amVAE: Age-aware multimorbidity clustering using variational autoencoders.

Area of Science:

• Computational epidemiology
• Artificial intelligence in healthcare
• Chronic disease research

Background:

• Multimorbidity, the coexistence of multiple chronic conditions, is a growing global health challenge.
• Existing research often overlooks the temporal dynamics of multimorbidity progression, relying on static data.
• Understanding multimorbidity patterns is crucial for managing patient burden and healthcare system strain.

Purpose of the Study:

• To develop and validate a novel AI-driven method for temporal disease-based clustering.
• To identify age-aware multimorbidity clusters and their progression over time.
• To generate new hypotheses regarding the development and associations of multiple chronic conditions.

Main Methods:

• Introduction of a two-step multimodal Variational Autoencoder (VAE) approach for temporal clustering.
• Quantitative experiments to assess the robustness of the VAE model and extracted clusters.
• Application of the model to a large Danish population dataset (1995-2015) focusing on chronic heart disease patients.

Main Results:

• Successfully extracted distinct temporal clusters representing multimorbidity progression.
• Demonstrated the robustness and validity of the proposed AI approach.
• Identified novel insights into the dynamic development of multiple chronic conditions over time.

Conclusions:

• The novel AI approach effectively captures temporal dynamics in multimorbidity.
• Temporal disease clusters provide a deeper understanding of multimorbidity development and associations.
• Findings can inform targeted interventions and future research on chronic disease management.