E-CV Lewis Sheiner

Research Interests

The overall research goal of Dr Sheiner's laboratory is to develop mathematical
models and data analysis techniques that will allow quantitative understanding
of outcomes of drug therapy from in vivo observations, especially those
made in the course of drug development trials and routine patient care.
He has used the application areas of AIDS therapy, Anesthesiology, and
to a lesser extent Health Care Research (no longer a primary focus of his
research) as "laboratories" for the development and testing of such methodologies.

Application of Non-linear Random Effects Models

Models of clinical phenomena, such as the relationship of drug dosage to drug
concentration (Pharmacokinetics, PK) or effect(Pharmacodynamics, PD)are
quantified by parameters describing mean effects, for example how drug clearance
relates to renal function, and variability, for example the extent of
interindividual variability remaining after the dependence of clearance on
renal function is accounted for. To estimate such parameters from sparse individual
data (e.g., from observational data or observations "added-on" to the standard
set in a clinical trial) requires sophisticated statistics to deal with
interindividual design variation, imbalance, and other "data" problems.
Hierarchical models, that is, models which embed a scientific individual model
in a second stage model for interindividual variability as a function of
covariates and random effects, is the natural, albeit statistically challenging
framework for inference and prediction based on such data. In collaboration with
his long-time colleague, Prof. Stuart Beal, investigation of methods for
analyzing data according to such models has been an ongoing interest. Current
work focuses primarily on improved methods of estimation and validation for
hierarchical models, and on extension of hierarchical models to deal with the
potential confounding due to deviations from protocol, such as noncompliance
and dropout. The results of this work are methods that allow one to learn and predict
how individuals will respond to various drug regimens, using experimental
designs that resemble therapy more than currently standard trial designs. The
ultimate value of these methods is in designing more efficient and informative
clinical trials (using clinical trial simulation, a technique engendering increasing
interest), optimizing dosage recommendations, and via empirical bayes
methods, optimizing individual therapy. The work has had considerable impact on,
and wide application in Clinical Pharmacology research, the Pharmaceutical
industry, and drug regulation.

PK/PD Models

Developing models to analyze simultaneous PK/PD data, especially
non-steady-state data have been a continuing interest. Semi-parametric models
build in justifiable assumptions about system structure without adding
unjustifiable ones, and development of such models has been done in
collaboration primarily with Prof. D. Verotta. The current emphasis in
the PK/PD modeling work is to apply sophisticated mathematical/statistical
and often non-parametric modeling approaches to non-steady-state (also possibly
non-linear and non-stationary)PK/PD systems with varying types of outputs (categorical,
ordered categorical, time to event, continuous). Modeling of high-dimensional
complex dynamic systems (such as the AIDS virus and the immune system) is also
of interest. The techniques used to accomplish this goal are borrowed
from linear systems theory, image reconstruction, engineering, and modern
statistics. The techniques developed in Dr Sheiner's laboratory are enjoying
increasing application by others, and are permitting researchers to isolate and
quantify steady-state PK/PD relationships from noisy dynamic in vivo observations.