AI Snips
Chapters
Transcript
Episode notes
From Medical Curiosity To Causal Research
- Miguel Hernán describes learning causal questions while training as a physician and wondering how clinicians know which treatment to give which patient.
- He recounts working on HIV where time-varying treatments and confounders made causal questions both urgent and methodologically rich.
Pick Any Good Method — If Time Zero Is Right
- Any adjustment method (propensity scores, outcome regression, standardization) works in simple two-arm comparisons if you have sufficient data.
- The harder problem is correctly defining groups and time zero; wrong definitions cause bias that adjustment cannot fix.
Time Zero Is A Fundamental Design Principle
- Defining the start of follow-up (time zero) as when eligibility and treatment assignment align is a fundamental design principle.
- Violating that principle (e.g., comparing current users to non-users) introduces selection biases that no statistical adjustment can reliably fix.