Spring 2021

Organizer: Huiyi Lin, PhD


Date and Time: Jan. 25 (Mon), 2021; 3-4 pm
Speaker: Jung-Ying Tzeng, PhD
Affiliation: Professor, Department of Statistics and Bioinformatics Research Center, North Carolina State University
Fellow, American Statistical Association
Title: Gene-Set Integrative Analysis of Multi-Omics Data Using Tensor-based Association Test
Location: Zoom meeting

Abstract: Multi-omics analyses integrate complementary levels of information from different molecular events, and have great potential in detecting novel disease genes and elucidating disease mechanisms. One major focus of integrative analysis has been on identifying gene-sets associated with clinical outcomes, and a common strategy is to regress clinical outcomes on all omics variables in a gene set. However, such joint modeling methods encounter the challenges of high-dimensional inference. We introduce a tensor-based framework to enhance model efficiency for variable-wise inference. By accounting for the inherent matrix structure of an individual’s multi-omics data, the proposed tensor methods naturally incorporate the relationship among omics effects, reduce the required number of parameters, and boost the efficiency of high-dimensional modeling. We study the variable-specific testing procedure under tensor regression and enhance computational efficiency of omics tensor modeling. We evaluate the performance of the tensor-based test using simulations and real data application on the Cancer Cell Line Encyclopedia.


Date and Time: Mar. 15 (Mon), 2021; 3-4 pm
Speaker: Bhupendra Rawal, M.S.
Affiliation: Associate Director, Biostatistics, Immunocore
Title: Biostatistician Pursuing Clinical Trials: Research and Industry Experiences
Location: Zoom meeting

Abstract: In this presentation, we will talk about experiences while working as a biostatistician focusing on clinical trials and drug development. This talk will be an interactive discussion which highlights the importance of a biostatistician’s work in collaborative clinical research and how a biostatistician contributes to solving public health problems within a team of clinical scientists and public health professionals.


Date and Time: Mar. 22 (Mon), 2021; 3-4 pm
Speaker: Jared Huling, Ph.D.
Affiliation: Assistant Professor, Division of Biostatistics, School of Public Health,
University of Minnesota
Title: Energy Balancing of Covariate Distributions for Estimation of Causal Effects
Location: Zoom meeting

Abstract: Bias in causal comparisons has a correspondence with distributional imbalance of covariates between treatment groups. Weighting strategies such as inverse propensity score weighting attempt to mitigate bias by either modeling the treatment assignment mechanism or balancing specified covariate moments. In practice, these approaches can be highly sensitive to modeling decisions. This talk introduces a new weighting method, called energy balancing, which instead aims to balance weighted covariate distributions. By directly targeting distributional imbalance, the proposed weighting strategy can be flexibly utilized in a wide variety of causal analyses, including the estimation of average treatment effects and individualized treatment rules. Our energy balancing weights (EBW) approach has several advantages over existing weighting techniques. First, it offers a model-free and robust approach for obtaining covariate balance that does not require tuning parameters, obviating the need for modeling decisions of secondary nature to the scientific question at hand. Second, since this approach is based on a genuine measure of distributional balance, it provides a means for assessing the balance induced by a given set of weights for a given dataset. Finally, the proposed method is computationally efficient and has desirable theoretical guarantees under mild conditions. The effectiveness of our approach is demonstrated in the analysis of two real-world applications, the first a study in the safety of right heart catheterization and the second a study of the effectiveness of a transitional care intervention at a large Midwestern academic medical center.


Date and Time: Mar. 29 (Mon), 2021; 3-4 pm
Speaker: Rajabather Velu, Ph.D.
Affiliation: Adjunct Professor, Biostatistics Program, LSUHSC
Title: Reduced-Rank Regression Models for Longitudinal Data Analysis
Location: Zoom meeting

Abstract: In an influential paper, Anderson (1951) introduced the reduced-rank regression model that incorporates constraints on the regression coefficient matrix, not known apriori but implied by the data. The model is rich in uncovering relationships that are latent. In the time series context, where the traditional dimension reduction techniques cannot be directly applied, the method of reduced-rank regression provides a natural way to do canonical analysis. In this talk I will focus on yet another application in the context of seemingly unrelated regression (SUR) which was introduced by a landmark paper by Zellner (1962).We discuss the estimation and inference aspects of the extended model and illustrate the methodology with an application on scanner data in marketing.


Date and Time: Apr. 5 (Mon), 2021; 1-2 pm
Speaker: Soham Mahato
Affiliation: PhD student, Biostatistics Program, LSUHSC
Title: Bayesian Mixtures of Beta Distributions for Biostatistics (PhD dissertation prospectus defense)
Location: Zoom meeting

Abstract: Bayesian methods can be used to deal with a range of statistical problems. Clinical trial designs and selecting important variables for a model are couple of them. Bayesian methods are used by incorporating prior information from clinicians and providing flexible semiparametric frameworks for clinical trial design and inference as well as for other complex clinical data. Three projects are discussed, and Beta prior is used in all of them – the first and second projects are dose finding for phase I clinical trial designs, the third is a Bayesian framework to investigate how patient compliance is dependent on patient specific covariates and which of those covariates’ effects compliance status in a patient. The continual reassessment method (CRM) is a popular dose-finding design for phase I clinical trials. This method requires that practitioners prespecify the toxicity probability at each dose. Such pre-specification can be arbitrary, and different specifications of toxicity probabilities may lead to very different design properties. To overcome the arbitrariness and further enhance the robustness of the design, we propose a Beta prior for each dose level parameterized by mean and effective sample size of the distribution. This method has been extended in the second project to include multiple pre-specified toxicity probabilities. For the third project, we will use compliance data for pregnant women with gestational diabetes and will try to find out which patient level characteristics effects their compliance behavior using mixture model and Bayesian variable selection.


Date and Time: Apr. 7 (Wed), 2021; 3-4 pm
Speaker: Erick D. Nguyen
Affiliation: MS student, Biostatistics Program, LSUHSC
Title: Exploration of Propensity Score Adjustment in Logistic Regression via Simulation Study (Master’s thesis defense)
Location: Zoom meeting

Abstract: Using propensity scores as covariates can control the effect of confounders in observational studies. However, the methods of variable selection for propensity score modeling are still under debate. To gain insight on the variables that should be used in the propensity model, a simulation study with randomly generated scenarios was conducted to examine confounding variables with varying effect sizes on exposure and outcome. We found that inclusion of variables related to one (exposure or outcome), but not related to or weakly related to other has little to no detrimental effect on the propensity model. Therefore, we recommend including all possible confounders as there is no appreciable negative affect on exposure effect estimation with unneeded variables.


Date and Time: Apr. 12 (Mon), 2021; 3-4 pm
Speaker: Chaoyi Zeng
Affiliation: MS student, Biostatistics Program, LSUHSC
Title: Detecting SNP-Alcohol Interactions Associated with Prostate Cancer Aggressiveness using SNP-Environment Interaction Pattern Identifier (SNPxE)
(Master’s thesis defense)
Location: Zoom meeting

Abstract: Previous studies mainly focused on individual effects of single nucleotide polymorphisms (SNPs) or alcohol intake associated with prostate cancer (PCa) aggressiveness, whereas the impact of their interplay was understudied. This study aims to examine whether SNPs would modify alcohol’s effect on PCa aggressiveness. We tested interactions between 175 SNPs and four alcohol types using SNP-Environment Interaction Pattern Identifier (SNPxE) in 5,038 PCa patients. We identified two SNP-alcohol pairs involved with KLK3 (rs266876 -beer and rs4802755-beer) that were significantly associated with PCa aggressiveness. Our study demonstrated that SNPxE is a powerful tool to detect gene-environment interactions. We observed that the impact of alcohol intake on PCa aggressiveness varies among genetic sub-groups in KLK3, PRKCA, ROBO1, CAMK2D, and ADH family genes. We identified PCa patients with rs266876 GG genotype or heavy beer intake, and PCa patients with the rs4802755 GA or AA and moderate beer intake were high-risk groups of PCa aggressiveness.


Date and Time: Apr. 14 (Wed), 2021; 9-10 am
Speaker: Ravan Moret
Affiliation: MS student, Biostatistics Program, LSUHSC
Title: Analysis of the effects of adjusting for binary non-confounders in a logistic regression model: a simulation study
(Master’s thesis defense)
Location: Zoom meeting

Abstract: Logistic regression models are used in many professional areas including business, marketing, research, etc. Logistic regression estimates the probability of a binary event, $Y$, based on a set of covariates. Researchers attempt to adjust for confounders (i.e. associated with exposure and outcome) however the ability to always adjust for all confounders is not always possible. Inappropriately adjusting for non-confounders, over-adjusting, is always a major concern in terms of how inference is carried out on covariate effects. The effects of omitting and adjusting for confounders have been intensely studied. We instead focused on the potential effects caused by adjusting for additional non-confounding binary covariates after all true confounders have been added in a logistic regression. In this thesis we perform a simulation study to investigate the effects of adding non-confounders to a logistic regression model on estimating binary exposure effect. We analyzed these effects over 1k randomly generated simulation scenarios with sample sizes of 1,000 and 10,000, and 30 and 50 covariates.


Date and Time: Apr. 19 (Mon), 2021; 3-4 pm
Speaker: Ruofei Du, Ph.D.
Affiliation: Assistant Professor, Department of Biostatistics, Univ. of Arkansas for Medical Sciences
Title: An approach for appropriately including observations affected by LOD issue in a least squares estimate based regression analysis
Location: Zoom meeting

Abstract: A limit of detection (LOD) is a threshold value below which a measure is considered too small to be reliably quantified. For handling LOD caused missing data, commonly applied approaches in association studies either utilize only the complete-case observations which then have to bear a limited sample size for analyses, or substitute those ≤ LOD by a unique value (e.g. LOD/) which however yield biased estimates. Within a LSE based regression analysis, we propose a data-driven method to substitute the missing values due to ≤ LODs and the Bootstrap-based variance estimate to facilitate the hypothesis testing of the association effects. The conducted simulation study shows the proposed approach is able to generate unbiased estimate, and has higher detection power compared to the complete-case approach.

Fall 2020 Archive
Spring 2020 Archive
Fall 2019 Archive
Spring 2019 Archive
Fall 2018 Archive