A class of distribution-free models for longitudinal mediation analysis

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Mediation analysis is crucial for understanding intervention mechanisms in treatment studies.
• Structural Equation Models (SEM) are common for causal relationships but have limitations.
• Existing SEM methods struggle with longitudinal data and missing data under non-parametric assumptions.

Purpose of the Study:

• To propose a novel, robust approach for longitudinal mediation analysis.
• To overcome limitations of current SEM, particularly with distribution and missing data.
• To provide valid inference for mediation analysis under missing completely at random (MCAR) and missing at random (MAR) mechanisms.

Main Methods:

• Utilized functional response models (FRM) for a distribution-free approach.
• Extended inverse probability weighted (IPW) estimates to the SEM context.
• Applied the FRM-based SEM to both simulated and real longitudinal data.

Main Results:

• The FRM-based SEM approach is distribution-free, removing parametric assumptions.
• The method provides valid causal inference for longitudinal mediation analysis.
• Successfully handled missing data under MCAR and MAR assumptions in analyses.

Conclusions:

• The proposed FRM-based SEM offers a robust and flexible alternative for longitudinal mediation analysis.
• This approach enhances the utility of mediation analysis in real-world studies with complex data.
• It effectively addresses challenges posed by missing data in longitudinal research.