Winter 2005

January 5, 2005
Title:	Comparing outcomes that only exist in a group chosen after randomization
Speaker:	Bryan Shepherd, Biostatistics graduate student
Abstract:	In many experiments researchers would like to compare between treatments an outcome that only exists in a subset of participants selected after randomization. For example, in preventive HIV vaccine efficacy trials it is of interest to determine whether randomization to vaccine causes lower HIV viral load, a quantity that only exists in participants who acquire HIV. I’ll talk about some of the challenges of making these comparisons and propose sensitivity analysis methods using causal inference techniques.
January 12, 2005
Title:	A significance method for time course microarray experiments
Speaker:	Jeff Leek, Biostatistics graduate student
Abstract:	The human genome project has shown that we have far fewer genes than originally expected. To account for our complexity much effort has turned to understanding the regulation of gene expression, that is when and to what extent genes are “turned on” or “turned off”. One powerful approach for gene expression analysis is through DNA microarray experiments, which measure expression for thousands of genes simultaneously. A number of methods have been developed to determine significant changes in expression for genes in distinct static biological conditions, but there has been a relative dearth of methods for timecourse studies. I will briefly introduce DNA microarrays and present a straightforward, computationally efficient, and powerful methodology for the analysis of timecourse microarray experiments.
January 19, 2005
Title:	Adjusting for Covariate Effects in Biomarker Studies Using the Subject-Specific Threshold ROC Curve
Speaker:	Holly Janes, Biostatistics graduate student
Abstract:	Recent scientific and technological innovations have produced an explosion of potential biomarkers which are being investigated for their use in disease screening and diagnosis. In evaluating these new markers, it is often necessary to account for covariates which are associated with the biomarker of interest. For example, age is strongly associated with prostate-specific antigen (PSA), a biomarker for prostate cancer, and the discriminatory accuracy of PSA may also vary with age. We propose the subject-specific threshold ROC (SST-ROC) as a covariate-adjusted measure of the diagnostic accuracy of the biomarker. The SST-ROC is the ROC curve for a rule which uses covariate-specific thresholds to define “test-positive”. It can also be interpreted as a weighted average of the covariate-specific sensitivities, holding the covariate-specific specificities constant. We motivate consideration of the SST-ROC, propose non-parametric and semi-parametric estimators, provide asymptotic distribution theory for these estimators, and explore the implications for efficient study design.
January 26, 2005
Speaker:	Panel Discussion
Abstract:	This week we will have a panel of upper level biostat students sharing their stories and answering questions about choosing an advisor, thesis/dissertation topics, and whatever else comes up! We encourage all students to come and learn from others’ experiences, and to ask the questions you always wanted to ask!
February 2, 2005
Title:	The Clustering of Regression Models Method with Applications in Gene Expression Data
Speaker:	Li-Xuan Qin, Biostat graduate student
Abstract:	Identification of differentially expressed genes and clustering of genes are two important and complementary objectives addressed with gene expression data. Methods for differential expression analysis can generally be characterized as using a regression function to independently model the observations for each gene; various adjustments for multiplicity are then used to interpret the statistical significance of these per-gene regression models over the collection of genes analyzed. Motivated by this common structure of per-gene models, we propose a new model-based clustering method — the clustering of regression models method, which groups genes that share a similar relationship to the covariate(s). This method provides a unified approach for a family of clustering procedures and can be applied for data collected with various experimental designs. The proposed methodology was applied to three real microarray datasets. This work is done under the supervision of Dr. Steve Self.
February 9, 2005
Speaker:	Michael LeBlanc, Ph.D., FHCRC and UW Biostatistics
February 16, 2005
Title:	Multivariate Survival Analysis for Case-Control Family Data
Speaker:	Li Hsu, Ph.D., Fred Hutchinson Cancer Research Center
Abstract:	Multivariate survival data arise when data consist of clusters in which the failure times may be dependent. Population-based case-control family data are one such example. However, it is challenging to obtain a consistent estimator for the marginal hazard function, because the data are complicated not only by the correlations among family members but also by the non-cohort study design. In this talk, I will describe the data structure and present some ongoing work. I will also talk about some of the issues in the methodologic development for non-cohort family data, e.g. generally unavailable large-sample theory and rather restrictive multivariate survival models.
February 16, 2005
Title:	Environmental Dose Reconstruction and Radiation Epidemiology: an Uncertain Connection
Speaker:	Ken Kopecky, Ph.D. Ph.D., Fred Hutchinson Cancer Research Center
Abstract:	Our knowledge of the health effects of exposure to ionizing radiation relies heavily on epidemiological studies of exposed but otherwise normal populations, e.g., Japanese A-bomb survivors, residents of fallout-contaminated areas (Chernobyl, Hanford, Nevada Test Site, Marshall Islands, Kazahkstan, ). To be as useful as possible, such studies require information about individual radiation doses to the target organs of study participants. However these doses were not measured: they must be estimated. Environmental dose reconstruction can be defined as the development and implementation of models for estimating levels of exposure people receive from potentially dangerous materials in the environment. Radiation dose reconstruction models are fraught with uncertainties that cannot be reduced and are in many settings quite complex (e.g., combining both classical and Berkson errors). This talk will describe the approaches for reconstructing radiation doses and their uncertainties, and the implications of those uncertainties, in two recent studies of populations exposed to iodine-131: the Hanford Thyroid Disease Study and the ICRHER Thyroid Cancer Case-Control Studies.
February 23, 2005
Title:	Challenges, statistical and otherwise, in a cooperative clinical trial group setting
Speaker:	Jackie Benedetti, Ph.D. UW and Fred Hutchinson Cancer Research Center
March 2, 2005
Title:	Dynamic treatment regimes
Speaker:	Erica Moodie, Biostatistics graduate student
Abstract:	A dynamic regime is a function that takes treatment and response history and baseline covariates as inputs and returns a decision to be made. Robins (2004) and Murphy (2003) have proposed models and developed semi-parametric methods for making inference about the optimal regime in a dynamic trial that provide clear advantages over the traditional approach of dynamic programming. I will show that Murphy’s model is a special case of Robins’ and that the methods are in some cases equivalent; in doing this, I show that Murphy’s estimates are not efficient. Interesting features of the methods are highlighted using a recent clinical study.
March 9, 2005
Title:	Estimating Treatment Effect in Randomized Clinical Trials with Non-compliance
Speaker:	Leslie Taylor, Biostatistics graduate student
Abstract:	Well-designed randomized clinical trials are a powerful tool for investigating causal relationships and producing valid estimates of a causal effect of treatment. But in trials involving human subjects there are oftentimes problems of non-compliance which standard analyses either ignore, which can lead to biased estimation, or account for in such a way that the estimand can no longer be considered a causal effect. Rubin developed an approach to causal inference using potential outcomes (Rubin, 1974, 1978; Holland, 1986) which has been referred to as the Rubin Causal Model (Holland, 1986). In the setting of a randomized clinical trial with cross-over non-compliance, this talk will go over the standard approaches to estimating treatment effect and then introduce the Rubin Causal Model.