Kernel machine score test for pathway analysis in the presence of semi-competing risks

Area of Science:

• Genomics
• Biostatistics
• Cancer Research

Background:

• Cancer patients face non-terminal events (recurrence, metastasis) and terminal events (death).
• Understanding gene pathways linked to these events is crucial for disease insight and therapeutic development.
• Current methods modeling progression-free survival may overlook pathways specific to recurrence or death alone.

Purpose of the Study:

• To develop a statistical method for assessing a gene pathway's association with both cancer recurrence and death.
• To account for the dependency between these two distinct event types without making assumptions about their relationship.
• To identify pathways influencing disease progression and mortality more comprehensively than existing models.

Main Methods:

• A combined testing procedure was developed to evaluate pathway associations with cause-specific hazards of recurrence and marginal hazards of death.
• Perturbation resampling was employed to approximate the null distribution, effectively handling the dependency between outcomes.
• A flexible kernel machine framework was utilized to detect complex, non-linear relationships between gene pathways and clinical events.

Main Results:

• The proposed method demonstrated superior statistical power in identifying relevant gene pathways compared to standard approaches.
• Numerical simulations confirmed the effectiveness of the combined testing procedure.
• Application to a breast cancer gene expression dataset revealed significant pathways associated with disease progression and survival.

Conclusions:

• The novel statistical approach provides a more robust framework for analyzing multi-event cancer data.
• This method enhances the identification of gene pathways critical to cancer development and patient outcomes.
• The findings offer potential for discovering new biomarkers and therapeutic targets in cancer research.