A New Linking Approach for Patient Medical Records in a Community Health Setting
Keywords: Record linkage, patient identifier, probabilistic matching
As US healthcare increasingly stresses data sharing and interoperability, the need to accurately and efficiently match patient information from disparate data sources will intensify. Moreover, the American Recovery and Reinvestment Act of 2009 set aside approximately $19 billion to encourage physicians to adopt electronic medical record systems. There are currently over 300 practice management systems in use with contrastingly few attempts made to link records among independent systems. Use of social security numbers as means of medical identification is a dwindling trend and without an unambiguous patient identifier, challenges will abound regarding patient information linkage. Much attention has been paid to candidate matching models on the basis of survival status for epidemiological and clinical studies, yet far less common are those techniques used to investigate patient matching over multiple data sources. We propose that a patient identification system reliant on a minimum amount of claim-driven information will meet our needs to link records over sources such as laboratory, hospital, practice management systems, and EMRs. Our simply-implemented probabilistic linking algorithm matches claim lines based on weighted information, if available, such as social security numbers, first and last names, date of birth, gender, and zip code. The model searches for probable matches using the information at hand and links data based upon a predetermined weighted-scoring threshold. The algorithm that we have constructed adapts current research to our position, which we expect to become more prevalent – a community healthcare model.