Representation of information about family relatives as structured data in electronic health records.

BACKGROUND: The ability to manage and leverage family history information in the electronic health record (EHR) is crucial to delivering high-quality clinical care.

OBJECTIVES: We aimed to evaluate existing standards in representing relative information, examine this information documented in EHRs, and develop a natural language processing (NLP) application to extract relative information from free-text clinical documents.

METHODS: We reviewed a random sample of 100 admission notes and 100 discharge summaries of 198 patients, and also reviewed the structured entries for these patients in an EHR system's family history module. We investigated the two standards used by Stage 2 of Meaningful Use (SNOMED CT and HL7 Family History Standard) and identified coverage gaps of each standard in coding relative information. Finally, we evaluated the performance of the MTERMS NLP system in identifying relative information from free-text documents.

RESULTS: The structure and content of SNOMED CT and HL7 for representing relative information are different in several ways. Both terminologies have high coverage to represent local relative concepts built in an ambulatory EHR system, but gaps in key concept coverage were detected; coverage rates for relative information in free-text clinical documents were 95.2% and 98.6%, respectively. Compared to structured entries, richer family history information was only available in free-text documents. Using a comprehensive lexicon that included concepts and terms of relative information from different sources, we expanded the MTERMS NLP system to extract and encode relative information in clinical documents and achieved a corresponding precision of 100% and recall of 97.4%.

CONCLUSIONS: Comprehensive assessment and user guidance are critical to adopting standards into EHR systems in a meaningful way. A significant portion of patients' family history information is only documented in free-text clinical documents and NLP can be used to extract this information.