Causal Inference With Outcomes Truncated by Death and Missing Not at Random

Area of Science:

• Biostatistics
• Clinical Trials
• Causal Inference

Background:

• Principal stratification analysis is used in clinical trials to handle outcomes affected by death.
• Missing outcomes in survivors, not due to death, challenge standard methods.
• Accurate treatment effect estimation requires addressing these missing data issues.

Purpose of the Study:

• To develop methods for identifying and estimating the average treatment effect in a subpopulation of potential survivors.
• To address challenges posed by uncollected or missing-not-at-random survivor outcomes.
• To derive nonparametric bounds for causal parameters when identification assumptions are not met.

Main Methods:

• Introduced a proxy variable to identify the causal parameter under specific assumptions.
• Developed an estimation approach for causal parameters.
• Derived nonparametric bounds for causal parameters when identification assumptions are violated.

Main Results:

• Demonstrated that the causal parameter can be identified using a proxy variable.
• Proposed a method for estimating causal parameters.
• Derived nonparametric bounds for situations with violated assumptions.

Conclusions:

• The proposed methods enhance causal inference in clinical trials with missing outcome data.
• The study provides a framework for handling complex missingness patterns in principal stratification.
• The approach was validated through simulations and a human immunodeficiency virus (HIV) study dataset.