Society for Medical Decision Making

Course Background:

One of the most important tasks of decision makers is to derive causal interpretations using both statistical analyses of original datasets and decision-analytic modeling. Often an intervention, action or risk factor is modeled to have a "causal effect" on one or more model parameters (e.g., probability, rate, or mean of outcome). Therefore, both the biostatistician and the decision analyst need tools to check: (1) when effect estimates have a causal interpretation and when they do not; and (2) the appropriate methods to derive causal effects instead of merely statistical associations (e.g., traditional multivariate regression analysis or causal g-methods). This course intends to provide basic knowledge on causal thinking and visual, structural, and statistical tools to be able to judge in practice whether estimates are suitable for causal interpretation and which parameters from the literature can be used for causal decision-analytic models and which not. We will also discuss structural elements that make a model causal, and therefore, more robust when transported to a different setting or research question.

Course Description and Objectives:

This course introduces to the principles of causation, the use of causal diagrams (directed acyclic graphs, DAGs), and methods for causal inference. We will use the framework of "target trial emulation" with counterfactual "clones" to apply causal methods such as “cloning-censoring-weighting” to big real-world observational data and pragmatic trials and link this framework with causal decision-analytic modeling.

The course will consist of lectures, case examples drawn from the published literature, polls and interactive discussion.

The intended audience includes researchers from all substance matter fields, statisticians, epidemiologists, outcome researchers, health economists and health policy decision makers interested either in methods of causal analysis or causal interpretation of results based on the underlying method.

The objectives of this course are to:
1. define causal interventions and actions, draw and interpret causal diagrams, and apply the rules of causal diagrams to distinguish causal from non-causal statistical associations and to identify bias.
2. understand the concept of “target trial emulation” and the use of “clones” from a medical decision making perspective, and develop a target trial protocol.
3. decide which biostatistical/epidemiological methods (e.g., propensity score methods or g-methods) must be used in different situations to derive causal effect parameters.
4. Perform a target trial emulation using observational real-world data.
5. Describe how causal methods and approaches are applied in big observational real-world data, single arm trials, and pragmatic randomized trials.
6. understand the conceptual approaches of causal modeling.
7. derive causal modeling input parameters to feed decision-analytic models for benefit-harm assessment and health-economic evaluations in health technology assessment.

Published examples from oncology, cardiovascular disease, HIV, nutrition and obstetrics will be used to demonstrate how to:
• Develop a target trial protocol and identify the potential of self-inflicted (i.e., avoidable) biases;
• Adjust for time-independent confounders (i.e., confounder affects both risk factor and disease), where standard stratification, regression analysis, propensity score methods and matching/balancing approaches yield valid causal effects if all confounders are measured;
• Adjust for time-dependent confounding (i.e., the confounder simultaneously acts as an intermediate step in the causal chain between risk factor and disease), where standard regression analysis fails and causal g-methods such as g-formula, inverse probability weighting of marginal structural models or g-estimation of structural nested models must be used;
• Adjust for non-adherence or treatment switching in randomized clinical trials, where both the intention-to-treat and the naïve per protocol analyses can fail to yield the true causal intervention effect;
• Apply the concept of “target trial emulation” and counterfactuals using “clones” to perform a valid per-protocol analysis of sustained treatment regimens with big real-world data and pragmatic trials with post-randomization confounding.
• Discuss different biases such as time-independent confounding, time dependent confounding, selection bias, and immortal time bias etc.
• Support participants in explaining and teaching causal concepts to colleagues within their fields.

Course Format:

The course provides insight into causal modeling. It will consist of lectures, exercises drawn from the published literature. It will facilitate interaction and foster discussion using exercises, polls, and Q&A. Participants are invited to discuss their own research problems. Course material includes all session handouts, exercises with solutions, a comprehensive background reading library, and software recommendations. The intended audience includes researchers from all substance matter fields, statisticians, epidemiologists, and decision analysts interested either in methods of causal analysis or causal interpretation of results based on the underlying method.

Learning Objectives

• Define causal interventions and actions, draw, and interpret causal diagrams, and apply the rules of causal diagrams to distinguish causal from non-causal statistical associations and to identify bias.
• Decide which biostatistical/epidemiological methods (e.g., propensity score methods or g-methods) must be used in different situations to derive causal effect estimates.
• Perform a target trial emulation using observational real-world data.
• Describe how causal methods and approaches are applied in big observational real-world data, single arm trials, and pragmatic randomized trials.
• Derive causal modeling input parameters to feed decision-analytic models for benefit-harm assessment and health-economic evaluations in health technology assessment.
• Develop a decision-analytic model along causal modeling principles.

Requirements and Pre-Course Preparation:
• Basic knowledge in epidemiologic methods (confounding).
• Three pre-read papers will be distributed to participants prior to the course

Time Allocation & Topic Outline

Time Topics
5 min Wait seating participants
5 min Brief Intro/Background
40 min Principle of Causation
15 min DAG Exercises
40 min Causal Inference Methods I (Baseline Confounding)
20 min BREAK
50 min Causal Inference Methods II (Time-Dependent Confounding)
25 min Target Trial Emulation
20 min Case Example
5 min Causal Modeling
5 min Summary and Poll
10 min Q&A

Faculty Background & Qualifications

Uwe Siebert, MD, MPH, MSc, ScD

• 25 years of experience in teaching causal inference
• History of updating and teaching this course
• Member of the ISPOR-SMDM Modeling Good Research Practices Task Force
• Professor of Public Health, Medical Decision Making at UMIT TIROL - University for health Sciences and Technology, Hall in Tirol, Austria
• Adj. Professor of Epidemiology and Health Policy & Management, Harvard Chan School of Public Health, Boston, MA, USA
• Future author of the chapter on causal modeling in the book Hunink et al. Decision Making in Health and Medicine
Integrating Evidence and Values. Cambridge University Press.
• Author of publications on the interface between causal inference and decision-analytic modeling.

COI
None.