|
|
|
Title: |
Day in the Life of
a Masters Statistician |
|
Speaker: |
Katie Gower and Carolyn Noonan, former Biostatistics
Masters Students |
|
Abstract: |
Carolyn Noonan received her MS in Biostatistics in 2001 and now works
for the UW in the Department of Dental Public Health Sciences. She is one of six statisticians/consultants that make up the Biometry Core for that department. The Biometry Core
provides statistical support and assistance regarding study design and implementation for the majority of the
trials that exist within the responsibilities as part of the Biometry Core, including: statistical analysis of a variety of trial data, consulting with researchers
regarding appropriate analysis techniques and proper study design, database
design and management, and teaching statistical computing to international scholars. Katie Gower received her MS in Biostatistics in 2001 and now works
for the responsible for the management and implementation of two large cancer prevention trials: the Prostate Cancer Prevention Trial (PCPT) and
the Selenium and Vitamin E Cancer Prevention Trial (SELECT). To provide a snapshot of what statisticians do in the field of cancer prevention,
she will discuss her specific responsibilities for these studies,
including: monitoring accrual, monitoring adherence, developing forms, creating reports, running analyses, reviewing study protocol, giving
presentations, writing papers and providing general statistical support to study
staff. |
|
|
|
Title: |
|
|
Speaker: |
Alan Dabney,
Biostatistics Graduate Student |
|
Abstract: |
Ecological studies consist of areal summaries of outcome and
predictor variables. Disease incidence
or mortality counts by geographic region is a common example. Due to the
ready availability of such data (SEER, NCI, DOH all have disease count data on their websites), ecological studies
are used frequently. Though convenient,
ecological bias and spatial, temporal dependence must be considered. The primary endpoint used in the analysis of ecological data is the standardized mortality/morbidity ratio (SMR). For rare diseases (i.e. cancer), a Poisson model is often assumed for incidence or mortality
counts. Various methods (Bayesian, in particular) can then be employed to
handle spatial and temporal dependence; in the Bayesian approach,
hierarchical models with structured and unstructured random effects are used. There are also specific methods for reducing ecological bias; "aggregate
studies", using survey information on randomly selected individuals within each
area, is one of them. In addition to these established methods, there are many problems
still left to solve in this setting. One
such problem is the modeling of spatio-temporal interactions, appropriate since it would be expected
that trends in space change over time.
Another interesting problem is the joint modeling of two diseases.
"Shared component models", specifying a separate random effects model for each disease but allowing for a common
random effect between the two, have been proposed for this purpose but are
largely unexplored. Also, it has
recently been suggested that one model survival times (rather than disease counts) over space. |
|
|
|
Title: |
Forensic Breath
Alcohol Analysis in the Context of Measurement Error |
|
Speaker: |
Rod Gullberg, Graduate
Student, Biostatistics |
|
Abstract: |
Alcohol impaired driving remains a major threat to traffic safety in year for driving while under the influence (DUI). Forensic breath alcohol testing provides significant evidence in the prosecution of arrested drivers. The Washington State
Patrol maintains all breath test instruments used within the state in addition to collecting and
managing a database of analytical and demographic data. The presentation will begin with a description of breath alcohol
testing methodology and a brief summary of several descriptive measures from
the DUI database. Next, we discuss
several scientific and statistical challenges to the program that are routinely experienced in
court. In many cases, these challenges focus on the uncertainty in measurement results, particularly when the results are near one of the critical
limits (e.g., 0.02 for minors, 0.04 for commercial drivers, 0.08 for all
others, and 0.15 for enhanced penalties). Ideally, the breath alcohol evidence will be accompanied by some
estimate of their uncertainty, which often takes the form of a confidence
interval. Reliable estimates of a confidence interval, however, require a
reliable estimate of the standard deviation.
Similar to many other analytical methods, breath alcohol analysis appears to have measurement error
that is proportional to concentration.
Duplicate breath alcohol data from each individual can be analyzed in order to obtain variance estimates throughout the range of concentrations. Identifying the appropriate measurement error model that best
describes forensic breath alcohol measurement is the primary focus of this
work. Actual field duplicate breath alcohol data from the year 2000
(n=28,321 subjects) is evaluated for its error structure by employing ordinary
least squares (OLS), double generalized linear models (DGLM) and standard approaches of evaluating differences squared. These analyses are also applied to data simulated from additive and multiplicative error
models and illustrate the importance of knowing the underlying error
structure when estimating variances. The results show that breath alcohol measurement has a non-constant measurement error that increases with concentration and appears to generally conform to a multiplicative error model. These results have important forensic implications in the prosecution of DUI cases. |
|
|
|
Title: |
Marginalized
Regression Models for Categorical Longitudinal Response Data |
|
Speaker: |
Jonathan
Schildcrout, Graduate Student, Biostatistics |
|
Abstract: |
The goal of this seminar is to discuss the analysis of categorical longitudinal response data when the focus is on estimating marginally specified parameters. I will first differentiate between marginal and conditional means and parameters, and give examples of situations
under which marginal model parameters are likely to be of scientific
interest. Then I will discuss situations under which parameter estimates using
a likelihood-based procedure might be more desirable than ones using a semi-parametric one (e.g., Generalized Estimating Equations,
Alternating Logistic Regressions).
Finally, I will introduce a new class of models, called marginalized models.
These models allow for likelihood inference in the marginal regression setting. |
Feb 19, 2003 |
|
|
Title: |
Bayesian design of
clinical trials and Semiparametric
approaches to inference in joint models for longitudinal and time-to-event
data |
|
Speaker: |
Lurdes Inoue,
Ph.D. Department of Biostatistics, UW Xiao Song,
Ph.D. Department of Biostatistics, UW |
|
Abstract: |
Lurdes and Xiao recently joined our department's faculty. The purpose of this seminar is to meet them and to get a brief idea of the research
they are working on. Lurdes
received her Ph.D. in 1999 from and is doing some work with FHCRC.
She will talk about Bayesian design of clinical trials. Xiao is
currently working at CHS and HPTN. She
got her Ph.D. in 2002 from approaches to inference in joint models for longitudinal and time-to-event data. |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Title: |
Weighted
Estimators for Proportional Hazards Model under Generalized Case/Control
Sampling |
|
Speaker: |
Lihong Qi,
Graduate Student Department of
Biostatistics, UW |
|
Abstract: |
In this talk, I will introduce the generalized case/control sampling
and discuss some weighted estimators for the Cox proportional hazards
model under this sampling scheme. These weighted estimators are also applicable to case-cohort and nested case-control designs. I will present the asymptotic properties of the weighted estimators and discuss their asymptotic relationships under a special situation. This is part of my dissertation work supervised by Dr. Ross Prentice
and Dr. C.Y. Wang. |
|
|
|
Title: |
Estimating
Lifetime Medical Costs Using a Joint Frailty Model of Survival Time and Cost as a
Mark Variable |
|
Speaker: |
Kristen Berry,
Graduate Student Department of
Biostatistics, UW |
|
Abstract: |
The analysis of lifetime medical costs with censored data presents statistical challenges. The assumption of independent
censoring may be valid on the time scale, but is not reasonable on the cost scale. The censoring pattern on the cost scale is typically
induced to be dependent. Of more concern is the fact that the cost distribution is potentially nowhere identifiable in a nonparametric setting
owing to the censoring. Methods to date have avoided this problem by arbitrarily estimating
costs only up to the time of the final failure. We propose a semi-parametric joint gamma frailty model for costs and survival. This model assumes a common frailty for an individual's costs and survival time. We will develop maximum likelihood estimates (MLE) for baseline hazards and
the gamma frailty parameter using the E-M algorithm. These MLE estimates can be combined to obtain the marginal cost distribution and mean. We will discuss the existence and consistency of these
estimates. We will also present results of these methods as applied to both simulated and
real data. |
|
Mar 12, 2003 |
|
|
Title: |
Placebo-Controlled Trials for Surgical Procedures |
|
Speaker: |
April Slee, Graduate Student
Department of Biostatistics, UW |
|
Abstract: |
In the era of evidence- based medicine, randomized double blind placebo-controlled clinical trials have become the gold standard for evaluating the safety and efficacy of new medical
treatments. Although there is some controversy surrounding the use of a placebo arm, this design has been widely accepted because, as long as the placebo is
truly inert, treatment with a placebo poses similar risks to patients as receiving no treatment at all. When there is no
evidence of an effective treatment, a placebo-controlled trial can reduce biases from
investigators and patients. In surgical procedures, however, risks inherent in
surgeries often render a placebo-controlled (sham surgery) trial unethical. Unlike many placebos for medical treatments, a sham surgery certainly
carries more risks than no treatment at all. Recently, especially in neurology, the potential for bias resulting from non-placebo-controlled trials
has caused some investigators to use a sham surgery design. I would l! ike to outline the history of placebo-controlled trials for surgical
procedures, and to share the insight of a few investigators who have recently conducted such trials. I would also like to describe
guidelines set by FDA and by our internal review board to attempt to simultaneously minimize risks to patients and to evaluate a new surgical treatment as objectively. |
Last Modification: 22 January 2003