PROBABILISTIC RECORD LINKAGE BETWEEN DIGITAL IMAGING NET

DESCRIPTION (provided by applicant): Ubiquitous and secure access by physicians to a patient's medical records regardless of geographic location has tremendous potential to improve the delivery of medical care along multiple dimensions, yet the lack of a unique patient identifier between institutions remains a critical obstacle. While a national patient identifier was to have been established by the federal HIPAA legislation, opposition from privacy advocacy groups led to the suspension of a notice of proposed rulemaking. Probabilistic record linkage, however, represents a cross-disciplinary set of techniques that should permit reliable identification of patients across institutions in the absence of a unique identifier. Here we propose to study the feasibility of these techniques and quantify their reliability in the context of radiologic records, the only digital medical data for which there is a standard for electronic data interchange. In collaboration with the Hospital of the University of Pennsylvania (HUP), we propose to create a virtual patient "film jacket" across two unconnected picture archiving and communication systems (PACS). Adapting tools initially developed at the US Census Bureau, we will use probabilistic techniques to map unique patients between the two PACS based on the standard demographic data available in the DICOM headers of their imaging studies excluding the medical record number (MRN). We propose to compare this statistical approach to a benchmark of "true matches" represented by HUP's own MRN and use frequency analysis to determine appropriate cutoff values for probabilistic matches and evaluate the discriminatory power of the different DICOM attributes. We will also extend this analysis to additional demographic data stored in HUP's radiology information system (RIS) to evaluate the potential benefit of interfacing with both PACS and RIS and explore system performance as the number of records scales upwards. Phase II of the project, intended as a proof-of-concept to set the stage for commercialization, will involve applying our approach to a far more realistic test case - the creation of a virtual patient film jacket across two unaffiliated medical institutions. Commercialization of this technology will take the form of a radiology-specific global master patient index platform that will enable the creation of seamless, patient-centric virtual file jackets across arbitrary sets of institutions and PACWRIS installations. Such a product based upon non-proprietary, open standards would enable true and practical integration of care delivery between hospitals within an affiliated integrated delivery network or even between unaffiliated medical centers.