The Effective Statistician - in association with PSI Estimands in Observational Studies
Nov 28, 2023
Artemis Koukounari, an expert in observational studies and Estimand frameworks, discusses the challenges and experiences of using Estimands in observational studies. Topics include enhancing controls with historical data, challenges with propensity score methods, importance of sensitivity analysis, and differences between clinical trials and observational studies
AI Snips
Chapters
Transcript
Episode notes
Schizophrenia Study Revealed Bias
- Early schizophrenia observational study showed treatment effects disappearing after one year due to patients switching medications.
- This revealed survival bias and highlighted the necessity to define different estimands carefully in observational studies.
Positivity Assumption Challenges
- Positivity assumption requires every patient to have a chance of receiving treatment under study.
- Violations can bias results and must be checked using diagnostics and possibly trimming extreme weights.
Population Heterogeneity in Observational Studies
- Observational studies have more heterogeneous populations, making average treatment effects among treated or untreated different.
- Different propensity score methods target different estimands aligned to specific research questions.