Curation of a Cardiology Interface Terminology for Highlighting Electronic Health Records using Machine Learning

Researchers developed a Cardiology Interface Terminology (CIT) system using machine learning to automatically highlight critical information in electronic health records, achieving 74.21% coverage with 98.2% completeness in identifying relevant clinical details.

This research addresses a significant challenge in clinical informatics: the cognitive overload physicians face when processing dense, jargon-laden electronic health records. The study presents a three-phase methodology that transforms how medical terminology gets extracted and applied to EHR documents. The approach combines domain expertise (SNOMED hierarchies, cardiology-specific concepts) with automated machine learning to create intelligent highlighting systems that reduce information oversight risks.

The work builds on decades of medical informatics research but innovates by reducing manual annotation burden through semi-automated candidate review and iterative refinement. By leveraging existing medical ontologies alongside ML model training, the methodology achieves scalability without requiring prohibitive manual effort. The 98.2% completeness rate on test data suggests the system reliably captures clinically relevant concepts.

For healthcare technology vendors and EHR developers, this represents a viable pathway to implementing intelligent content highlighting features that could reduce diagnostic errors and improve clinical workflow efficiency. The metrics demonstrate practical viability: high completeness ensures minimal missed information while reasonable breadth (1.68) prevents excessive highlighting that reduces usefulness. The 84.2% conciseness score indicates the system maintains signal quality.

Future applications extend beyond cardiology to other medical specialties, suggesting a generalizable framework. However, the 74.21% coverage gap indicates refinement opportunities, particularly for edge-case terminology and newly emerging clinical concepts. Real-world deployment requires validation across diverse EHR systems and patient populations to ensure the model generalizes beyond the training environment.

• →Machine learning combined with medical ontologies enables semi-automated creation of specialized clinical terminologies without excessive manual annotation.
• →The system achieved 98.2% completeness in identifying relevant cardiology concepts, suggesting high clinical safety potential.
• →Three-phase methodology balances automation with human review, reducing development costs while maintaining quality standards.
• →74.21% coverage indicates the system captures most but not all relevant clinical details, requiring ongoing refinement.
• →Approach is generalizable to other medical specialties beyond cardiology for broader healthcare IT applications.