[CoWy ASA] Agenda - Fall Chapter Meeting - Friday November 7th

[Matt Pocernich]

The chapter fall meeting will be held on Friday November 7th on the
Anschutz Medical Campus in the Education 2 North building,  Room
P28-1303.  This is a wonderful opportunity to visit this amazing
campus which has been created in the last decade.  The selection of
talks is biased towards topics in medical research.  But everyone is
encouraged to attend.  Refreshments will be provided.

Directions are found below.  More specific instructions will be forthcoming.

http://www.uchsc.edu/anschutzmedicalcampus/maps/



Tentative
Fall meeting for the CO/WY Chapter of the ASA

Date:  November 7, 2008
Location:  Ed2 North, P28-1303, AMC Campus
Time:  1:00 to 3:30

Speakers:

Lorri Ogden		A discussion of methods for statistical validation of intermediate
endpoints

Miranda Grote	Simplification of the Freedman method for statistical
validation of
intermediate endpoints

Matt Pocernich	The History and Challenges of Statistics in the Field of Weather
Modification

Brandie Wagner	The Use of Averaged Expression Values for Gene Selection in
Microarray Analysis

Xian Lu	Use of Marginal Structural Models to Estimate the Causal
Effect of Environmental Tobacco Smoke Exposure on Asthma Rescue
Medication Use in Asthmatic Children

Loren Cobb		Statistical Catastrophe Theory

Abstracts:

TITLE: A discussion of methods for statistical validation of
intermediate endpoints
SPEAKER:  Lorri Ogden
AUTHORS: Lorri Ogden, Miranda Grote
AFFILIATION:  Colorado School of Public Health, University of Colorado Denver

A variety of methods have been proposed for testing and quantifying
intermediate variable effects with a binary outcome measure. The
statistical validation approach suggested by Freedman, et. al. (Stat.
Med., 1992) compares coefficients from two models: Model 1 which
estimates the effect of treatment on the outcome unadjusted for the
intermediate variable, and Model 2 which estimates the effect of
treatment on the outcome adjusted for the intermediate variable. The
criticisms of this method for assessing intermediate endpoints are
discussed and alternative strategies based on standardizing the
coefficients from Model 1 and Model 2 (MacKinnon & Dwyer, Evaluation
Review, 1993) or re-defining the total effect (Li, Meredith, &
Hoseyni, Stat. Med., 2001) are described. We recommend a simpler
method for testing the intermediate endpoint effect, along with an
alternative method for calculating the total effect.



TITLE: Simplification of the Freedman method for statistical
validation of intermediate endpoints
SPEAKER:  Miranda Grote
AUTHORS: Miranda Grote, Lorri Ogden
AFFILIATION:  Colorado School of Public Health, University of Colorado Denver

The use of an intermediate endpoint is commonly used in clinical
research to assess treatment effects on an outcome through the
intermediate variable.  These endpoints, also called surrogate
endpoints, are often used to evaluate the primary outcome of a study
at the earliest time possible.  One common method to validate the use
of an intermediate endpoint with a binary outcome is proposed by
Freedman, et. al. (Stat. Med., 1992).  His statistical validation
approach compares beta coefficients for the treatment effect from two
models: Model 1 is the effect of treatment on the outcome unadjusted
for the intermediate variable, and Model 2 is the effect of treatment
on the outcome adjusted for the intermediate variable.  To estimate
the covariance between model 1 and model 2 beta coefficients, Freedman
provides a formula based on a first-order Taylor expansion, which has
also been reported by Buyse and Molenberghs (Biometrics, 1998) using
matrix notation.  We provide a simplification of this formula and
discuss its similarity to a method proposed by Greenland and Mickey
(Appl. Statist., 1988) for testing strict collapsibility in
contingency tables.

TITLE:  The History and Challenges of Statistics in the Field of
Weather Modification
SPEAKER:  Matt Pocernich
AFFILIATION:  National Center for Atmospheric Research

In the 1970's, John Tukey, Jerzy Neyman, William Kruskal and many
other prominent statisticians were actively involved in the statistics
supporting the field of weather modification.  Debates on methods used
by different projects were often contentious. This talk provides a
brief overview the  challenges that faced researchers of that period
and why many of the same questions still remain unresolved nearly
three decades later.  Possibly surpising to some, weather modification
projects still are an active area of research.  This talk will also
describe several active projects.

TITLE:  The Use of Averaged Expression Values for Gene Selection in
Microarray Analysis
SPEAKER:  Brandie Wagner
AFFILIATION:  Colorado School of Public Health, University of Colorado Denver

The commonly used microarray gene selection methods based on
permutation theory are restricted to simple comparisons between groups
and test genes individually, ignoring the opportunity to utilize the
correlation that exists between some genes. The proposed method was
aimed towards addressing this limitation by applying common testing
and selection methods to the average expression of genes clustered
together, rather than at the individual gene level. This approach
lessens the multiplicity issue as well as reduces noise in the
expression values. Genes were assigned to clusters based on their
correlations or the observed similar expression patterns across
subjects. These methods were applied to a microarray dataset
investigating the association of gene expression with schizophrenia
after accounting for smoking status, pH and age, which in previous
studies have been determined to be potential confounders. The results
from this method and from testing each gene separately are compared
using information obtained from independent, previously published
genetic association studies. This investigation suggested an
improvement over testing genes individually.

TITLE:  Use of Marginal Structural Models to Estimate the Causal
Effect of Environmental Tobacco Smoke Exposure on Asthma Rescue
Medication Use in Asthmatic Children
SPEAKER:  Xian Lu
AFFILIATION:  Nurse-Family Partnership (Denver, CO)

Marginal structural models (MSMs) are widely used to obtain causal
effect estimates in observational studies with non-randomized exposure
or treatment. Inverse probability of treatment weight (IPTW)
estimation of an MSM was applied to observational data from an EPA
funded study at National Jewish Medical and Research Center
(2002-2003) involving children at the Kunsberg School. Effects of
Environmental Tobacco Smoke (ETS) exposure on asthma rescue medication
(albuterol) use for children were evaluated, and the results were
compared with those obtained through traditional regression methods.
We obtained an estimated causal albuterol use odds ratio of 3.1 (95%
CI based on empirical standard errors: 0.94, 10.4, p = 0.0636) for ETS
exposure versus non-exposure, whereas the traditional analysis yielded
an odds ratio of 1.97 (95% CI: 0.81, 4.6, p = 0.1240). On the
assumption that our models were correctly specified, these IPTW
estimates will be consistent for the true marginal causal odds ratio
as if data have been obtained from a randomized controlled trial, in
which confounding effects are eliminated.

TITLE:  Statistical Catastrophe Theory
SPEAKER:  Loren Cobb
AFFILIATION:  Visiting Professor, UCD

The so-called "catastrophe" models of differential topology exhibit a
variety of useful but highly nonlinear behaviors: bifurcations, sudden
transitions, hysteresis effects, etc. But the malleability of these
topological models also makes them highly problematic for statistical
analysis, because most statistical models are at best invariant only
up to a linear transformation. However, recent research may have
uncovered a way to introduce a lot more malleability into statistical
versions of these models.

In this presentation I will show how to create stochastic differential
equations for catastrophe models, and how to derive their
corresponding equilibrium probability density functions. The results
are multimodal exponential families which generalize many of the
probability density functions of classical statistics. This extension
of classical statistics opens up large domains of statistical models
with multiple stable states and rich nonlinear dynamics. Finally, I
will show how the recent work of Hartelman & Wagenmakers may make
possible a form of statistical analysis that is invariant under smooth
deformations of the underlying space. The implications for time series
analysis are potentially large, if we can fully integrate these
topological ideas into current statistical theory. There is plenty of
material here for several enterprising and talented PhD students.