Identification and estimation of survivor average causal effects

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Longitudinal studies often face missing outcome data due to participant death before follow-up.
• This
• truncation by death
• can bias causal effect estimates, even with randomized treatment assignment.

Purpose of the Study:

• To nonparametrically identify and estimate the survivor average causal effect (SACE).
• SACE quantifies treatment effects in the subpopulation that would survive irrespective of treatment status.
• To address bias introduced by death before follow-up in longitudinal studies.

Main Methods:

• Leveraging post-exposure longitudinal correlates of survival and outcome.
• Employing a nonparametric structural equations model with a monotonicity assumption on treatment's effect on survival.
• Utilizing a novel weighted analysis with a consistent estimate of the survival process.

Main Results:

• The proposed weighted analysis yields consistent estimates of SACE.
• The methods are extended to handle time-varying exposures.
• A sensitivity analysis framework is presented to assess assumption violations.

Conclusions:

• The study provides a robust method for estimating causal effects in the presence of death.
• The approach enhances the validity of findings from longitudinal studies with potential loss to follow-up due to mortality.
• The developed techniques offer valuable tools for biostatisticians and epidemiologists.