September 12, 2024
Speaker: Ahmedin Jemal, DVM, PHD
Using Surveillance Data to Promote Cancer Prevention and Control: Examples from the American Cancer Society
Abstract: Descriptive study (surveillance research) is not taught in most public heath graduate schools despite its importance in informing etiologic hypothesis and setting priorities for and monitoring the effectiveness of disease control and prevention. In this lecture, I will discuss the use of surveillance data (research) to promote cancer prevention and control using examples from the American Cancer Society.
October 23, 2023
Speaker: Dr. Ruth Pfeiffer
An Introduction to Risk Prediction
Abstract: First we define notions of “risk;” a clinically very relevant quantity is absolute (or “crude”) risk, the probability that an individual who is free of a given disease at an initial age, a, will develop that disease in the subsequent interval (a, t]. Absolute risk is reduced by mortality from competing risks. Models of absolute risk that depend on covariates have been used to design intervention studies, to counsel patients regarding their risks of disease and to inform clinical decisions. Methodological issues relevant to the development and evaluation of absolute risk prediction models will be discussed, and various study designs and data for model building will be presented, including cohort, nested case-control, and case-control data combined with population registry data. Issues relating to the evaluation of risk prediction models and the strengths and limitations of risk prediction models for various applications will be discussed. Standard criteria for model assessment will be presented, as well as loss function-based criteria applied to the use of risk models to screen a population and the use of risk models to decide whether to take a preventive intervention that has both beneficial and adverse effects. Various applications will be discussed. If time permits, model updating and model validation with missing predictors will be addressed.
May 2, 2023
Speaker: Dr. Matthew Wheeler
Bayesian Model Averaging and ToxicR
Abstract: The World Health Organization and European Food Safety Authority have recommended that researchers and regulators use Bayesian model averaging for dose-response analyses. Additionally, BMD-Express allows for using Bayesian model averaging for gene expression data. With the proliferation of these approaches, it is difficult for practitioners to understand what methods to use in their research. This talk explains Bayesian model-averaging and Bayesian dose-response analysis, showing why these methods are superior to traditional maximum likelihood approaches. All examples are done using the R dose-response package ToxicR, and reproducible code is given to allow participants to gain intuition with Bayesian dose-response analyses in ToxicR.
March 9, 2023
Speaker: Dr. Dimitris Rizopoulos
Dynamic Risk Predictions from Joint Models, with Applications in R
Abstract: This workshop focuses on data collected in follow-up studies. Outcomes from these studies typically include longitudinally measured responses (e.g., biomarkers) and the time until an event of interest occurs (e.g., death, dropout). The aim often is to utilize the longitudinal information to predict the risk of the event. An important attribute of these predictions is that they have a time-dynamic nature, i.e., they are updated each time new longitudinal measurements are recorded. In this workshop, we will introduce the framework of joint models for longitudinal and time-to-event data and explain how it can be used to estimate and evaluate such dynamic risk predictions. We will use the R package JMbayes2 to showcase the capabilities of these models.
September 29, 2021
Speaker: Professor Richard Smith
Title: Climate Change, Extremes, and Risks
Introduction: 2021 has been the year that climate change finally became a subject everyone was talking about. A series of extreme climate events have covered the US and Canada, many parts of Europe, and other parts of the world. The Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) delivers dire warnings about what will happen if we fail to curb greenhouse gas emissions quickly. An international conference will take place in Britain in November where world leaders including President Biden will be expected to reveal their plans for action. So, where do statistics and data science fit into this picture?
It has often been stated that we cannot attribute a single event, such as the recent extreme heat conditions in the Pacific Northwest or the wildfires affecting many parts of the world, to climate change. What we can calculate is how the probability of such an event, or the size of the event given its occurrence, may change as a result of greenhouse gas emissions compared with the atmospheric conditions that existed 200 years ago. First, we need to define the event itself, for example, that the average temperature over a specific region of space and time exceeded a certain threshold level. Second, we can estimate the probability of such an event by studying historical records and comparing them with climate model output, in effect, computer simulations of climate under both present-day and historical conditions. Extreme value theory is the branch of statistics concerned with estimating probabilities of extreme events, and is widely used to characterize probabilities of extreme weather events. However, that theory itself raises many questions about the appropriate choice of distribution, method of estimating parameters, and how to account for uncertainty.
This talk will introduce these concepts to statisticians and data scientists not previously familiar with this field. No prior knowledge of climate science will be assumed, and only a basic graduate-level knowledge of statistics. The talk will introduce extreme value theory, show how these methods are applied in the climate context, discuss some of the pitfalls, and suggest directions for future research.
Richard Smith has been performing research in extreme value theory for several decades, has authored many papers on climatological statistics with research groups including the Statistical and Applied Mathematical Sciences Institute (SAMSI) and the National Center for Atmospheric Research (NCAR), and has interacted with numerous climate scientists in North America and worldwide. He has just been reappointed to the EPA’s Science Advisory Board. About the speaker, please visit http://rls.sites.oasis.unc.edu/ or https://sph.unc.edu/adv_profile/richard-smith-phd/
June 17, 2021
Speaker: Professor David Banks
Title: Statistical Issues in Agent-Based Models for Risk Assessment.
Abstract: Agent-based models (ABMs) are computational models used to simulate the actions and interactions of agents within a system. Usually, each agent has a relatively simple set of rules for how it responds to its environment and to other agents. These models are used to gain insight into the emergent behavior of complex systems with many agents, in which the emergent behavior depends upon the micro-level behavior of the individuals. ABMs are widely used in many fields, and this talk emphasizes the challenges that arise in the context various risk analyses (e.g., epidemics, invasive species, insurance). Relatively little work has been done on statistical theory for such models, this talk also points out some of those gaps and recent strategies to address them.
About the speaker, please visit https://www2.stat.duke.edu/~banks/
April 22, 2021
Speaker: Professor Nilanjan Chatterjee
Title: Polygenic Risk Prediction and Equitable Disease Prevention.
Abstract: Recent discoveries from large scale genome-wide association studies (GWAS) have raised the prospect of using polygenic risk scores in routine health care setting for the prediction of future incidence of large variety of complex diseases. However, as GWAS studies to date have been heavily biased towards European origin populations, current polygenic risk scores often underperform in non-European populations and thus use of them can further exacerbate healthcare inequality. In this talk, I will review simple and advanced statistical methods for generating polygenic risk score using high-dimensional SNP data and describe theoretical characterizations of their expected performance, both in the population that underlies original studies and in a different population that is expected to have different distribution of allele frequencies and linkage disequilibrium (SNP-correlation). I will further describe novel Bayesian and machine learning based methods for building polygenic risk scores that can borrow information across GWAS studies of different ethnic groups, and thus makes best use of available data to generate more powerful polygenic risk scores across different ethnic groups. I will demonstrate potential utility for PRS in precision medicine using our recent studies on breast cancer.
About the speaker, please visit http://www.nilanjanchatterjee.org/