Explore the concept of assignment control plots for causal inference study design in this informative conference talk by Rocky Aikens from Stanford University. Delve into the importance of understanding the distribution of treated and untreated subjects in terms of measured baseline covariates for causal inference studies. Learn about the proposed set of visualizations that decompose the space of measured covariates into different types of baseline variation crucial for study design. Discover how these assignment-control plots visually illustrate core concepts of causal inference and suggest new directions for methodological research. Gain practical insights through an application of assignment-control plots to a cardiothoracic surgery study. Understand the potential of simple visual tools in education, application, and methods development for causality studies. Benefit from Aikens' expertise as a collaborative biostatistician focused on developing data-centered tools for stronger observational studies.
Assignment Control Plots for Causal Inference Study Design - Rocky Aikens
Stanford University via YouTube
Overview
Syllabus
Intro
I am a collaborative biostatistician
Example: Patient-physician concordance and heart attack
A confounder influences the treatment assignment and the outcome
Decomposing baseline variation
Maybe a covariate space we're really interested in is the subspace of covariates important to treatment and outcome
Methods hurdle: How do you fit your prognostic scores?
Application: Diagnostic
Example: Patient-Surgeon concordance and CABG
Application: Matching
Exploration: Unmeasured Confounding Variation
Exploration: Breaking down treatment assignment
Randomization Assignment-Control Plots
Characterizing Uncertainty in Practice
Other dimensions to consider
Key references
Taught by
Stanford MedAI
