Featured Module (Archived)
(Week of October 28, 2024)
A new educational offering from the Statistics section of the curriculum wheel has been posted (1-1.5 hours of primary open access content).
This website will be updated every Monday (by 12:00 PM Eastern) or Tuesday (if Monday is a holiday). Given that the design, implementation, and management of pragmatic trials is a non-linear process, featured modules will relate to various sections of the curriculum wheel over time.
Statistics Section
Confounding and randomization: an introduction
Primary content:
Epidemiology Whiteboard Series - Clearing up Confounding (S. Roche & N. Lachowsky for the Department of Population Medicine, University of Guelph) (February 22, 2017 via “ACER Consulting”): 4-min video.
Summary: A brief visual depiction of confounding, one of the most important issues to consider when mitigating bias in an epidemiologic study. A working definition of a confounding variable is provided, where potential confounders must be: 1. Associated with the exposure of interest, 2. Associated with the outcome of interest, 3. Not be a result or consequence of the exposure (i.e., not be a mediator on the causal pathway between the exposure and outcome).
Lecture 8: Confounding in Health Research: Definition and Conceptual Issues – Part I (EPIB-601: Fundamentals of Epidemiology, McGill University, Prof. Madhukar Pai): 72-slide presentation.
Summary: Overviews the main objective of explanatory epidemiology (causal inference) and a central concern with establishing causality (i.e., confounding), emphasizing that a understanding of various approaches to identifying and controlling for confounding is essential for all who engage in health research (both observational research and clinical trials). Discusses four approaches to understanding confounding bias: 1. “Mixing of effects,” 2. Confounder selection based on a priori working criteria, 3. Collapsibility and data-based criteria, 4. “Counterfactual” and non-exchangeability-based criteria. Highlights that randomization helps to make the compared groups (e.g., intervention vs. comparator) exchangeable with respect to known and unknown confounders, highlighting examples of contradictory findings in observational studies vs. randomized trials.
Health in Numbers: Quantitative Methods in Clinical & Public Health Research - Week 9: CONFOUNDING: INTRO (HarvardX: PH207x) (December 18, 2012 via “DataLearner”): 11-min video.
Summary: Dr. Earl F. Cook introduces the concept of confounding using an example (i.e., the relationship between tobacco smoking and heart disease in a hypothetical observational cohort study, where age is a confounder). An introductory explanation of counterfactuals, exchangeability (“comparability”), and how randomization helps a researcher achieve exchangeability in a clinical trial (by reducing confounding) is also provided. Lastly, causal diagrams, which show a visual representation of the exposure (e.g., an intervention), outcome, and confounder(s) are introduced.
Randomized Controlled Trials and Confounding (November 10, 2013 via “Global Health with Gerg Martin”): 4-min video.
Summary: Dr. Greg Martin briefly connects the concept of confounding bias with randomization: a design-based approach to mitigate this bias in epidemiologic research. Ultimately, randomization balances both known and unknown confounding variables between the intervention (exposed) and comparator (unexposed) groups, meaning that any hypothetical confounding variables are no longer associated with the exposure; this nullifies working criteria #1 (“1. A confounder must be associated with the exposure of interest), mitigating confounding (effectively “deleting” the arrow connecting the confounding variable[s] to the exposure in a causal diagram).
International Society of Nephrology (ISN): ISN-ACT Clinical Trials Toolkit - Randomization (Study stage I: Design and development): 1-page website.
Summary: Further emphasizes that confounding bias is due to an imbalance in confounding variable(s) between the intervention and comparator groups. Highlights that, when done correctly, randomization is a design-based approach that balances known and unknown confounders between those who receive the intervention, and those who do not. Briefly introduces fixed allocation randomization and adaptive randomization.
Optional content:
Confounding Bias Part I (Second Edition No.11). Epidemiologic Research and Information Center (ERIC) Notebook - UNC Gillings School of Global Public Health. (Updated 2014-15): 5-page chapter.
Summary: This chapter describes confounding as one type of systematic error that can occur in epidemiologic studies. Elaborates on the working definition of confounding (i.e., a method to identify confounders) and provides practice questions involving 2x2 tables.