The Role of Statistical Principles in
                   Quantitative Biomedical Modeling

                             Marie Davidian
        Department of Statistics, North Carolina State University

It is increasingly recognized in the biomedical research community, in
the pharmaceutical industry, and among regulatory authorities (e.g.,
the FDA) that mathematical models of biological systems have the
potential to reveal new insights into mechanisms underlying disease
that may be exploited to develop new treatment strategies and design
clinical studies. This enterprise requires that the models be tested
on and applied to clinical data and that the inherent variation among
individuals in the population of diseased subjects be characterized.
In order that this be carried out in a principled way, the
mathematical models should be embedded into a statistical model
framework in which the variation in data on individual subjects (e.g.,
errors in measurement) and across individuals may be faithfully
represented and from which appropriate approaches to fitting the
models to data may be derived.  Focusing on the particular situation
where a mathematical model for within-individual disease mechanisms is
available, we will describe such an integrated
mathematical-statistical framework and emphasize why it is essential
that use of mathematical models in this setting take place within it.
A project involving the use of HIV dynamic models to design treatment
strategies and clinical trials will illustrate the model formulation
and concepts.