Designs and Models in Drug Development
Monday 2 June, 2:40--4:30
- Moderator: Alfred BARRON
- Johnson \& Johnson
- Krishnendu GHOSH
- Wyeth-Ayerst Research, Princeton, NJ, USA
The potencies of pharmacological competitive antagonists are commonly
expressed in terms of PA2 values (negative of log of antagonist
concentration corresponding to a dose-ratio of 2).
Commonly used estimation method for PA2 is Schild's
method. This method does not utilize the variabilities in the ED50s in
calculating the PA2. Alternative design and model
was suggested by Lazareno and
Birdsall. This method requires much less experimental units and
use both the agonist and antagonist concentrations simultaneously.
Both the methods assume that the antagonist administered alone
(in the absence of the agonist) has no effect. Many recent studies show
that antagonist administered alone may have a dose-response relationship.
In such situations we have shown that both the above methods will
fail to provide appropriate PA2 estimates and indicate that
the antagonists are non-competitive. We have derived alternative designs
and models for estimating PA2, Kd, and IC50 for the antagonist
in such situations.
[Krishnendu Ghosh, Biometrics Research, Wyeth-Ayerst Research, ghoshk@war.wyeth.com]
- Paulette CEESAY, Sanat K. SARKAR
- Temple University, Philadelphia, PA, USA
- Steve SNAPINN
- Merck Research Laboratories, Blue Bell, PA, USA
Many researchers have studied the problem of testing the superiority
of a combination
of two drugs to its components using frequentist methodologies. However, the
frequentist
approach is complicated by the presence of the nuisance parameter measuring the
mean
difference between components, and this can lead to low power.
By using a prior that
reflects one's knowledge about the component drugs, we provide a formal Bayesian
approach
for studying several fixed-dose combinations of two drugs witn unknown variance
where the design of the experiment is a multilevel factorial clinical trial in
order to determine if at least one
provides a more effective treatment than its components. This approach avoids
the problem of dealing with the nuisance parameter, and is better at detecting
the superiority of a combination when the gains over its respective components
are moderate. The Bayesian procedure could be especially useful in Phase II
clinical trials.
[Paulette Ceesay, Temple University, Department of Statistics, Phila., PA 19122, USA, paulette@vm.temple.edu]
- George ZELIGER
- Trilogy Consulting Corp., Princeton, NJ, USA and
Anvical-Simplex, Boston, MA, USA
A modification of the Sequential Simplex Design
originally proposed by Spendley, Hext, and Himsworth is presented.
The technique is aimed to the situation where it is easier to
recognize the best outcome of those obtained in vertices of a regular
simplex in the factor space, rather than the worst one as it is in the
original approach.
The situation embraces cases of sensory as well as of multidimensional
evaluation of the optimization criterion, e.g., quality of a product or
performance of a process.
Formulae for simplex rotation are given, stopping rules and approaches
to dealing with constraints are discussed.
Use of sorting algorithms well known in Computer Science is proposed for
finding the pivotal vertex of the simplex. (The proposal has some
application value of its own.)
Convergence of the process is briefly discussed, results of numerical
experiments and comparison with other methods are reported.
As a by-product of the approach, the opportunity to run parallel
experiments or calculations is stated that allows to reduce the total
experimental time nearly proportionally to the number of the parameters
of the problem compared to the original approach. In other words, it
becomes possible to treat any multiparametric optimization problem as
one-dimensional.
Potential applications of the approach range from experimental design in
drug, food, and cosmetics development to car manufacturing to
architectural acoustic design.
[George Zeliger, 1725 Commonwealth Ave., Apt. 2, Brighton, MA, 02135, gzeliger@genetics.com]
Back to JRC program
Back to JRC Home Page