Disease Prediction Using Graph Machine Learning Based on Electronic Health Data: A Review of Approaches and Trends.

Area of Science:

• Health Informatics
• Machine Learning
• Graph Theory

Background:

• Graph machine learning (ML) methods are advancing rapidly, yet their application in health informatics, particularly for disease prediction using electronic health data, remains underexplored.
• Existing reviews primarily focus on social networks, necessitating a focused review within the health domain.

Purpose of the Study:

• To comprehensively review graph ML methods and their applications in disease prediction using electronic health data.
• To identify current trends and challenges in this emerging field.

Main Methods:

• A systematic literature search was conducted across PubMed, Scopus, ACM digital library, and IEEEXplore.
• Articles applying or proposing graph ML models for disease prediction using electronic health data were identified and analyzed.
• Methods were categorized based on node classification and link prediction tasks, with a focus on shallow embedding and graph neural networks (GNNs).

Main Results:

• Graph neural networks (GNNs) demonstrate superior performance compared to traditional ML methods in various disease prediction tasks.
• Key challenges identified include model interpretability and handling dynamic graph structures.
• The field of ML-based disease prediction using electronic health data is still emerging.

Conclusions:

• GNN-based models hold significant potential for advancing disease prediction in medical diagnosis, treatment, and prognosis.
• Addressing interpretability and dynamic graph challenges is crucial for the wider adoption of GNNs in health informatics.
• Further research is warranted to fully leverage graph ML for improved healthcare outcomes.