Identifying Alzheimer's Disease Progression Subphenotypes Via a Graph-based Framework Using Electronic Health Records

Area of Science:

• Neuroscience
• Medical Informatics
• Computational Biology

Background:

• Alzheimer's disease (AD) neurodegeneration exhibits significant heterogeneity.
• Identifying distinct disease progression pathways is crucial for effective diagnosis, treatment, and prevention strategies.
• Current understanding often overlooks the diverse trajectories of cognitive decline.

Purpose of the Study:

• To identify and characterize distinct Alzheimer's disease (AD) progression subphenotypes.
• To analyze progression pathways from mild cognitive impairment (MCI) to AD using real-world data.
• To develop a novel computational framework for delineating neurodegenerative subphenotypes.

Main Methods:

• Development of a novel framework combining graph neural networks (GNNs) and time series clustering.
• Application of the framework to a large cohort of 2,525 patients with MCI and AD from electronic health records (EHRs).
• Analysis of clinical patterns and progression times associated with identified subphenotypes.

Main Results:

• Identification of four distinct MCI-to-AD progression subphenotypes.
• Characterization of unique clinical patterns within each subphenotype.
• Quantification of average MCI-to-AD progression times, ranging from 805 to 1,236 days, highlighting significant variability.

Conclusions:

• Alzheimer's disease (AD) progression is not uniform but follows heterogeneous pathways.
• The proposed GNN-based framework offers an explainable, data-driven method for subphenotyping AD progression.
• Findings provide actionable insights for healthcare informatics and personalized clinical management of AD patients.