Shape Mediation Analysis in Alzheimer's Disease Studies.

Area of Science:

• Neuroscience
• Biostatistics
• Medical Imaging

Background:

• Mediation analysis is vital in neuroscience for understanding intermediary variables from neuroimaging.
• Current methods, often using structural equation models (SEMs), assume linear relationships and are limited for shape-space mediators.
• Existing SEMs may suffer efficiency losses and reduced predictive accuracy with complex data.

Purpose of the Study:

• To develop a novel framework for shape mediation analysis in neuroscience.
• To explore causal relationships between genetic exposures and clinical outcomes, mediated by shape factors.
• To address limitations of linear assumptions in existing mediation models.

Main Methods:

• Applied the square-root velocity function to extract elastic shape representations within a linear Hilbert space.
• Introduced a two-layer shape regression model to link neurocognitive outcomes, shape mediators, genetic exposures, and confounders.
• Developed estimation and inference procedures for unknown parameters and causal estimands, investigating asymptotic properties.

Main Results:

• The proposed shape mediation analysis framework demonstrated superior performance in simulated and real-data analyses.
• Achieved higher estimation accuracy and robustness compared to existing approaches for causal estimands.
• Effectively modeled complex relationships involving shape-based mediators.

Conclusions:

• The novel shape mediation analysis framework offers a powerful tool for neuroscience research involving shape-space mediators.
• This method enhances the understanding of causal pathways between genetic factors and clinical outcomes.
• The approach provides more accurate and robust estimations than traditional linear models.