|Friday, February 16|
|PS2 Poster Session 2 and Refreshments||
Fri, Feb 16, 5:15 PM - 6:30 PM
An Algorithm to Identify Family Linkages Using Electronic Health Record Data (303689)
Stuart Cowburn, OCHIN
*Megan Hoopes, OCHIN, Inc.
Jon Puro, OCHIN
Pedro Rivera, OCHIN
Keywords: Linkage, Electronic Health Records, Validation
Many electronic health records (EHRs) lack explicit linkage of family members, hindering health services research on many health and policy outcomes. Objective: To develop a reproducible method to link family member EHR data. Method: We identified a set of data elements commonly captured in EHRs and often shared among family members (e.g. address, phone number). We tested combinations of target elements to predict family linkages. Results were evaluated against a ‘gold standard’ dataset of 25.5K explicit family links (i.e., emergency contacts, billing guarantor accounts) and measures of agreement were computed. Results: We identified 382K mother-child links in a dataset of 2.1 million patients; only 44.5% were explicitly documented in the EHR. The use of home phone numbers resulted in the highest number of links (sensitivity 0.8, precision 0.99), followed by Medicaid case number (sensitivity 0.34, precision 0.98) and geocoded last known address (sensitivity 0.18, precision 0.94). Conclusion: Accurate family linkages can be imputed using standardized methods and available EHR data, laying the groundwork for studies needing linked family data.