Re: Sparse (pediatric) and rich (adult) data
Steve - Thanks for making the point about the importance of experimental design. Often times when pooling adult and pediatric data, data are imbalanced, and pediatric PK designs are much less informative than the adult data. If, for a particular drug and disease state, pediatric patients really are just small adult patients, the design deficiency isn't much of a concern - but that's not always the case.
Although very useful for scaling body-size related differences in PK parameters from adults to peds, the allometric "small adult" assumption, doesn't always provide the complete story. There are other bits of information about pediatric PK (e.g. developmental changes, pediatric disease state effects) that we'd like to learn about directly from the pediatric data.
The analysis of the pooled data in this case (sparse, poorly- optimized pediatric data with more informative adult data) is similar to a Bayesian data analysis, with informative prior distributions for most/all model parameters. An alternative approach to analyzing the sparse pediatric data could be:
1. Assess the expected precision of PK parameters under the pediatric data alone, using a PFIM-type method. 2. Analyze the pediatric data, using a full Bayesian estimation method. Informative prior distributions based on adults would be selectively applied to those parameters with poor design support in the pediatric data alone, while other parameters which are of particular interest in the pediatric population could be estimated with diffuse prior distributions.
This approach allows the pediatric data alone to influence the estimation of a subset of parameters (hopefully, those components you'd like to learn about), while relying on prior adult information to anchor some of the more poorly supported components of the model.
Marc
Marc R. Gastonguay, Ph.D.
President & CEO, Metrum Research Group LLC [www.metrumrg.com]
Scientific Director, Metrum Institute [www.metruminstitute.org]
Direct: 860-670-0744 Main: 860-735-7043
Email: [EMAIL PROTECTED]
Quoted reply history
On May 28, 2008, at 9:49 PM, Stephen Duffull wrote:
> Leonid
>
> > I hope that you do not dispute that in this particular case
> > you need to use adult data (50 full profiles) rather than
> > discard them and use only kids data (3 sample per subject, 20
> > subjects)?
>
> I definitely do not dispute the need to have both adult and paediatric data in the analysis (so I agree :-) ). I see two reasons for this (perhaps more if I took more time). The first and most important reason is combining
>
> adult and paediatric data together is a great (only) way to learn how
>
> children differ pharmacokinetically from adults and how doses can be scaled to achieve equivalent exposures. Secondly, especially in this case, it is often helpful to combine data sets together to improve the informativeness
>
> of the overall design. This latter point, however was the point of my
>
> previous email. Some care must be taken to assess the accuracy of covariate
>
> effects given the unbalanced nature of the design.
>
> > While optimal design can be used to extract more
> > information from the same number of samples, it is not a
> > substitute for the real data. Even with optimal design of the
> > pediatric study (with the same 20 subjects, 3 optimal sample
> > points) I bet you would gain by using adult data as well.
>
> You always gain by summing over data (unless the new data is negatively informative which is unlikely in any PK situation). So I don't exactly follow your point. The question to me is simply, what chance do I have of
>
> identifying a model that allows me to draw appropriately accurate
>
> conclusions. Optimal design is a way that allows investigators to improve
>
> the informativeness of data. Obviously, no data = no information.
>
> Steve
> --
> Professor Stephen Duffull
> Chair of Clinical Pharmacy
> School of Pharmacy
> University of Otago
> PO Box 913 Dunedin
> New Zealand
> E: [EMAIL PROTECTED]
> P: +64 3 479 5044
> F: +64 3 479 7034
>
> Design software: www.winpopt.com