Cross-direct effects in settings with two mediators.

Area of Science:

• Causal inference
• Biostatistics
• Immunology

Background:

• Traditional causal inference focuses on natural (NDE) and controlled direct effects (CDE).
• Multiple mediators in biological systems, like immune responses, necessitate advanced analytical approaches.
• Existing methods may not fully capture complex interactions when multiple mediators are present.

Purpose of the Study:

• To introduce novel causal effect estimands for situations involving two mediators.
• To address cross-controlled direct effects and cross-natural direct effects.
• To provide methods applicable to both sequential and non-sequential mediator scenarios.

Main Methods:

• Development of five new estimands for cross-CDE and cross-NDE.
• Formulation of identifying expressions for observational data without residual confounding.
• Derivation of tight symbolic bounds for randomized settings with potential residual confounding.

Main Results:

• Introduction of five novel causal estimands for dual-mediator systems.
• Methods provided for arbitrary mediator and outcome types in observational studies.
• Bounds established for randomized experiments with binary variables and potential confounding.

Conclusions:

• The proposed estimands offer a more comprehensive understanding of causal pathways with multiple mediators.
• These methods have direct applications in fields like immunology, particularly for vaccine response studies.
• The work extends causal inference techniques to complex biological and observational data settings.