|
|
|
|
Title: |
A hierarchical aggregate data model with allowance for spatially correlated disease rates |
|
Speakers: |
Katherine Guthrie, Graduate student, Biostatistics |
|
Abstract: |
The aggregate data study design (Prentice and Sheppard, 1995) aims to estimate exposure effects by regressing population-based disease rates on covariate data from survey samples in each population group. The design is motivated by the need to accurately estimate individual-level associations between exposures with limited within-population variability, such as dietary fat intake, and the risk of chronic diseases. By incorporating individual-level exposure and confounder data, the aggregate data study design can overcome many of the sources of bias that are inherent to ecological studies. In this work, we further develop the aggregate data model in the context of Bayesian disease mapping in order to allow for spatial correlation among disease rates across populations. Disease mapping is a process of describing geographical variation in disease incidence and mortality. Our model differs from the standard disease-mapping model by focusing on the exposure effect, instead of on prediction of disease outcomes. We show how we can integrate the aggregate and disease-mapping models in order to provide an intuitive and generalizable approach to the modeling of spatial effects while retaining the efficiency of the exposure effect estimation. |
|
|
|
|
Title: |
A comparative study of the Genentech dynamic randomization method |
|
Speakers: |
Hao Liu, Graduate Student, Biostatistics |
|
Abstract: |
In a controlled clinical trial, treatment groups of equal size within prognostic groupings are usually desired for both statistical and clinical reasons. Dynamic randomization methods that are based on the patient covariates are usually used to achieved this goal, especially when the total sample size is small, and the number of prognostic grouping is large. In this talk, I review a dynamic randomization method used at the Genentech Inc. The statistical properties of the Genentech dynamic randomization method are studied via simulations. Furthermore, the performance of the Genentech method is compared with three classical randomization methods: Pocock and Simon's method, Efron's method, and the randomly permuted blocks design. Recommendations based on the simulation studies are provided. |
|
|
|
|
Title: |
Semiparametric regression models
and spatial binary data |
|
Speaker: |
Chuan Zhou, Graduate Student, Biostatistics |
|
Abstract: |
Every year large areas of
forest in the |
|
|
|
|
Title: |
Finding a dissertation/thesis advisor |
|
Speaker: |
Panel Discussion |
|
Abstract: |
A panel discussion will be lead by four students, 2 Master's level students and 2 Ph.D. students, on their experiences of finding a thesis/dissertation adviser and their recommendations. The panel will also briefly discus their thesis/dissertation topics to give students an idea of what qualifies as an appropriate topic. The remaining time will be devoted to a question and answer session. |
|
|
|
|
Title: |
Uncertainty in outcomes for survival analysis |
|
Speaker: |
Amalia Meier, Graduate Student, Biostatistics |
|
Abstract: |
Estimation of failure time has mainly been done in the context in which failure is observed with certainty. This project describes a failure time estimation technique that addresses information loss due to imperfect sensitivity/specificity in the outcome measure. Following the methods of Richardson and Hughes (2000) that handle discrete time data with uncertain outcomes, estimation methods are developed that include other types of data. Methods described are appropriate for data with missed visits and/or covariate measures which influence survival rates. The final goal of this project is to adapt methods for interval censoring survival analysis with uncertain outcomes. Most of the methods described make use of the E-M algorithm. This talk is being given in preparation for a general exam the following week. |
|
|
|
|
Title: |
Variability in diagnostic mammography accuracy |
|
Speaker: |
AYingye Zheng, Graduate Student, Biostatistics |
|
Abstract: |
I will present an analysis of correlated ROC data using ordinal regression models. The study is based on diagnostic mammography outcomes from Breast Cancer Surveillance Consortium (1994-2000). The goal of the study is to provide an overall summary of accuracy for diagnostic mammography in a community setting, identify systematic variation in accuracy and characterize random variations due to radiologists. Results from generalized estimating equation, random effect model and Bayesian hierarchical model approaches will be discussed. This is part of my RA work under the supervision of Dr. William Barlow. |
|
|
|
|
Title: |
Screening based on the risk of cancer calculation from
Bayesian hierarchical change-point models of longitudinal markers |
|
Speaker: |
Donna Pauler, Assistant Professor, Biostatistics |
|
Abstract: |
The standard approach to early detection of
disease with a quantitative marker is to set a population-based fixed
reference level for making individual screening or referral decisions. For
many types of disease, additional information is contained in the
subject-specific temporal behavior of the marker, which exhibits a
characteristic alteration early in the course of the disease. In this talk I
present a Bayesian approach to screening based on calculation of the
posterior probability of disease from longitudinal biomarker levels. The
method is motivated by a randomized ovarian cancer screening trial in the |
|
|
|
|
Title: |
Parametric identifiability and related
problems |
|
Speaker: |
Abhijit Dasgupta. Graduate Student, Statistics |
|
Abstract: |
Identifiability is often the first assumption made in methods of statistical inference. I consider the case when we do not have even local identifiability of the model. This results in the information matrix being singular. I present a method of reparametrization so that we can get a transformed parameter set under which the model is at least locally identifiable. I also present a sufficient condition so that the model will be globally identifiable under the new parameters. Often constraints are placed on the parameters so that the constrained model is then identifiable (e.g. ANOVA). The problem of finding the constraints is, in a sense, the dual problem to the reparametrization problem. I suggest a method of construction constraints, and describe what properties constraints need to have to be model-preserving. The objective of reparametrization here is to be able to make inference on the original parameter space. I suggest a method for obtaining likelihood-based confidence regions on the original parameter space. |
Last Modification: