From:"Bonate, Peter" 
Subject:[NMusers] weight as a covariate in kids
Date:Wed, 30 Oct 2002 08:52:50 -0600

Dear all,
I am sure this will generate a lot of traffic but I have a question.  In developing
a model in kids (infants,neonates, and children) should weight be included as an a priori
covariate used in the development of the structural model even if its inclusion does not
improve the goodness of fit.  I have heard alot of people say 'yes' but I wanted to get
a group opinion.  Then if so, should it go on all structural parameters, except perhaps
absorption-related terms? And should it always be modeled as a power function?

Let the games begin.  Thanks,

pete


Peter L. Bonate, PhD
Director, Pharmacokinetics
ILEX Oncology, Inc
4545 Horizon Hill Blvd
San Antonio, TX  78229
phone: 210-949-8662
fax: 210-949-8487
email: pbonate@ilexonc.com
___________________________________

From:Edmund Capparelli 
Subject:Re: [NMusers] weight as a covariate in kids
Date:Wed, 30 Oct 2002 13:18:02 -0800

After being involved primarily in pediatric population pharmacokinetic 
analyses over the past 12 plus years, I feel very strongly that accounting 
for size upfront is critical to modeling pediatric pharmacokinetic 
data.  It is my underlying assumption that "the truth" is that 
pharmacokinetic parameters scale with subject size.  Since size is highly 
correlated with many other covariates (age, creatinine clearance etc.) and 
is an extremely powerful covariate, any evaluation of other covariates in a 
structural model without size will have little resemblance to their impact 
in a multivariate analysis with size and thus provide limited insight.  In 
general pediatric pharmacokinetic studies minimize the data collected and 
are not designed to robustly answer the size question in regards to all 
pharmacokinetic parameters.  However in these situations, if one does not 
include size in PK parameters because it does improve the fit by some 
statistical criteria, there needs to be recognition that an assumption has 
been made, that the study design was powerful enough to determine the 
impact of size on all pharmacokinetic parameters.  In this setting ignoring 
size make fit the data as just as well, but it leads to bizarre unrealistic 
extrapolations just outside age, covariate, dose-sampling domain that 
generated the data.  And even if we cautions against extrapolation in these 
settings, we modelers need to recognize that the general lack and 
fragmentation of pediatric pharmacokinetic information make extrapolation 
of pediatric PK data a common practice.  Not including size also ignores a 
large knowledge of pharmacokinetics in children were we have good 
pharmacokinetic information from toddler to adults and size has born out to 
be a significant covariate on Vd and Cl (without exception to my knowledge).

I include size on all of my size dependant parameters which may include 
absorption. if zero order.  In the modeling process it is also important to 
recognize that size affects multiple pharmacokinetic parameters and there 
are interactions in these influences, so the standard forward covariate 
selection approach can grossly underestimate the impact of size on 
individual parameters or miss a size covariate entirely.  Use of a 
backwards elimination approach prevents missing these complex interactions 
but rarely are there sufficient data in pediatric pharmacokinetic studies 
to support this approach for all covariates.

Lastly getting to how specifics of how to incorporate size into the model 
really depends on what questions one is asking.  As a starting point I 
agree with Nick Holford's approach to used standardized allometric 
scaling.  It promotes comparability of data, has a sound scientific 
theoretic basis, is frequently very close to fitted exponents and in most 
situations superior to a linear weight function.  However, I would caution 
that it does not account for the ontogeny of clearance processes and other 
potential age related pharmacokinetic differences.  I also keep in the 
forefront of mind the quote by Box that "all models are wrong but some are 
useful" and do not believe in an ultimate "final" correct model.  There are 
"final" models given any specific approach (and underlying assumptions) and 
there may be usefulness in developing various "final" models with different 
sizing approaches to develop and justify pediatric dosing paradigms.

Best Regards,

Edmund
___________________________________