Modeling Disease Progression via Multisource Multitask Learners: A Case Study With Alzheimer's Disease

Area of Science:

• Biomedical Informatics
• Machine Learning
• Chronic Disease Research

Background:

• Chronic disease progression is complex, involving multiple data sources and temporal dependencies.
• Existing machine learning models often fail to address dual heterogeneities in disease progression data.
• Accurate prediction of future disease status is crucial for proactive patient management.

Purpose of the Study:

• To develop a novel, unified machine learning scheme for predicting chronic disease progression.
• To coregularize source consistency and temporal smoothness in disease prediction tasks.
• To demonstrate the model's effectiveness and generalizability across various chronic diseases.

Main Methods:

• Proposed a unified scheme integrating source consistency and temporal smoothness priors.
• Theoretically proved the proposed model to be a linear model.
• Employed matrix factorization to handle missing data before model training.
• Evaluated the model on a real-world Alzheimer's disease dataset.

Main Results:

• The proposed model effectively predicts chronic disease status at future time points.
• Demonstrated the model's effectiveness and efficiency on Alzheimer's disease data.
• The model's linear nature simplifies interpretation and application.

Conclusions:

• The novel machine learning scheme successfully addresses dual heterogeneities in chronic disease progression.
• The model shows promise for proactive patient care across a range of chronic diseases.
• Matrix factorization effectively pre-processes missing data, enhancing model performance.