Extending electronic medical records vector models with knowledge graphs to improve hospitalization prediction

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Data Science

Background:

• Artificial intelligence (AI) and machine learning (ML) can enhance analysis of electronic medical records (EMRs) for improved patient outcomes.
• Knowledge graphs (KGs) capture human knowledge and support reasoning, offering a way to pre-process and enrich data for ML algorithms.
• Standardizing medical data allows for the combined use of KGs and ML capabilities.

Purpose of the Study:

• To improve hospitalization prediction by enriching EMR vector representations with information from medical knowledge graphs.
• To automatically select relevant features derived from knowledge graphs to enhance predictive model performance.

Main Methods:

• Enriching electronic medical record (EMR) vector representations with information extracted from various knowledge graphs.
• Implementing an automatic feature selection process for KG-derived features.
• Conducting experiments on the PRIMEGE PACA database with over 600,000 patient consultations.

Main Results:

• The proposed approach significantly improves hospitalization prediction accuracy.
• Injecting features from cross-domain knowledge graphs into EMR representations increases the F1 score of the prediction algorithm.

Conclusions:

• Integrating knowledge from recognized sources into EMR representations enhances medical event prediction.
• Future work includes evaluating feature selection and combining features from multiple KGs, exploring hierarchical concept properties, and integrating semantic annotators for unstructured data.