Modeling electronic health record data using an end-to-end knowledge-graph-informed topic model.

Area of Science:

• Computational biology
• Medical informatics
• Machine learning

Background:

• Electronic Health Record (EHR) datasets are growing rapidly, offering potential for disease understanding.
• Extracting clinical knowledge from sparse and noisy EHR data remains a challenge.

Purpose of the Study:

• To develop an effective method for distilling latent disease topics from EHR data.
• To improve patient representation for stratification and drug recommendations using EHR information.

Main Methods:

• Introduced Graph ATtention-Embedded Topic Model (GAT-ETM), a multimodal embedded topic model.
• Utilized a constructed medical knowledge graph to learn embeddings from EHR data.
• Applied GAT-ETM to a large-scale EHR dataset of over 1 million patients.

Main Results:

• GAT-ETM demonstrated superior performance in topic quality, drug imputation, and disease diagnosis prediction compared to alternative methods.
• Learned clinically meaningful, graph-informed embeddings of EHR codes.
• Discovered interpretable and accurate patient representations.

Conclusions:

• GAT-ETM effectively extracts clinical knowledge from EHR data by integrating medical knowledge graphs.
• The model provides enhanced patient representations for clinical applications like stratification and drug recommendations.