A causal framework for classical statistical estimands in failure-time settings with competing events.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Competing events complicate survival analysis by preventing the event of interest from occurring.
• Classical competing risks literature defines various statistical estimands but lacks a formal causal framework.
• Difficulty in interpreting effect estimates and analytic recommendations stems from the absence of causal characterization.

Purpose of the Study:

• To formally define classical competing risks estimands using a counterfactual framework.
• To clarify the interpretation of treatment effect estimates in the presence of competing events.
• To illustrate the use of causal diagrams for representing identifying assumptions.

Main Methods:

• Application of a counterfactual framework to define statistical estimands in competing risks.
• Distinction between contrasts of risks (total/direct effects) and counterfactual hazard contrasts.
• Representation of identifying assumptions using causal diagrams with time-varying covariates for competing events.

Main Results:

• Contrasts of risks can define total or direct causal effects of a treatment on the event of interest, depending on how competing events are treated.
• Counterfactual hazard contrasts generally cannot be interpreted as causal effects, irrespective of how competing events are defined.
• Causal diagrams effectively visualize identifying assumptions for counterfactual estimands.

Conclusions:

• The counterfactual framework provides a rigorous approach to defining and interpreting causal effects in competing risks settings.
• Distinguishing between risk contrasts and hazard contrasts is crucial for valid causal inference.
• The study provides a method for analyzing treatment effects on prostate cancer mortality using estrogen therapy trial data.