method is known to

produce biased estimates of the variances because the true between

subject variability is confounded with the individual parameter

estimation error (Sheiner 1984). There are fancier two stage methods

that can account for this but they are somewhat complicated. All two

stage methods require that there is sufficient data per individual to

estimate all parameters of interest. While this is desirable from a

design viewpoint, even for a full population analysis, the reality is

that PKPD studies are usually sub-optimally designed and individuals may

not have enough observations. The two stage approach cannot deal with

this but the full population approach can.

The full population approach allows you to simultaneously estimate the

relationship between the parameter of interest eg EC50 and the covariate

e.g. age on a continuous scale. This means you get a more realistic

estimate fo parameter uncertainty because you are not making the

assumption that the EC50 values are estimated without error e.g. bias

arising from not understanding the covariate relationship. In addition,

covariate relationships can be non-linear. ANOVA cannot handle

non-linear covariate relationships as far as I know. Thus the population

approach is more honest and more flexible than ANOVA.

Finally as Sir Michael Rawlins (Chairman of the NICE in the UK) pointed

out yesterday the traditional statistical approach to clinical trials

does not adequately describe the clinical pharmacology and benefits of

medicines. The flexibility of the population approach allows it to used

for 'learning' as well as 'confirming' (Sheiner 1997). This combination

of approaches is in keeping with the broader philosophy posed by Rawlins.

http://www.politics.co.uk/opinion-formers/press-releases/royal-college-physicians-sir-michael-rawlins-attacks-traditional-ways-assessing-evidence-$1245035$365674.htm

Sheiner LB. The population approach to pharmacokinetic data analysis:

rationale and standard data analysis methods. Drug Metab Rev.

1984;15(1-2):153-71.

Sheiner LB. Learning versus confirming in clinical drug development.

Clinical Pharmacology & Therapeutics. 1997;61(3):275-91.

ayyappa.5.chaturvedula

--

Nick Holford, Dept Pharmacology & Clinical Pharmacology

University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand

n.holford

http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford

Received on Thu Oct 16 2008 - 16:27:35 EDT