Applications of Mixed Effects Modeling
Tuesday 3 June, 2:40--4:30
- Organizer: Jose' PINHEIRO
- Bell Labs
- Moderator: Chuanhai LIU
- Bell Labs
- Jos\'e C. PINHEIRO
- Lucent Technologies, Murray Hill, NJ, USA
- Douglas M. BATES
- University of Wisconsin, Madison, WI, USA
Grouped data can be loosely defined as data in which the
observations are classified into disjoint groups. Some examples are
data from longitudinal studies or from split-plot designs.
Observations in the same group are usually correlated and
mixed-effects models constitute a powerful tool for modeling this
within-group dependence. Although grouped data with a single
grouping factor are the rule in Biology and Medicine, data with
multiple nested classification factors are frequently collected in
industrial experiments and in sociological studies. Displaying and
fitting mixed-effects models to multiple-nested data is considerably
more challenging than for grouped data with a single grouping
factor. We describe extensions to our \textsf{nlme} library for
\textsf{S-PLUS} to handle data with multiple levels of nesting.
These include different classes of data structures, graphical
displays based on the \textsf{trellis} library, and efficient
algorithms for maximum likelihood estimation. A real example from
integrated circuits manufacturing is used to illustrate the proposed
methods.
[Jos\'e C. Pinheiro, Bell Labs, Lucent Technologies, 600 Mountain Avenue Rm 2C-258, Murray
Hill, NJ 07974-2070, USA, jcp@bell-labs.com]
- David M. GILTINAN
- Genentech Inc., San Francisco, CA, USA
Determining an appropriate dosing regimen for a new drug in
clinical development is usually guided by the idea of a ``therapeutic
window'' for circulating drug concentration. The aim of dosing is to
maintain the drug concentration in the bloodstream within a target
range; concentrations which are too low may be subtherapeutic, while
exceeding the upper limit of the range may lead to toxicity.
Maintaining drug levels within the desired limits may be complicated
by large inter-subject variability in both the pharmacokinetic and
the pharmacodynamic response to the drug. The former can result in
large disparities in circulating drug levels in two subjects
following the same dosage regimen, while the latter corresponds to
differences in the concentration-effect relationship, so that a
common target range may not be appropriate for all subjects.
Accurate characterization of inter-subject variation in
pharmacokinetics and pharmacodynamics is thus essential to determine
an appropriate dosing strategy. Typically, such characterization
must be accomplished on the basis of infrequent, irregularly spaced,
measurements of pharmacokinetic and pharmacodynamic response in
subjects receiving the drug, though the number of subjects may be
large. The nonlinear mixed effects model provides a useful framework
for analyzing repeated measurement data of this type. Methods for
inference in this model framework will be discussed and their
application to investigating dosing strategies will be illustrated
using data from a clinical study of a novel cardiovascular agent.
[David M. Giltinan, Genentech Inc., San Francisco, CA, USA, giltinan@gene.com]
Back to JRC program
Back to JRC Home Page