Causal inference with outcomes truncated by death in multiarm studies.

Area of Science:

• Biostatistics
• Epidemiology
• Toxicology

Background:

• Evaluating causal effects is difficult when outcomes are subject to truncation-by-death, a common issue in clinical and toxicological studies.
• Existing methods for binary treatments using principal stratification are insufficient for multiarm studies with multivalued ordinal treatments.

Purpose of the Study:

• To develop methods for identifying, estimating, and bounding causal effects in the presence of truncation-by-death with multivalued ordinal treatments.
• To address the complexities of causal inference in multiarm studies beyond binary treatment scenarios.

Main Methods:

• Defined causal parameters for multivalued ordinal treatments under truncation-by-death.
• Proposed a semiparametric estimation method with asymptotic results.
• Derived sharp bounds for causal effects using covariate adjustment when identification conditions are not met.

Main Results:

• Causal parameters were shown to be identifiable using an auxiliary variable or a linear model assumption.
• The proposed semiparametric estimator demonstrated good performance in simulation studies.
• Analysis of a developmental toxicity trial showed a high dose of chronic toxin significantly reduced pup weights.

Conclusions:

• The developed methods provide a framework for causal effect estimation in complex scenarios involving truncation-by-death and multivalued treatments.
• The findings highlight the significant impact of chronic toxins on fetal developmental outcomes, specifically pup weight.