Beyond the Hazard Ratio: Causal Inference from Time-to-Event Data with Dependent Censoring, Confounding, and...

Area of Science:

• Epidemiological research
• Biostatistics
• Causal inference

Background:

• Time-to-event outcomes are central to epidemiological research.
• Standard survival analysis methods, like hazard ratios, have limitations for causal inference.
• Confounding, dependent censoring, and competing risks are common challenges in observational studies.

Purpose of the Study:

• To review the necessity and limitations of survival analysis methods.
• To explain advanced techniques for addressing confounding, dependent censoring, and competing risks.
• To emphasize the importance of causal estimands for interpretable results.

Main Methods:

• Review of survival analysis techniques.
• Explanation of inverse probability weighting and parametric g-formula estimators.
• Discussion of recent developments in defining causal estimands with competing risks.
• Empirical comparison of various estimands through simulations.

Main Results:

• Hazard ratios are insufficient for causal inference.
• Inverse probability weighting and g-formula estimators can address confounding and dependent censoring.
• New methods for defining causal estimands in the presence of competing risks are summarized.
• Counterfactual risks and survival functions are interpretable causal parameters.

Conclusions:

• Move beyond routine survival analysis methods for causal inference.
• Causal survival analysis provides interpretable parameters crucial for investigators.
• Advanced methods are necessary to accurately estimate causal effects from time-to-event data.