Causal inference with recurrent and competing events.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Recurrent events (e.g., hospitalizations, injuries) are common in clinical research.
• Competing events (e.g., death) complicate the analysis of recurrent events.
• Formal causal interpretations of statistical estimands in these settings are lacking.

Purpose of the Study:

• To formalize causal estimands for recurrent events with and without competing events.
• To clarify the causal interpretation of existing statistical estimands.
• To introduce novel causal estimands relevant to clinical practice.

Main Methods:

• Utilized a formal causal inference framework, including causal directed acyclic graphs and single world intervention graphs.
• Applied counting process theory to connect discrete and continuous time analyses.
• Developed and validated estimators for proposed causal estimands.

Main Results:

• Established formal causal interpretations for classical estimands in the presence of competing events.
• Defined new interventionist mediation estimands for recurrent and competing events.
• Demonstrated the convergence of discrete-time causal estimands to continuous-time counterparts.

Conclusions:

• The proposed causal framework provides a rigorous approach to analyzing recurrent events with competing risks.
• New causal estimands offer enhanced clinical relevance for understanding treatment effects.
• The methods were successfully applied to real-world data on blood pressure treatment and kidney injury recurrence.