Survivor Average Causal Effects for Continuous Time: A Principal Stratification Approach to Causal Inference With...

Area of Science:

• Biostatistics
• Causal Inference
• Health Services Research

Background:

• Semicompeting risks data, where nonterminal events (e.g., hospital readmission) are truncated by a terminal event (death), pose challenges for causal effect estimation.
• Traditional hazard models struggle with causal inference due to conditioning on survival, a posttreatment outcome.

Purpose of the Study:

• To extend the survivor average causal effect (SACE) framework for causal inference in semicompeting risks settings.
• To introduce novel causal estimands, the time-varying SACE (TV-SACE) and restricted mean SACE (RM-SACE).

Main Methods:

• Utilizing principal stratification to define causal effects among individuals who would survive regardless of treatment.
• Employing a Bayesian estimation procedure with parameterized illness-death models for both treatment arms.
• Incorporating a frailty specification to handle within-person correlation between event times.

Main Results:

• The proposed TV-SACE and RM-SACE estimands provide a robust framework for evaluating causal treatment effects in the presence of semicompeting risks.
• The Bayesian approach with frailty specification allows for flexible modeling of event time correlations.

Conclusions:

• The developed causal inference methods offer a valuable tool for analyzing time-to-event data in complex clinical scenarios, such as hospital readmission in cancer patients.
• This approach enhances the ability to draw valid causal conclusions from observational or trial data with competing risks.