Making valid causal inferences from observational data.

Area of Science:

• Epidemiology
• Causal Inference

Background:

• Randomized control trials (RCTs) are gold standard for causal inference.
• Observational studies present challenges for establishing causality.
• Counterfactual causal frameworks are central to modern causal theory.

Purpose of the Study:

• To review concepts of causation and counterfactual causality.
• To discuss methods for improving causal inferences from observational data.
• To highlight limitations of traditional multivariable modeling for confounding.

Main Methods:

• Review of causal inference concepts and counterfactual theory.
• Discussion of confounding and its impact on observational studies.
• Exploration of advanced methods like propensity scores and marginal structural models.

Main Results:

• Confounding significantly limits causal inferences in observational studies.
• Multivariable modeling has limitations in controlling confounding.
• Propensity scores can identify confounders and improve exchangeability.
• Marginal structural models are necessary for time-dependent confounders.

Conclusions:

• Improved causal inference from observational data requires better study design and advanced statistical methods.
• Asking precise questions and utilizing counterfactual thinking enhances causal validity.
• Propensity scores and methods for time-dependent confounders are crucial for robust causal claims.