Summary
Provides consensus practices and evaluation methods for AI systems that retrieve relevant diagnoses, medications, laboratory results, and other information from electronic health records, addressing retrieval accuracy, completeness, relevance, traceability, usability, safety, fairness, and monitoring.
Healthcare Implications
Healthcare organizations and developers should validate retrieval against local records and clinical workflows, measure missing and incorrect information, preserve links to source data, protect health information, test across patient groups and record types, and monitor whether updates or workflow changes degrade performance.