This webinar is part of the Advanced Methods Webinar Series
Health administrative data is longitudinal with measures captured on individuals over time. Conventional regression-based methods applied to longitudinal data do not explicitly account for time-varying confounders and can produce biased estimates for causal effects. Marginal structural models are an estimation process used in longitudinal data for causal inference analysis and the control of time-varying confounding. These approaches require careful conceptual consideration of assumptions. This webinar will focus on introducing marginal structural models and specifically their utility in the context of population-wide health administrative data.
The webinar will:
- Describe what marginal structural models are and how these models compare to other regression-based approaches
- Cover the most common applications of marginal structural models, including time-varying confounding, causal inference, and causal mediation
- Discuss challenges to consider when implementing marginal structural models using health administrative data
- Include walkthrough examples of marginal structural models applied to health administrative data
- Cover the implementation considerations to be aware of when using these models, including assumptions, limitations, and interpretive cautions.
Dr. Laura Rosella, is an epidemiologist and Associate Professor in the Dalla Lana School of Public Health (DLSPH) at the University of Toronto, where she holds a Canada Research Chair in Population Health Analytics.
She is the Site Director for ICES U of T and Faculty Affiliate at the Vector Institute. In 2020, she was made the Inaugural Stephen Family Research Chair in Community Health at the Institute for Better Health, Trillium Health Partners. She leads the Population Health Analytics Lab our of DLSPH, which is focused on using population databases to inform population health and health system planning.