Mediation analysis for mixture Cox proportional hazards cure models

Area of Science:

• Biostatistics
• Survival Analysis
• Causal Inference

Background:

• Mediation analysis decomposes total effects into causal pathways.
• Existing survival models struggle with a significant 'cure fraction' (subjects never experiencing the event).

Purpose of the Study:

• To develop mediation analysis for cure fraction problems using mixture cure models.
• To assess path-specific effects on survival outcomes like restricted mean survival time.

Main Methods:

• Utilizes a three-stage mediation framework with a partially latent group indicator.
• Employs a Bayesian approach with P-splines for the baseline hazard function.
• Applies the mediation formula approach to estimate effects.

Main Results:

• The proposed method demonstrates satisfactory performance in simulation studies.
• Successfully applied to investigate APOE-$$ allele effects on Alzheimer's disease progression using real-world data.

Conclusions:

• The novel method effectively handles mediation analysis in the presence of cure fractions.
• Provides a robust framework for understanding causal mechanisms in survival data, applicable to complex diseases like Alzheimer's.