University of Minnesota
Discover Public Health at the University of Minnesota -- one of the top schools of public health in the country, a distinction that reflects our firm commitment to academic and research excellence.
Biostatistics Seminars

The Biostatistics seminar series includes research-focused talks by division faculty and other guests. All seminars are free and open to faculty, students and staff.

Upcoming and Recent Seminars

Date Speaker Affiliation Title
May 7 Abhyuday Mandal University of Georgia Multi-objective Optimal Experimental Designs for Event-related fMRI Studies
May 2 Jeff Gill Washington University Product Partitioned Dirichlet Process Prior Models for Identifying Substantive Clusters and Fitted Subclusters in Social Science Data
Apr 30 Jing Ning MD Anderson Cancer Center Statistical Inference on Right-censored Length-biased Data
Apr 23 Debajyoti Sinha Florida State University Safe Methods and Sample Size Determination for
Non-Inferiority Trial for Survival Response using a Proportional Odds Survival Model
Apr 16 Jianwen Cai University of North Carolina More Efficient Estimator for Additive Hazard Model for Case-Cohort Studies
Mar 26 Michael Rosenblum Johns Hopkins University Optimal Tests of Treatment Effects for the Overall Population and Two Subpopulations in Randomized Trials, using Sparse Linear Programming
Feb 26 Eric Vance Virginia Tech LISA 2020: Creating a Network of Statistical Collaboration Laboratories
Feb 11 Eric Lock Duke University Exploratory Methods for the Integrated Analysis of Multi-Source Data
Feb 7 Fan Yang University of Pennsylvania Using Post-Quality of Life Measurement Information in Censoring by Death Problems
Feb 5 Yingying Wei Johns Hopkins University Integrative Statistical Models for Genomic Signal Detection and Ultra-high-dimensional Prediction
Jan 31 Susan Wei University of North Carolina Latent Supervised Learning
Jan 29 Yong Chen University of Texas – Houston Novel Statistical Tests for Homogeneity in Semiparametric Mixture Models with Application to Methylation Data.
Jan 22 Prabhani K. Don Penn State Finding Latent Structures: Model-based Approaches and Applications in Genomics
Jan 15 Matt McCall University of Rochester Gene Regulatory Network Estimation
Dec 11 Bruce Swihart Johns Hopkins University Sleep Hypnograms, Insurance Claims, & Hand Movement After Stroke: Big Data and Potentially Many Weak, Predictive Signals
Dec 4 Bryan Shepherd Vanderbilt Comparing Different Methods for Assessing When-to-Start Treatment
Nov 20 Sijian Wang University of Wisconsin – Madison Regularized Outcome Weighted Subgroup Identification for Differential Treatment Effects
Nov 6 Doug Schaubel University of Michigan Semiparametric Methods for Survival Analysis of Case-control Data Subject to Dependent Censoring
Oct 30 Seunggeun Lee University of Michigan Rare Variant Analysis in Sequencing-based Association Studies
Oct 23 Rob Weiss UCLA The Markov Chinese Restaurant Process: A Non-parametric Bayesian Cluster Memory Model for Longitudinal Data
Oct 16 Daniela Witten University of Washington Graph Estimation with Joint Additive Models
Oct 2 Yong Chen University of Texas – Houston Regression Analysis of Longitudinal Data with Irregular and Informative Observation Times
Sep 25 John Fricks Penn State Stochastic Modeling and Inference of Molecular Motors across Scales
Sep 18 Nidhi Kohli University of Minnesota Fitting a Linear-linear Piecewise Growth Mixture Model with Unknown Knots: A Comparison of Different Software Procedures
«  August  2014 »
Events on August 4, 2014
  • Biostatistics Student Presentation
    Starts: 10:00 am
    Ends: August 4, 2014 - 11:00 am
    Location: Mayo Memorial Building, room A301
    Description: Masters candidate in Biostatistics, Ryan Tieu, will present: "Two-stage Method for Linear Mixed Model Estimation: Efficient Estimation of Linear Mixed Models for Genome-wide Association Studies"

    This paper analyzes the estimation of SNP (single-nucleotide polymorphism) effects in genetic association studies using a Two-stage method for estimating linear mixed effect models. In genetic association studies, SNPs are analyzed to detect association with phenotypes linked to clinical risk factors. Many of these risk factors are continuous (ie. blood pressure, cholesterol level, body mass index) and measured at multiple times on a given patient. Though these measurements are longitudinal and change over time, the SNP of interest for each subject is time-invariant. Mixed effects models can be used to incorporate this longitudinal data, but estimating thousands or millions of such models can become computationally cumbersome. The method proposed in this paper separates estimation into two stages utilizing the fact that SNPs are time-invariant in the mixed model. In the first stage, the mixed effects model is fitted with only time-variant variables. In this step, the time-invariant SNP effect is not directly estimated, but is captured in the estimation of the random effects. In the second step, the estimated random effects are used as the dependent variable and a simple linear regression can then be performed to estimate the SNP effect. Thus, for a particular phenotype of interest, only one mixed effect model is estimated followed by simple linear regressions for each SNP. Through simulation we show that that the Two-stage method greatly decreases computation time while still being able to detect significant signals coming from associated SNPs.

    Refreshments will be served prior to the presentation.
Events on August 5, 2014
  • Biostatistics Student Presentation
    Starts: 9:00 am
    Ends: August 5, 2014 - 10:00 am
    Location: Mayo Memorial Building, Room A301
    Description: Masters Candidate in Biostatistics, Nicole Bettes, will present: "Using Pseudo Observations to Estimate the Optimal Lung Transplant Type: Single or Bilateral"

    Lung transplantation can involve transplanting either one or both lungs; however, it is unclear which patients would benefit most from which transplant type. We aimed to estimate which transplant type is most beneficial, in terms of restricted survival, for recipients given donor and recipient clinical characteristics. 7,776 Patients with idiopathic pulmonary fibrosis (IPF) and chronic obstructive pulmonary disease (COPD), two native diseases for whom both transplant types are an appropriate treatment, that received lung transplants, single or bilateral, between May 4, 2005 and September 30, 2011 were included in the analysis. Leave-one-out jackknife pseudo-observations were computed to non-parametrically estimate the treatment effect (difference in survival between each transplant type) within each subject and linear regression was implemented to determine the patient and donor characteristics that contributed to a specific transplant type producing better post-transplant survival time. Post-transplant two-year restricted mean survival time was higher for a single lung transplant given a recipient had a lung allocation score below $39$. Covariates influential upon difference in post-transplant survival times between the two transplant types included oxygen delivered and functional status of the patient. Survival was different between the two transplantation types dependent on clinical characteristics and the overall health of the patient.

    Refreshments will be served prior to the presentation.
Events on August 6, 2014
  • Biostatistics Student Presentation
    Starts: 2:00 pm
    Ends: August 6, 2014 - 3:00 pm
    Location: Mayo Memorial Building, Room A110
    Description: Masters candidate in Biostatistics, Zeqing Lu, will present: "Generalized Non-linear Mixed Model for Cigarette Purchase Survey"

    Examining the relationship between tobacco price and consumption helps taxation management, thereby helping promote smoking cessation. This paper proposes a generalized nonlinear mixed effects model to estimate the demand curve for cigarette consumption by analyzing data from a cigarette purchase task (CPT), a self-reported survey of cigarette consumption at escalating levels of price per cigarette. In this CPT study, 308 participants aged from 14 to 65 were selected. Because cigarette consumption is a decreasing function with increasing price, a properly scaled survival function provides a rich class of possible models. We used a scaled Weibull survival function with subject-specific scale and rate parameters to parameterize the subject-specific expected consumption at each price. The self-reported cigarette consumption of course deviates from the subject-specific expected consumption. Therefore, conditioned these subject-specific parameters, the self-reported cigarette consumption was assumed to follow a Poisson distribution. The elasticity (proportionate price sensitivity) of demand, which is related to hazard function, was used to evaluate the effect of price on consumption. This modeling approach also allowed us to estimate the anticipated consumption of cigarettes under price zero, maximum daily total revenue, and price at maximum daily total revenue. For a typical participant in this study, demand for cigarettes was initially at lower prices not sensitive to price changes (inelastic), but at price $0.90, it started to become increasingly sensitive (elastic) as prices increased. Comparing to other approaches, we modeled consumption directly instead of modeling the log consumption. It will be important for future studies to improve accuracy of self-reported responses in a hypothetical situation.

    Refreshments will be served prior to the presentation.
Events on August 8, 2014
  • Biostatistics Student Presentation
    Starts: 10:30 am
    Ends: August 8, 2014 - 11:30 am
    Location: Mayo Memorial Building, Room A301
    Description: Masters Candidate in Biostatistics, Tien Vo, will present " A Bayesian Approach to Estimating the Demand Curve"

    Exponential demand curve analysis, known as behavioral – economic demand analysis offers a great tool to provide conceptual and methodological framework for research in economics, public policies, humans and animal studies. To the best of our knowledge, exponential demand curve parameters have often been estimated using the classical Frequentist approach. However, there are a number of issues with the traditional Frequentist approach, e.g, the variances and distributions of two important measures of the demand curve namely Pmax and Omax can’t be estimated using the Frequentist approach via the delta method. In addition, when the sample size is small, inferences based on t-distribution approximation might be problematic. The purpose of this paper was to propose Bayesian hierarchical models as an alternative means to estimate the demand curve and its parameters such as Pmax and Omax. The performance of the proposed Bayesian hierarchical models was compared with the maximum likelihood method through simulation studies and a case study.
    KEY WORDS: Bayesian hierarchical model; Maximum likelihood; behavioral economics; demand curve analysis.

    Refreshments will be served prior to the presentation.
Events on August 12, 2014
  • Biostatistics Student Presentation
    Starts: 10:00 am
    Ends: August 12, 2014 - 11:00 am
    Location: Mayo Memorial Building, room A301
    Description: Ph.D. Candidate in Biostatistics, Tom Murray, will present: "Hierarchical Models that Flexibly Incorporate Supplemental Information for Settings with Unknown Nonlinear Functions"

    Conventional approaches to statistical inference preclude structures that facilitate incorporation of partially informative supplemental information acquired from similar circumstances. Borrowing strength from supplemental data promises to facilitate greater efficiency in the scientific investigative process, but neglecting to account for heterogeneity across the sources of information obscures understanding of the complex underlying mechanisms that produced the primary data, and may lead to biased inference. As such, inference should derive from flexible statistical models that account for inherent uncertainty while increasingly favoring the primary information as evidence for between-source heterogeneity arises.

    In this talk, I will discuss extensions of existing flexible borrowing methods to settings where the estimation of a curve is of primary interest, and the amount of borrowing reflects congruence in curve shape across sources of information. I address flexible borrowing in the context of a piecewise exponential survival model with respect to the baseline hazard function. I then discuss an alternative piecewise linear log-hazard regression model. Finally, I discuss a general framework for flexible borrowing with respect to a set of parameters that characterize a functional object (e.g. a curve or surface), and apply this framework to functional liver imaging data using a model with Bayesian penalized splines.

    Refreshments will be served prior to the presentation.
Events on August 13, 2014
  • Biostatistics Student Presentation
    Starts: 10:00 am
    Ends: August 13, 2014 - 11:00 am
    Location: Mayo Memorial Building, Room A110
    Description: Masters Candidate in Biostatistics, Jincheng Zhou, will present "A Bayesian Approach to Estimating Causal Vaccine Effects on Binary Post-infection Outcomes"

    To estimate causal effect of vaccine on post-infection outcomes, Hudgens and Halloran (2006) defined a post-infection causal vaccine efficacy estimand based on principal stratification framework using the maximum likelihood estimation method. Extending their research, we propose a Bayesian approach to estimate the causal vaccine effects on binary post-infection outcomes. The performance of the proposed Bayesian method is compared with the maximum likelihood method through simulation studies and two case studies — a clinical trials of a rotavirus vaccine candidate and a field study of a pertussis vaccine. For both of the case studies, the Bayesian approach has provided similar inference with the frequentist analysis. However, simulation studies with small sample sizes suggest that the Bayesian approach provides better convergence, smaller bias and shorter confidence interval length.

    KEYWORDS: Bayesian methods; causal inferences; principal stratification; vaccine effects

    Refreshments will be served prior to the presentation.
Events on August 19, 2014
  • Biostatistics Student PhD Presentation
    Starts: 1:00 pm
    Ends: August 19, 2014 - 2:00 pm
    Location: Mayo Memorial Building, Room 3-100
    Description: PhD Candidate in Biostatistics, Andrew Wey, will present "Estimating Restricted Mean Treatment Effects with Stacked Survival Models".

    The restricted mean difference is a useful and clinically relevant summary measure for comparing the difference in anticipated survival between two groups. With observational data, there may be imbalances in confounding variables between the two treatment groups. One approach to account for such imbalances estimates a covariate adjusted restricted mean difference by modeling the covariate adjusted survival distribution. Traditionally, the survival distribution is related to covariates through a proportional hazards model. However, this assumption may be overly restrictive in many applications and, when violated, can lead to bias in the estimated difference of restricted means. We propose estimating restricted mean differences by modeling the covariate adjusted survival time distribution with stacked survival models. Stacked survival models estimate an optimally weighted average of several survival models by minimizing predicted error. By including parametric and semi-parametric models that extend beyond proportional hazards, stacked survival models consistently estimate the restricted mean treatment effect in a wider range of data generating mechanisms than the traditional proportional hazards model. A Monte Carlo simulation study demonstrates that the new estimator remains nearly as efficient when the survival distribution is related to covariates through a proportional hazards model while improving estimation when the proportional hazards assumption is violated. The proposed estimator is illustrated in post-lung transplant survival between large and small-volume centers with data from the United Network for Organ Sharing.

    Refreshments will be served prior to the presentation.
Events on August 28, 2014
  • Biostatistics Plan B Presentation
    Starts: 12:00 pm
    Ends: August 28, 2014 - 1:00 pm
    Location: 717 Delaware Street, Room 202
    Description: Masters candidate in Biostatistics, Yiwen Zhang, will present "Analysis of Missing Data for a Smoking Cessation Study"

    Missing outcome data are a common and challenging problem in smoking cessation studies. In this paper, we focused on a large college smoking cessation study with missing binary cessation outcomes. In this study, a two-by-two randomized factorial design was applied to examine the effects of extended (4-month) vs. standard (1-month) Quit and Win contests and counseling vs. contact control. Self-report tobacco abstinence through an online survey and urine-verified tobacco abstinence were collected at two main follow-up points, the end of treatment at 4 months and the end of follow-up at 6 months. At either follow-up point, the missing cessation outcomes could occur at two different stages: (1) when the survey was not responded or (2) when the urine sample was not provided among those who self-reported abstinence and were therefore invited for urine verification. To deal with the missing data from this two-stage process, we developed two-stage imputation procedures, which allow the missing status dependent upon the cessation outcome itself (called missing not at random or MNAR) with different levels of dependence. Two special cases of the proposed methods include (1) the widely adopted “missing = smoking” imputation method which corresponds to the situation of a perfect correlation between smoking and missing and (2) the imputation methods based on a missing at random (MAR) assumption which corresponds to a zero smoking-missing correlation within each treatment. As a sensitivity analysis, we also performed a multiple imputation analysis based on the MAR assumption by taking into account baseline covariates in the imputations.

    Key words: binary outcome, imputation, missing data, smoking cessation.

    Refreshments will be served prior to the presentation.
  • © 2014 Regents of the University of Minnesota. All rights reserved.
  • The University of Minnesota is an equal opportunity educator and employer. Privacy