|
|
Title: |
Current research in the department |
Speakers: |
Diana Miglioretti, Jon Wakefield, and Thomas Lumley |
Abstract: |
Selected faculty will present a brief description of their current research. The hope will be to give students who are looking for a thesis or dissertation advisor a clearer understanding of the topics that are being pursued and who is pursuing them. |
|
|
Title: |
Current research in the department |
Speakers: |
David Yanez, Carolyn Rutter, and CY Wang |
Abstract: |
Selected faculty will present a brief description of their current research. The hope will be to give students who are looking for a thesis or dissertation advisor a clearer understanding of the topics that are being pursued and who is pursuing them. |
|
|
Title: |
Semiparametric regression models for recurrent events |
Speaker: |
Hao Liu, Graduate student, Biostatistics |
Abstract: |
In many longitudinal medical studies, the patients may experience repeated occurrences of the same type of event. Examples include recurrent superficial tumors in bladder cancer patients, recurrent pyogenic infections in individuals with chronic granulomatous disease, and repeated pulmonary exacerbation for patients with cystic fibrosis. Complicated situation arises when death or censoring prevents the further observation of the recurrent event, especially when death or censoring is dependent to the recurrent event conditional on the covariate. I will review some existing methods in the literature for this complicated situation. In my dissertation research, we study a semiparametric marginal mean model proposed for the joint inference of the recurrent event and death. The estimation procedure is derived from the full likelihood based on the Gamma Frailty nonhomogeneous Poisson processes. When the Poisson assumption fails but the marginal mean structure holds, asymptotic properties of the MLE estimators will be established. |
|
|
Title: |
Incorporating death into the statistical analysis of
longitudinal health status data |
Speaker: |
Laura Lee Johnson, Graduate student, Biostatistics |
Abstract: |
This is a practice talk for a 10 minute talk I am giving next week. Because of the audience at the conference and the short amount of time, I have removed most of the statistical theory from my talk, so it should be accessible to everyone in the department. The topic: Incorporating Death into the Statistical Analysis of Longitudinal Health Status Data. I will briefly describe the type of data of interest, how this data can be analyzed in cohort and randomized trials, and a few of the different questions that we may want to answer. |
|
|
Title: |
High-throughput SNP genotyping with finite mixture models |
Speaker: |
Bart Burlington, Graduate student, Biostatistics |
Abstract: |
Work in last decade on model-based clustering has highlighted the manner in which a likelihood-based approach leverages cluster shape to make clustering calls. This often involves a multivariate Gaussian distribution for the data-generating components, parameterized according to a spectral decomposition of the variance/covariance matrix. I will show some ongoing consulting work that has taken the approach that cluster shape may sometimes be usefully represented by a semi-parametric mean model. I use a simple measurement error model to separate linear features in a bivariate scatter plot; points are then clustered (i.e. the genotype is called) according to their conditional probabilities of group membership, which is, under conditions, related to the variance-weighted, perpendicular distance of the observed points to the estimated linear features. The data-generating processes are not normal, but a least-squares fit is hard to beat, with respect to speed and accuracy. I will present results for fitting this model to reference data and describe the problems still to be solved. |
|
|
Title: |
Linkage disequilibrium and local recombination rate |
Speaker: |
Michael Li, Graduate student, Biostatistics |
Abstract: |
We already knew the distribution of recombination across the genome is not uniform. But not until recently, with the availability of dense genetic markers (e.g. SNPs) and high-throughput genotyping, were we able to study the variation of recombination rate (in terms of physical distance) on a very fine scale. LD refers to the non-independence of alleles at different sites (read: the correlation between two categorical, often binary, random variables). The extent and distribution of LD in humans is of great interest because LD reflects past biological events (mutation, recombination, etc.) and population history (migration, admixture, etc.). Studies of LD may offer some insight on variations in recombination rate over short distance. On the other hand, knowledge about local recombination rate is very important in interpreting the patterns of LD. In particular, within a population, LD is a function of the local recombination rate which thus provides a summary of LD across multiple sites. In this talk, I'll review current methods for studying local recombination rate. Explanation to relevant genetic background will be given and no genetics knowledge is assumed. It should be accessible to anyone who has not been bored by all the talks about genetics. |
|
|
Title: |
The intensity-score approach to adjusting for confounding |
Speaker: |
Mary Redman, Graduate student, Biostatistics |
Abstract: |
In a recent article on the efficacy of antihypertensive therapy, Berlowitz et al (1998) introduced an ad hoc method of adjusting for serial confounding assessed via an intensity score, which records cumulative differences over time between therapy actually received and therapy predicted by prior medical history. Outcomes are subsequently regressed on the intensity score and baseline covariates to determine whether intense treatment or exposure predicts a favorable response. We supply sufficient conditions for interpreting results of the Berlowitz approach based on a causal model for the effect of a time-varying exposure. We also consider a modified version in which the intensity score records cumulative scaled differences over time between therapy actually received and therapy predicted by prior medical history. This leads to a simple, two-step implementation of G-estimation if we assume a nonstandard but useful structural nested mean model which implies that subjects less likely to receive treatment are more likely to be helped by it. These modeling assumptions might apply, for instance, to certain health services research contexts in which differential access to care is a primary concern. We also extend the methods to accommodate repeated outcomes and time-varying effects of time-varying exposures. |
|
|
Title: |
Hierarchical clustering for gene expression array data |
Speaker: |
Chuan Zhou, Graduate student, Biostatistics |
Abstract: |
The advent of microarray technology has enabled biologists to measure the expression levels of thousands of genes in parallel. These experiments have generated enormous amounts of data, and raised many exciting questions in experimental design and data analysis. We propose a general framework which deals specifically with time course data. This framework includes filtering using Bayes factors, clustering based on hierarchical models, inferences through MCMC, and prior information incorporation. Difficulties, (dis)advantages and potential generalizations will also be discussed. I will try to keep the theories at minimal, so the talk should be accessible to everyone. As this is still work in progress, questions and suggestions are very much welcome. |
|
|
Title: |
Complex traits and reduced penetrance |
Speaker: |
Elizabeth Rosenthal, Graduate student, Biostatistics |
Abstract: |
Current statistical genetics methods have low
power in the case of complex disease. Parametric tests make simplifying
assumptions that may not be true and non-parametric tests do not perform well
under certain alternatives. I will discuss what properties make a genetic
disease complex. I will also present
simulation results that illustrate some of the problems encountered, with a
focus on reduced penetrance. Some knowledge of
genetics may be helpful, although I will give a brief overview of the
genetics pertinent to the talk. |
Last Modification: