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
«  July  2014 »
Events on July 8, 2014
  • Biostatistics Student Presentation
    Starts: 2:00 pm
    Ends: July 8, 2014 - 3:00 pm
    Location: Mayo Memorial Building, room A301
    Description: Masters Candidate in Biostatistics, Ashwini Venkatasubramaniam, will present: "Prediction via Recursive Partitioning: Comparing Conditional Inference Trees to Classification and Regression Trees"

    Recursive binary partitioning or decision trees seek to determine those sub-populations that are particularly associated with a response of interest. A popular method, Classification and Regression Trees (CART) by Breiman et al., utilizes improvements in the Gini index of node impurity or mean squared errors to build a tree. However, this method is prone to over-fitting and displays biasedness towards covariates with large number of splits. An alternative method, Conditional Inference Trees by Hothorn, et al., utilizes a statistical inference based framework. This method seeks to overcome the problems of CART by separating the process of variable selection and determination of splits. We provide a comparison of the two tree types through a simulation study. In several of the examined scenarios, we found the conditional inference tree to have better predictive performance. We also illustrate an application of these tree types to data from a recently completed trial, the Box Lunch Study.

    Refreshments will be served prior to the presentation.
Events on July 17, 2014
  • Biostatistics Student Presentation
    Starts: 1:00 pm
    Ends: July 17, 2014 - 2:00 pm
    Location: Mayo Memorial Building, Room A301
    Description: Masters Candidate in Biostatistics, Fan Wang, will present:" Effect of Different Normalization Methods on Meta-analysis for Illumina Infinium 450k DNA Methylation Data"

    DNA methylation is a biological process that is closely related to many phenotypes and diseases. Since association test between methylation level and trait can be underpowered due to small sample size or effect size, meta analysis serves as a practical tool in methylation data studies, to obtain higher statistical power and reduce false positives. One particular topic of interest in such studies is potential heterogeneity introduced by practical issues. Illumina Infinium HumanMethylation450 (HM450) BeadChip provides capture-associated bias free methylation measurements, with large coverage capacity; and it contains both Infinium I and infinium II assays. Since methylation measures derived from these two probe types exhibit different distributions, many normalization methods have been introduced to reduce this difference. To determine whether different normalization methods amongst studies will introduce heterogeneity in meta analysis, we generated synthetic methylation studies with different normalization methods, using Atherosclerosis Risk in Communities (ARIC) data set, to find primary sources for heterogeneity. Our results show that heterogeneity usually coincides with strong association signal, and that different normalization methods don't have significant effect on heterogeneity. In addition, fixed effect meta analysis yields consistent estimation as pooled analysis. Thus, it suggests using fixed effect model rather than random effect model in meta analysis of methylation data, to increase statistical power.

    Refreshments will be served prior to the presentation.
Events on July 22, 2014
  • Biostatistics Student Presentation
    Starts: 1:00 pm
    Ends: July 22, 2014 - 2:00 pm
    Location: Mayo Memorial Building, Room A301
    Description: Ph.D. Candidate in Biostatistics, Jing Zhang, will present: "Bayesian Hierarchical Methods for Network Meta-Analysis"

    Abstract: Biomedical decision makers confronted with questions about the comparative effectiveness and safety of interventions often wish to combine all sources of data. Network meta-analysis (NMA) expands the scope of a conventional pairwise meta-analysis to simultaneously handle multiple treatment comparisons, synthesizing both direct and indirect information and thus strengthening inference. The most popular methods to date for NMA are contrast-based (CB). They focus on reporting merely odds ratios (ORs), and have serious issues in estimating the event rates that sometimes are essential in decision-making. In this talk I will present our proposed novel arm-based (AB) method in the missing data framework by considering unobserved treatment arms as missing data to be imputed. This method provides more proper and comprehensive reporting summaries and is less-biased compared with the CB method. I will also talk about the extensions of the above method to handle the thorny situations: nonignorable missingness and outlying trials. We incorporated nonignorable missingness with selection models method and proposed several detection measures to diagnose outlying trials. I will illustrate our methods with real data analyses and also simulation studies.

    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