Neural Network-Based Dynamic Prediction for Interval-Censored Data With Time-Varying Covariates: Application to...

Area of Science:

• Neuroscience
• Biostatistics
• Medical Informatics

Background:

• Alzheimer's disease (AD) is a leading cause of dementia with no effective treatments.
• Dynamic prediction models are crucial for timely intervention in AD.
• Existing models face challenges with intermittent data and complex covariate effects.

Purpose of the Study:

• To develop a novel dynamic prediction method for Alzheimer's disease.
• To accurately predict AD development using longitudinal cognitive and functional data.
• To identify high- and low-risk subgroups for personalized interventions.

Main Methods:

• Utilized data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) study (1702 individuals).
• Integrated multivariate functional principal component analysis (FPCA) with a neural network.
• Addressed interval-censored time-to-AD, multiple time-varying covariates, and nonlinear effects.

Main Results:

• The proposed method demonstrated superior prediction accuracy compared to existing approaches.
• Successfully identified distinct high- and low-risk subgroups based on progression profiles.
• Facilitated individualized and dynamic risk predictions for AD development.

Conclusions:

• The developed method offers accurate and dynamic risk prediction for Alzheimer's disease.
• Enables early identification of individuals needing timely intervention.
• An online platform is available for practical application of dynamic predictions.