Fall 2006 Meeting
Held on October 19,
2006 at the Renaissance Chicago North Shore.
The Program consisted of three
- Identifiability Assumptions for Missing
Covariate Data in Failure Time Regression Models
Paul Rathouz, Ph.D.,
University of Chicago
- Phase II Selection Design With Adaptive
Randomization in a Limited-Resource Environment
Woo Kim, Ph.D., TGRD
Observed Confidence Levels: Theory and Application
M. Polansky, Ph.D., Northern Illinois University
Identifiability Assumptions for Missing Covariate Data in Failure Time
Regression Models by Paul Rathouz, Ph.D.,
University of Chicago
Paul Rathouz is currently an Associate Professor of Biostatistics in the
Department of Health Studies at the University of Chicago.
He has published many peer reviewed papers covering a wide range of
topics in health statistics. Dr. Rathouz's statistical interests include work on
nuisance parameters, estimating functions, missing data, and longitudinal data.
His areas of application include aging research, childhood psychiatric
disorders (especially natural history studies) and some work in environmental
epidemiology. He has been a recipient of several fellowships, honors, and
awards, including the James E. Grizzle Distinguished Alumnus Award, for
outstanding contributions to Biostatistical methodology, consulting and/or
teaching in Department of Biostatistics at the University of North Carolina at
Chapel Hill. He has also been involved in organizing sessions at several ENAR/IMS
meetings. Dr. Rathouz holds a Bachelors degree in Mathematics from Rice
University, a Masters degree in Biostatistics from the University of North
Carolina at Chapel Hill, and a PhD in Biostatistics from Johns Hopkins
Prospective, Methods in the literature for missing covariate data in survival
models have relied on the missing at random (MAR) assumption to render
regression parameters identifiable. MAR means that missingness can depend
on the observed exit time, and whether or not that exit is a failure or a
censoring event. By considering ways in which missingness of covariate ~ X
could depend on the true but possibly censored failure time T and the true
censoring time C, we attempt to identify missingness
mechanisms which would yield MAR data. We find that, under various reasonable
assumptions about how missingness might depend on T and/or C, additional strong
assumptions are needed to obtain MAR. We conclude that MAR is difficult to
justify in practical applications. One exception arises when missingness
is independent of T, and C is independent of the value of the missing ~ X. As
alternatives to MAR, we propose two new missingness assumptions. In one, the
missingness depends on T but not on ~ C; in the other, the situation is
reversed. For each, we show that the failure time model is identifiable.
When missingness is independent of T, we show that the naive complete record
analysis will yield a consistent estimator of the failure time distribution.
When missingness is independent of C, we develop a complete-record likelihood
function and a corresponding estimator for parametric failure time models. We
propose analyses to evaluate the plausibility of either assumption in a
particular data set, and illustrate the ideas using data from the literature on
Phase II Selection Design With Adaptive Randomization in a Limited-Resource
Hyung Woo Kim, Ph.D., TGRD
Hyung Woo Kim is currently working at Takeda Global Research & Development
as a Senior Statistician. Before he joined Takeda, he spent six years at MD
Anderson Cancer Center, Fred Hutchinson Cancer Research Center, and
Bristol-Myers Squibb Company working on clinical trials in oncology. He received
MS in Statistics from Iowa State University and PhD in Biostatistics from
University of Texas, School of Public Health. He has published several
statistical papers and co-authored many medical papers.
In clinical research where there are several competing treatments E1,
E2,…, ET of interest, there is limited number of
patients that can participate in clinical trials. With standard approaches of today, such as Simon’s ,
Gehan’s , and Fleming’s  multi stage design, the
choice of a treatment to which patients are assigned with priority will be in
question. In addition, researchers
conduct these trials one at a time. In
case competing treatments are tested simultaneously, they need to compete with
each other to accrue the required number of patients.
We propose a design for testing a null hypothesis H0:
p0 against an alternative hypothesis H1:
p ³ p1 for all competing treatments within one clinical
trial setup. Patients are
randomized to one of the competing treatments in an adaptive randomized fashion.
Initially, the randomization is balanced, and it will shift in favor of
treatments that are performing better. As
a result, a treatment with better performance will have higher patient accrual
rates and be advanced to the next level of trials sooner than other treatments.
We will restrict our attention to the case where the
number of available patients, N, is
limited. Using simulation, we will compare two approaches to phase II
trials. One uses the standard
approach of today, and the other uses adaptive assignment.
We will focus our attention to the number of drugs considered, number of
false positives, number of true positives, proportion of patients who respond,
and the time to find a drug with the best performance.
 Simon, R. ‘Optimal Two-Stage Designs
for Phase II Clinical Trials’, Controlled Clinical Trials, 10,
 Gehan, E. ‘The Determination of The Number of Patients Required in a
Preliminary and a Follow-up Trial of a New Chemotherapeutic Agent’, Journal
of Chronic Disease, 13, 346-353 (1961)
Fleming, T. ‘One-Sample Multiple Testing Procedure for Phase II Clinical
Trials, Biometrics, 38, 143-151 (1982)
Confidence Levels: Theory and Application by Alan
M. Polansky, Ph.D., Northern Illinois University
Alan M. Polansky received his Ph.D. from Southern Methodist University in 1995
under the direction of Dr. William Schucany. He is currently an Associate
Professor in the Division of Statistics at Northern Illinois University.
measure represents the amount of confidence there is that the true parameter
value is in that subset. Such a measure provides a simple method to account for
the inherent variability in the data and to simultaneously consider the
plausibility that the parameter is within each of the regions. This talk
investigates the asymptotic properties of several methods for computing such a
measure for the case of a parameter vector following the smooth function model
using multivariate Edgeworth expansion theory. The applicability of the results
to finite samples is investigated through an empirical study, and the methods
are demonstrated using an example.
Suppose that a sample is taken from a
population whose parameter is a member of a parameter space that has been
divided into a countable set of possibly overlapping subsets. The problem of
regions is concerned with establishing which of these subsets contains the true
parameter value on the basis of the sample. The problem has many applications
that include model selection and classification. A new approach to this problem
is based on assigning a measure to each of the subsets.