Opportunities and Challenges in Using National EHR Networks for AI in Learning Health Systems.

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Learning Health Systems

Background:

• National electronic health record (EHR) networks offer infrastructure for learning health systems (LHSs), facilitating data aggregation and benchmarking.
• However, their effectiveness in generating reliable and locally deployable machine learning and artificial intelligence (ML/AI) models is not well-established.
• This study investigates the capacity of these networks for ML/AI development and deployment within the LHS framework.

Purpose of the Study:

• To characterize major US national EHR networks.
• To examine the barriers hindering ML/AI model development and deployment throughout the LHS cycle.
• To assess the current state and future potential of EHR networks as ML/AI platforms.

Main Methods:

• Conducted an environmental scan using PubMed searches and reviews of network websites, governance documents, and white papers.
• Included national EHR networks aggregating patient-level data.
• Abstracted network characteristics and identified ML/AI studies, mapping barriers to a seven-step LHS-AI cycle.

Main Results:

• Identified 23 national EHR networks with varying scales and data aggregation models.
• Found 34 ML/AI studies, with few prospectively evaluated or integrated into clinical workflows.
• Common barriers included data heterogeneity, privacy concerns, limited representativeness, and sociotechnical implementation challenges.

Conclusions:

• National EHR networks are crucial infrastructure for LHSs but currently serve more as research tools than ML/AI platforms.
• Overcoming data, implementation, and evaluation barriers is essential for realizing the potential of ML/AI in these networks.
• Enabling ML/AI development and local deployment is critical for advancing healthcare through EHR data.