BIS 537: Statistical Methods for Causal Inference (Fall 2020, 2021, 2022)

Course Description: This course introduces the potential outcome framework and the associated statistical methods that allow unconfounded comparisons of treatments in biomedical observational studies. Although randomization is the gold standard for unbiased treatment evaluation, observational studies are increasingly common for comparative effectiveness research. Principles and methods for the design and analysis of observational studies are discussed throughout the course. We define the causal effect as the difference in potential outcomes averaged over a target population, and introduce identification conditions for estimation and inference. In the first part of the course, we formalize the comparison of a point treatment, and develop regression, subclassification, matching, weighting and doubly robust methods. In the second part, we will focus on the more complex time-varying treatments in longitudinal observational studies, and introduce methods to account for time-dependent confounding and censoring bias. We explain why traditional regression adjustment fails, and discuss the methods of g-computation, marginal structural models and structural nested models. Advanced topics such as Bayesian methods and machine learning methods for causal inference will be discussed if time permits. Examples will be drawn from various biomedical intervention studies. Different from CDE/EHS 566a, this course emphasizes the statistical methodology driven by contemporary causal inference applications.