CIMPOD

'Learning-by-doing' -- By registering for CIMPOD’s case-study driven interactive workshops, attendees will be given the opportunity to learn 3 different causal inference techniques. Instructors will utilize real-world examples to demonstrate how these powerful methods can be implemented and utilized effectively in various research settings. CIMPOD is a unique opportunity for research and patient stakeholders to learn cutting-edge and practical causal methods from a roster of internationally recognized experts.

Learning Objectives

•  Articulate your CER question effectively and select a specific causal inference method that is most appropriate for your research question

• Step-by-step implementation of causal methods with related software and program code

• Hands-on experience to help tackle the challenges related to applying the selected causal inference methods

If you want to understand how causal inference methods should be selected and applied in practice, the CIMPOD workshops are for you. The workshop learning format will illustrate causal methods through the use of detailed case-studies (derived from highly regarded real-world research), software implementation examples (covering SAS, Stata, and R analytics packages), and general “tips & tricks” that researchers can use to overcome the most common challenges faced when applying the methods to individual CER questions and databases.

These workshops are most appropriate for researchers in industry, academia, and government agencies.

1. Participants should have: a solid understanding of basic principles of statistical inference, including such concepts as bias, sampling distributions, standard errors, confidence intervals, and hypothesis testing

2. A good working knowledge of multivariate regression analysis

3. Experience using at least one of the following statistical packages: SAS, STATA, or R (see the specific workshop descriptions for more details)

The workshops will cover three topics: CER study design via explicit emulation of a target trial (Workshop #1); implementation of targeted maximum likelihood estimation (TMLE) with Super Learning (Workshop #2); and theoretical foundation and practical implementation of g-methods including inverse probability weighting and g-computation (Workshop #3).

All three workshops are interactive, case-study driven, and with a focus of research questions involving time-varying exposures. Instructors will utilize real-world examples to illustrate how covered topics can be implemented and utilized effectively. 

Workshop # 1 : CER study design via explicit emulation of a target trial 

Yi Zhang, Medical Technology & Practice Patterns Institute

Abstract : Both patients and clinicians want to understand what treatments work best when making complicated health-related decisions. These decisions may be supported by findings from randomized trials or, when these are not available, by the findings from observational data analyses that explicitly emulate a hypothetical randomized trial: the target trial . The Target Trial Framework is consistent with the formal counterfactual theory of causal inference and prevents common methodological pitfalls, such as immortal time bias and selection bias. This workshop will introduce a structured standardized algorithm to define and emulate a target trial. By following the provided algorithm or template, researcher can complete their CER study design with elements that are consistent with the fundamental principles of causal inference. Dynamic ESA treatment to correct anemia among dialysis patients will be used to illustrate the steps of emulating a target trial using Medicare claims data.

 We developed CERBOT (Comparative Effectiveness Research Based on Observational data to Emulate a Target Trial), a web-based tool that provides a structured standardized algorithm to define and emulate a target trial. Please go to cerbot.org for details. This tool prepare you with a good understanding of study design elements we will go over during Workshop #1 ; 2) a PDF of my article (see attached) with the following description:

Case studies to be used:

1.   Zhang Y, Young JG, Thamer M, Hernán MA. Comparing the Effectiveness of Dynamic Treatment Strategies Using Electronic Health Records: An Application of the Parametric g-formula to Anemia Management Strategies. Health services research. 2018 Jun;53(3):1900-18;  (link)

2. Zhang Y, Thamer M, Kaufman J, Cotter D, Hernán M. Comparative effectiveness of two anemia management strategies for complex elderly dialysis patients. Medical care. 2014 Mar;52(0 3):S132. (link)

3. Comparative Effectiveness of Two Anemia Management Strategies for Complex Elderly Dialysis Patients (link) 

For Workshop slides click here

For Workshop recording, click here

Workshop # 2 : Targeted Maximum Likelihood Estimation, integrating machine-learning, to evaluate the effects of longitudinal interventions

Maya Petersen, MD, PhD

School of Public Health

University of California, Berkeley

Abstract: Targeted Maximum Likelihood Estimation (TMLE) provides an approach for estimating the causal effects of longitudinal interventions with several attractive properties. TMLE uses estimates of both the propensity score (as used in inverse probability weighting) and of a series of outcome regressions (as can be used in parametric G-computation). Machine-learning methods, such as Super Learning (an ensemble approach) can be used to estimate both the propensity score and outcome regressions. TMLE, which is a double robust semiparametric efficient estimator, has the potential to reduce bias and variance and to improve the validity of statistical inferences compared to alternative approaches. However, as with other methods, challenges remain, particularly when some treatment regimes of interest have poor data support given confounder values.

This workshop will provide an introduction to implementation of TMLE with Super Learning. Methods will be illustrated using applied case studies drawn from HIV implementation science. A brief introduction to the R-package ltmle, which can be used to implement all methods described in the workshop, will also be provided.

For Workshop slides click here

For Workshop recording, click here

Workshop # 3 : Inverse probability weighting and g-computation for estimating effects of time-varying dynamic treatment strategies in observational studies

Jessica Young

Harvard Medical School

Abstract : Dynamic strategies are time-varying treatment rules under which a patient’s treatment assignment at a given follow-up time depends on her time-evolving risk factors. G-methods can recover causal effects of time-varying dynamic strategies on a future outcome even when time-varying confounders are affected by past treatment. In this workshop, we will learn about two g-methods: inverse probability weighting and g-computation. Using real-world examples, we will review the theoretical foundation for these methods, the relationship between them, along with details of practical implementation. We will consider advantages and disadvantages of these approaches relative to each other and to alternative approaches.

For Workshop slides click here

For Workshop recording, click here