General Identification of Dynamic Treatment Regimes Under Interference

Area of Science:

• Causal Inference
• Statistical Modeling
• Public Health

Background:

• Tailoring treatments to individual characteristics is crucial for optimizing outcomes in applied fields.
• Inter-subject dependence, or interference, where one subject's outcome is influenced by a neighbor's exposure, violates standard data assumptions.
• Existing dynamic treatment regime methods often assume data independence, limiting their applicability in settings with interference.

Purpose of the Study:

• To address the challenge of identifying optimal treatment policies in the presence of interference.
• To extend existing causal inference identification theory to settings with inter-subject dependence.
• To provide a framework for formalizing policy interventions under interference.

Main Methods:

• Utilized Lauritzen-Wermuth-Freydenburg chain graphs to represent interference generally.
• Formalized various policy interventions within this chain graph framework.
• Extended identification theory for optimal treatment policies under interference.

Main Results:

• Developed a general representation of interference using chain graphs.
• Established a formal framework for policy interventions in the presence of interference.
• Demonstrated the efficacy of policy maximization under interference through a simulation study.

Conclusions:

• The proposed methods enhance the ability to identify optimal treatment policies in complex scenarios with inter-subject dependence.
• The framework provides a robust approach for causal inference in the presence of interference.
• This research contributes to optimizing interventions in fields where outcomes are influenced by social or network effects.