DYNAMIC PREDICTION WITH MULTIVARIATE LONGITUDINAL OUTCOMES AND LONGITUDINAL MAGNETIC RESONANCE IMAGING DATA

Area of Science:

• Neuroscience
• Biostatistics
• Medical Imaging

Background:

• Alzheimer's Disease (AD) is a prevalent neurodegenerative disorder impacting cognitive functions.
• Multimodal data from studies like the Alzheimer's Disease Neuroimaging Initiative (ADNI) are crucial for understanding AD progression.
• Accurate predictive models are needed for personalized medicine in high-risk individuals.

Purpose of the Study:

• To develop a multivariate functional mixed model with longitudinal MRI data (MFMM-LMRI) for predicting dementia onset.
• To integrate longitudinal neurological scores, voxelwise MRI data, and survival outcomes into a unified predictive framework.
• To establish a dynamic prediction system for individual dementia risk assessment.

Main Methods:

• Proposed a multivariate functional mixed model with longitudinal MRI data (MFMM-LMRI).
• Utilized the joint and individual variation explained (JIVE) approach for longitudinal MRI data modeling.
• Employed Markov chain Monte Carlo (MCMC) for posterior sampling and developed a dynamic prediction framework.

Main Results:

• Simulation studies confirmed the method's validity across various sample sizes and event rates.
• Application to the ADNI study indicated that ApoE-ϵ4 alleles and a higher latent disease profile increase dementia risk.
• A significant association was found between longitudinal MRI data and dementia onset, with the instantaneous model showing superior performance.

Conclusions:

• The MFMM-LMRI model effectively integrates multimodal data for predicting Alzheimer's Disease progression and onset.
• Identified key risk factors, including genetic markers and disease trajectory, associated with dementia.
• The developed dynamic prediction framework offers valuable insights for personalized risk assessment in AD.