From EHR to Machine Learning: A Preliminary Report on an Ingestion Pipeline Based on JSON-LD

Area of Science:

• Computer Science
• Health Informatics
• Data Engineering

Background:

• Electronic Health Records (EHR) contain valuable data for machine learning.
• Integrating EHR data into machine learning workflows is challenging due to data heterogeneity and complexity.
• Existing methods often lack standardization and interoperability.

Purpose of the Study:

• To develop and present preliminary experiments for an ingestion mechanism.
• To facilitate the seamless transfer of data from Electronic Health Records (EHR) to machine learning (ML) processes.
• To leverage Linked Data concepts and the JSON-LD format for improved data integration.

Main Methods:

• Exploration of Linked Data principles for data representation.
• Utilization of the JSON-LD (JavaScript Object Notation for Linked Data) format for structuring EHR data.
• Development of a prototype ingestion mechanism for data transfer.
• Preliminary experimental validation of the mechanism's functionality.

Main Results:

• Demonstration of a functional ingestion mechanism for EHR data.
• Successful application of Linked Data and JSON-LD for data transformation.
• Preliminary evidence of efficient data movement from EHR to ML processes.
• Identification of key challenges and areas for future development.

Conclusions:

• The proposed ingestion mechanism shows promise for integrating EHR data into ML.
• Linked Data and JSON-LD offer a viable approach for standardized EHR data exchange.
• Further research and development are needed to optimize the mechanism for production environments.