A Bayesian joint model for mediation analysis with matrix-valued mediators

Area of Science:

• Biostatistics
• Radiation Oncology
• Medical Physics

Background:

• Unscheduled treatment interruptions in radiation therapy (RT) can compromise patient care quality.
• Understanding the relationship between RT prescription dose, organs at risk (OARs) radiation exposure, and treatment interruptions is crucial for optimizing treatment planning.
• Dose-volume histograms (DVH) provide a matrix-valued representation of OARs radiation exposure, posing challenges for traditional mediation analysis.

Purpose of the Study:

• To propose a novel Bayesian joint mediation model capable of handling high-dimensional matrix-valued mediators, specifically DVH data.
• To investigate the mediation effects of OARs radiation exposure on the relationship between RT prescription dose and treatment interruptions.
• To develop methods for extracting latent features from matrix-valued data and identifying significant mediation pathways.

Main Methods:

• Development of a Bayesian joint mediation model incorporating an adaptation of probabilistic multilinear principal components analysis (MPCA) for matrix-valued DVH data.
• Implementation of a Gibbs sampling algorithm for joint estimation of model parameters.
• Application of Varimax rotation to identify active mediation indicators within the matrix-valued data.
• Simulation studies to compare the proposed model's efficiency against a two-step method.

Main Results:

• The proposed Bayesian joint model demonstrates higher efficiency in estimating causal decomposition effects compared to a two-step approach.
• The model successfully identifies and visualizes mediation effects within the matrix structure of DVH data.
• The method was applied to analyze the impact of prescription dose on treatment interruptions in anal canal cancer patients.

Conclusions:

• The novel Bayesian joint mediation model provides an effective framework for analyzing high-dimensional matrix-valued mediators like DVH.
• This approach enhances the understanding of how radiation dose distribution in OARs influences treatment interruptions.
• The findings can inform future radiation therapy planning to minimize interruptions and improve patient outcomes.