Enhancing Personalized Healthcare via Capturing Disease Severity, Interaction, and Progression.

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Informatics
• Computational Health

Background:

• Personalized diagnosis prediction from electronic health records (EHR) is crucial but challenging.
• Existing AI models often overlook disease heterogeneity, including varying severity, complex interactions, and dynamic progression in patients.

Purpose of the Study:

• To develop an AI model for personalized diagnosis prediction that accounts for disease severity, interaction, and progression.
• To improve the accuracy of diagnostic predictions by addressing limitations in current EHR-based AI approaches.

Main Methods:

• Severity-driven embeddings for personalized disease representation.
• Hypergraph-based aggregation to capture higher-order disease interactions at the visit level.
• Neural ordinary differential equations for modeling continuous-time disease progression at the patient level.

Main Results:

• Significant performance improvements in diagnosis prediction compared to state-of-the-art methods.
• Average accuracy gains of 10.70% demonstrated on two real-world EHR datasets.
• The proposed approach effectively captures disease severity, complex interactions, and dynamic progression.

Conclusions:

• The developed AI model offers a more personalized and accurate approach to diagnosis prediction using EHR data.
• Addressing disease heterogeneity is key to advancing AI applications in personalized healthcare.
• This work provides a robust framework for leveraging complex EHR data for improved patient outcomes.