2017 Workshop

This workshop will introduce participants to a “causal roadmap” approach to epidemiologic questions: 1) clear statement of the scientific question, 2) definition of the causal model and parameter of interest, 3) assessment of identifiability – that is, linking the causal effect to a parameter estimable from the observed data distribution, 4) choice and implementation of estimators including parametric and semi-parametric, and 5) interpretation of findings. The focus will be on estimation with a simple substitution estimator (parametric g-formula), inverse probability of treatment weighting (IPTW), and targeted maximum likelihood estimation (TMLE) with SuperLearner. Participants will work through the roadmap using an applied example and implement these estimators in R during the workshop session.

Workshop Chairs:
Jennifer Ahern, University of California, Berkeley
Laura Balzer, Harvard University

This workshop introduces concepts of causal inference and confounding control for causal effect estimation. We will introduce potential outcomes, and articulate the conceptual basis and assumptions for two g-methods – standardization via g-computation and inverse probability weighting. Starting with a simple point-treatment setting we will explore how these methods estimate a causal effect, comparing them to more conventional techniques such as multivariable regression and propensity score control. We will then build to the more complex scenario of time-dependent confounding. Participants will learn how to apply these methods in SAS and R using an observational dataset with the primary goal of unpacking any “black boxes” to clarify the links among the causal effect of interest, the mechanics of these g-methods, and the programming code.  

Workshop Chairs:
Nicolle Gatto
Ulka Campbell

Drawing valid causal conclusions from observational data is challenging, yet crucial for predicting how changing exposures would affect public or occupational health.  Current machine-learning methods can help discover, test, and validate causal models and predictions by using data to test their implications, such as that (a) Causes provide unique information about their effects; (b) Changes in causes help to predict and explain subsequent changes in their effects; and (c) Information flows from causes to their effects over time.  This workshop will discuss and illustrate information-based methods of causal inference and show how progress in causal analytics updates the classical Hill considerations by replacing statistical associations with information-theoretic relations.  Practical illustrations are given using a Causal Analysis Toolkit (CAT) that allows Excel users to apply advanced R packages to identify and quantify potential causal relationships in epidemiological data.

Workshop Chair:
Tony Cox

This workshop is for the cheerleaders and critics of systems science in epidemiologic research. Participants will gain a better understanding of the historical and scientific development of systems thinking by focusing on three systems science methods (system dynamics, network analysis, and agent-based models). After this overview, participants will gain hands-on training with software and data analysis for each systems science method and, furthermore, learn how to interpret and translate research findings into actions of consequence in public health and healthcare policy. Workshop presenters bring expertise in using systems science thinking and methods to address a range of outcomes, including infectious diseases, pediatric asthma, birth outcomes, and obesity.

Workshop Chair:
Ayaz Hyder

Unmeasured confounders can lead to substantial bias during causal analysis. Indeed, no uncontrolled confounding is a key assumption needed for effect identification in causal inference. Investigators hope to have measured and controlled for all the variables that confound the effect of the exposure or treatment on the outcome under study. Often, after controlling for the measured confounders, there is residual confounding that should be addressed. This workshop will feature detailed discussions of the impact of uncontrolled confounding on estimates of the effect of the exposure and of the associations of measured confounders with the outcome. Modern bias analysis tools including bias formulas for handling uncontrolled confounding will be introduced and demonstrated. We will also discuss how to reason about, specify and use bias parameters in fixed and probabilistic bias analysis. Single and multiple unmeasured confounders will be considered. Interpretation and reporting will be discussed. Ample illustration and software codes will be provided and distributed.

Workshop Chair:
Onyebuchi Arah

This participatory workshop will take participants through several aspects of successful grant writing strategies: planning, the all-important specific aims, how to introduce rigor and reproducibility into the proposal and the (now) all-important biosketch development. Planning a proposal entails several preliminary stages including establishing the “back story”, critically reviewing the pertinent literature and developing testable hypotheses. Developing the specific aims, which are the cornerstone of a proposal, involves science, art and marketing strategies. Rigor and reproducibility are new NIH criteria for review, which need to be incorporated into all aspects of a proposal. Finally, the development of a biosketch which is pertinent to the specific proposal is accounted for in the scoring of proposals. This workshop will use examples from the leader’s own work and examples from participants to highlight these important aspects of proposal development.

Participants are asked to send drafts of specific aims (they will be blinded) for inclusion in the workshop or to bring sets of specific aims that they wish to work on. They are also asked to bring drafts of a biosketch to match the specific aims in their proposal. There will be ample time for interaction between participants and the instructor. New drafts of these portions of a proposal will be the deliverables.

Workshop Chair:
Pam Factor-Litvak, Columbia University

Bayesian methods formalize the process of updating prior knowledge with current observation, and have been employed in a number of settings relevant for public health research. However, epidemiologists typically receive little formal training in these techniques. This workshop will present the fundamentals of Bayesian methods applied to the analysis of epidemiologic data, including hierarchical regression modeling and Bayesian methods for bias analysis and missing data. We will use the R statistical programming language and JAGS (Just Another Gibbs Sampler); participants are encouraged to bring a laptop with these programs installed. Participants should be familiar with basic statistical analysis (probability and regression); acquaintance with the basics of R (reading data, basic variable manipulation, regression modeling) is helpful but not required. At the completion of this workshop learners should be able to incorporate Bayesian methods into the analysis of their own data.

Workshop Chair:
Patrick Bradshaw, University of California, Berkeley

In this half-day workshop, participants will critically review a paper as initially submitted to the American Journal of Epidemiology (AjE), but not yet published. The paper will be sent to participants in advance of the workshop for their critical review. During the workshop, a presentation will be made on some of the main points to be considered when preparing or reviewing a manuscript. Small group work will follow the presentation so that participants can compare their reviews and prepare a consolidated list of critical comments on the paper. Each group will designate a rapporteur, who will present the group’s review of the paper and decision letter to the whole group of participants. At the end of the workshop, students will receive copies of the manuscript’s AjE reviews, the initial editorial decision and the final version of the paper. Attendance will be limited to 30 participants.  

Workshop Chair:
Moyses Szko, Editor-in-Chief, AjE

This workshop will introduce participants to the R statistical computing platform for use in epidemiologic analysis. It is not intended to transform untested novices into R wizards in a mere half-day; rather, the goal will be to introduce the conceptual underpinnings, tools, and external resources that participants will need to overcome barriers to using R that they might encounter on their own, later. The material is designed for epidemiologists who are already familiar performing analyses using other statistical software (e.g. SAS/Stata/SPSS) but who have no first-hand experience with the R language. More specifically, the course will cover 1) basic R syntax, 2) importing data, 3) constructing, cleaning, and manipulating data objects, 4) loading and using external packages, 5) simple statistical modeling, and 6) graphics. Participants must bring a laptop with R installed; the instructor will be available by email beforehand to assist with R installation if difficulties arise.

Workshop Chairs:
Stephen Mooney, University of Washington 

This workshop will provide a broad introduction to the topic of interaction. We will discuss interaction on additive and multiplicative scales, and their relation to statistical models (e.g. linear, log-linear and logistic models).  We will describe procedures for interaction when logistic models are fit to data but when additive and not just multiplicative measures of interaction are desired. We discuss issues of confounding for interaction analyses and how whether control has been made for only one or both of two exposures affects interpretation. We further discuss conditions under which interaction gives evidence of synergism within the sufficient cause framework, when interaction is robust to unmeasured confounding, methods attributing effects to interaction, case-only estimators of interaction, and power and sample size calculations for interaction. Illustrations will be given from environmental, genetic, and infectious disease epidemiology. Software code will be provided.

Workshop Chair:
Tyler VanderWeele, Harvard University

This workshop will introduce participants to directed acyclic graphs (DAGs). We will review the basic principles and show how they can be used to determine appropriate sets of variables for estimating total causal effects of exposure (treatment).  Participants will work through concrete examples of increasing complexity.  We will also discuss more advanced applications, including natural and controlled direct and indirect effects, bias-amplification, how DAGs might be helpful in evidence synthesis, selection diagrams and SWIGs. Bringing a laptop will be helpful, but is not essential.

Workshop Chairs:
Ian Shrier, McGill University

Applied health scientists are increasingly dealing with complex data structures to answer questions about exposure effects and mediation. In such settings, feedback between confounders, exposures, and mediators render standard adjustment methods (regression, restriction, stratification, matching) inappropriate. The parametric g formula—one of three “g” methods—is a versatile tool that can be used to quantify a variety of exposure effects with complex data structures.

This workshop will provide a comprehensive overview of the g formula for identifying and estimating causal effects. After a brief introduction to the potential outcomes framework, we will review obstacles to effect estimation and mediation analysis with complex longitudinal data. The g formula will then be introduced with three examples using actual data and software code: (i) a simple simulated analysis that minimizes technical details and emphasizes core concepts; (ii) a mediation analysis setting where interest lies in direct/indirect effects; and (iii) a complex longitudinal data setting where interest lies in estimating the total effect of an exposure measured repeatedly over many months of follow-up. The goal of this workshop will be to enable participants to implement the parametric g formula in a range of settings, to articulate and evaluate key assumptions/limitations, and to implement critical model validation techniques. No prior knowledge of causal modeling, counterfactuals, or g methods is required.

Workshop Chair:
Ashley Naimi, University of Pittsburgh

Epidemiologists can use well-designed statistical graphics to understand data and to guide us toward correct inferences. These graphics can also be powerful tools for communicating study findings. While statistical software makes it easy to produce standard figures, default options often leave much to be desired and can produce figures that distract, confuse, or even distort data. In this workshop, participants will learn the fundamentals of effective data visualization and how to apply these best practices to create their own graphics. We will begin by reviewing general principles for displaying data, enabling viewers to make comparisons and identify trends and correlations. We will then show how to put these principles into practice, such as creating charts that are appropriate for a given dataset and the effective use of color, size, and annotation. We will also walk through techniques for going beyond the default settings of some software packages to produce well-designed figures.

Workshop Chairs:
Mike Jackson

R. Yates Coley

Bayesian analysis has become a popular tool for many statistical applications.  Yet many statisticians have little training in the theory of Bayesian analysis and software used to fit Bayesian models.  This talk will provide an intuitive introduction to the concepts of Bayesian analysis and demonstrate how to fit Bayesian models using Stata.  No prior knowledge of Bayesian analysis is necessary and specific topics will include the relationship between likelihood functions, prior, and posterior distributions, Markov Chain Monte Carlo (MCMC) using the Metropolis-Hastings algorithm, and how to use Stata’s graphical user interface and command syntax to fit Bayesian models.

Workshop Chair:
Chuck Huber