High-dimensional causal mediation analysis by partial sum statistic and sample splitting strategy in imaging genetics...

Area of Science:

• Biostatistics
• Genomics
• Medical Imaging
• Causal Inference

Background:

• Causal mediation analysis is crucial for understanding exposure-outcome relationships via mediators.
• High-dimensional omics and imaging data present challenges like multicollinearity and poor interpretability in mediation analysis.
• Existing methods often struggle with mediator prioritization and interpretability in complex datasets.

Purpose of the Study:

• To introduce the Partial Sum Statistic and Sample Splitting Strategy (PS5) framework for high-dimensional causal mediation analysis.
• To address limitations of existing methods in handling high-dimensional mediators.
• To improve mediator selection accuracy and estimation of mediation effects.

Main Methods:

• Proposed the Partial Sum Statistic and Sample Splitting Strategy (PS5) framework.
• Conducted extensive simulations to evaluate PS5's performance.
• Applied PS5 to an imaging genetics dataset from the COPDGene study.

Main Results:

• PS5 demonstrated superior type I error control, increased statistical power, and reduced bias in mediation effect estimation compared to existing methods.
• PS5 achieved more accurate mediator selection.
• Analysis of COPDGene data successfully estimated the global indirect effect and identified mediating image regions, including a specific lower lung lobe region.

Conclusions:

• The PS5 framework effectively handles high-dimensional mediators in causal mediation analysis.
• PS5 offers improved performance in terms of error control, power, bias reduction, and mediator selection.
• Identified a key lung region mediating genetic and environmental effects in COPD, suggesting potential therapeutic targets.