| October 8, 2003 | |
| Title: | Least Squares Based ROC Analysis |
| Speaker: | Zheng Zhang, graduate student UW Biostatistics |
| Abstract: |
I will first introduce ROC curve and its application, then I will present
a least squares based approach to estimate, compare and incorporate
covariate effects into the analysis of ROC curves. Data analysis will be
presented using a pancreatic cancer data set.
|
| October 15, 2003 | |
| Title: | A New Class of Models for the Design and Analysis of Multiple Sclerosis Clinical Trials |
| Speaker: | Rachel MacKay, graduate student UW Biostatistics |
| Abstract: |
Multiple sclerosis (MS) is a debilitating disease of the central nervous
system. Patients experience a range of neurologic symptoms, including
lesions in the brain and spinal cord. Magnetic resonance imaging (MRI)
can detect these lesions, and is a promising new technique for monitoring
disease onset, activity, and progression.
Two major issues concerning MS clinical trials are the development of an adequate model for the observed data, and the determination of necessary sample sizes (number of patients, number of measurements per patient, and frequency of scanning). In this talk, we focus on an MRI outcome, the number of lesions on each scan. Using data from a large, multi-centre trial with threetreatment arms, we first motivate the need for an appropriate longitudinal model. We then present our preliminary work in identifying a model that captures the behaviour of the lesion counts over time, as well as the effect of treatment. We discuss the interpretation of the model, and examine its adequacy for both the placebo and treated groups. We then address the problem of sample size computations based on the assumed model. We conclude with a discussion of some of the issues surrounding the modelling of longitudinal data, along with our ideas for a new class of general models for this purpose.
|
| October 22, 2003 | |
| Title: | Bayesian Data Analysis using BUGS |
| Speaker: | Chuan Zhou, graduate student UW Biostatistics |
| Abstract: |
The advances in computing power and simulation based methodology have
made Bayesian methods popular for analysing complex statistical
problems. The only expsoure to Bayesian data analysis in biostat's
curriculum is professor Wakefield's 570 class, unless you are taking
courses from stat department. In this talk, I will try to give some
introductions to Bayesian inference, MCMC especially Gibbs sampling, and
uses of BUGS software (both Classic BUGS under linux and WinBUGS). The
use of BUGS will be ilustrated through simple examples. The talk is
intended to give you some basic ideas of how to actually carry out
Bayesian data analysis using BUGS program. The technical level is low so
it should be accessible to everyone.
|
| October 29, 2003 | |
| Title: | MCMC Analysis of Complex Traits caused by Multiallelic Loci |
| Speaker: | Elisabeth Rosenthal, graduate student UW Biostatistics |
| Abstract: |
Complex traits are affected by multiple loci. While many analytic methods
used in the analysis of complex traits are based on a diallelic trait
model, many traits are known or suspected to be multiallelic. For example,
APOE, a triallelic locus, plays a role in Alzheimer's disease and
cardiovascular disease. Multiallelism is problematic when alleles are
common, in which case more than one risk allele may be segregating in a
family. Analyses of complex traits are further complicated by the lack of
information concerning the number of underlying loci and the number of
alleles at each trait locus. In order to study common complex traits,
linkage analyses may need to allow for multiple underlying loci each with
multiple alleles. Loki is a program that uses reversible jump MCMC
(RJMCMC), which allows the number of QTL to be a variable in a multipoint
linkage analysis of a quantitative trait. However, an underlying
assumption of Loki is that each QTL is diallelic. This assumption may
reduce the ability to detect and localize underlying QTL. I test Loki's
performance when a trait is caused by a triallelic locus. Traits are
simulated using a real data set consisting of multilocus marker data from
chromosome 19 on four large pedigrees, ranging in size from 48 to 87
individuals. By grouping marker alleles for one locus into a three allele
system and simulating quantitative traits based on this system, I can
incorporate the complexities of chromosomal transmission inherent in real
data while maintaining investigative control over the trait model.
Analysis results vary depending on the amount of dominance and
heritability of the linked trait locus.
|
| November 5, 2003 | |
| Title: | |
| Speaker: | Xuesong Yu, graduate student UW Biostatistics |
| Abstract: |
Association studies of candidate genes or loci are commonly conducted in
the field of genetic epidemiology, using both family or case-control
population designs. In this talk i am going to introduce three newly
developed methods for multiple markers in family design setting: 1)
haplotype analysis, 2) GEE approach, 3)GLM approach
|
| November 12, 2003 | |
| Title: | Analysis of Failure Time Data Under Risk Set Sampling and Missing Covariates |
| Speaker: | Lihong Qi, graduate student UW Biostatistics |
| Abstract: |
Missing covariate data are common in biomedical studies. Inconsistent and
inefficient estimates can be generated by naively discarding subjects
with missing covariate data. In this talk, regression parameter
estimation in the Cox proportional hazards model is considered when
certain covariates are observed for all study subjects and other
covariate data of interest are collected only for a subset. I will talk
about both simple weighted and fully augmented weighted estimators that
use partially incomplete data nonparametrically. The weighted methods are
nonparametric in the sense that they require neither a model for the
missing-data mechanism nor specification of the conditional distribution
of missing covariates given observed covariates. These weighted
estimators are asymptotically consistent and improve the efficiency of the
estimators based on only complete data. In addition, they allow the
missing-data mechanism to depend on outcome variables and observed
covariates, and are applicable to various cohort sampling
procedures, including case-cohort and nested case-control designs.
|
| November 19, 2003 | |
| Title: | Large, Simple Trials and other Mythical Creatures |
| Speaker: | April Slee, graduate student UW Biostatistics |
| Abstract: |
AFFIRM (Atrial Fibrillation Follow-up Investigation of Rhythm Management)
was designed to be a large, simple trial comparing two treatment strategies for
subjects with Atrial Fibrillation. In this 6-year, 4060 subject trial, less
than 5% of the primary data were missing. I would like to share some of the
dedication and innovation of the 400 investigators and study coordinators
who made data collection during this trial successful.
This talk will include some interesting and practical solutions to problems such as loss to follow-up and slow recruitment, as well as the impact of these solutions on power and follow-up duration.
|
| December 3, 2003 | |
| Title: | Semiparametric Marginal Mean Models for Multiple Type Recurrent Events |
| Speaker: | Hao Liu, graduate student UW Biostatistics |
| Abstract: |
Data of multiple type recurrent events are routinely
collected in longitudinal medical studies. Examples include
multiple type infectious episodes among patients after stem
cell transplantation, recurrent wheezing and cough among
patients with bronchial asthma and repeated basal cell and
squamous cell carcinomas of the skin among patients with the
skin cancers.
In this study, we consider a semiparametric model for the regression analysis of the marginal means of two-type recurrent events. To develop some nice estimation procedures, we deliberately introduce a dependence structure between the two types of recurrent events via an analytically tractable Gamma frailty. The proposed estimation procedures derived from the full likelihood are relatively simple. For large sample properties, we allow a general underlying probability that may not belong to the assumed Gamma frailty bivariate Poisson process model. This complicates the technical matters and the usual martingale methods for counting process are no longer valid. We establish consistency and weak convergence of our estimators by the applications of modern empirical process theory. We perform Monte-Carlo simulations to study the finite sample properties of our estimators, and demonstrate that our estimators are relatively efficient compared to the method that analyzes the two-type recurrent event data independently. Finally, we illustrate our method with the data of recurrent wheezing and cough collected from a clinical trial on bronchial asthmas.
|