Patients Visiting Multiple Sites in HealthLNK Clinical Data Research Network

Research Objective: A clinical data research network (CDRN) with several EHR-data-contributing institutions in a single city offers the opportunity to describe healthcare use across institutions and to mitigate the incomplete follow-up seen in single-institution studies. CDRNs thereby also have been used to study “care fragmentation,” based on a patient’s having EHR records from >1 institution. But a CDRN itself may reflect incomplete data, either sent in that way by the contributors and/or as a result of the CDRN’s procedures for merging clinical data elements and patient identities across the institutions. And no CDRN comprises every healthcare facility in a city—every CDRN is, ipso facto, incomplete. Such limitations will become increasingly appreciated. Our objective is to contribute to this metadata discussion by describing patterns and correlates of the use of >1 institution. Study Design: In a retrospective (2006-2012) cohort study of 6 large healthcare institutions, we calculated the number of diagnosis records by patient by institution. Among those with records in exactly two institutions, we calculated each patient’s record counts in her/his two institutions. We similarly compare patients with ≥12 records in each of ≥3 institutions versus those who visited exactly 1 institution with ≥36 records there. Population Studied: We studied the International Classification of Diagnosis, Ninth Revision (ICD-9) diagnosis data from six institutions (four academic medical systems, one county hospital and clinic system, and one community health center network) in the HealthLNK repository—a forerunner to the Chicago Area Patient-Centered Outcomes Research Network (CAPriCORN) CDRN in PCORnet. HealthLNK is a de-identified assembly of EHR data of adults age 18-89. At its inception HealthLNK had merged and had de-duplicated patient identifiers across the institutions. HeathLNK’s method for individual-level linkage via de-identified hashing and matching now has been adopted by CAPriCORN and other PCORnet CDRNs. Principal Findings: Among 2,524,019 patients with at least one record in the Diagnosis Table in the 7-year period, 392,051 (16%) had records from ≥2 institutions. Most of these (327,396 [84% of 392,051; 13% of 2,524,019]) had records in exactly 2 institutions, and among them 40% had no-more-than 2 records in the less-frequently-visited institution in her/his pair. Another 64,655 (2.6% of 2,524,019) had records in ≥3 institutions. Compared to patients who visited exactly one institution and had ≥36 records there, those with ≥12 records in each of ≥3 institutions had higher rates of various ICD-9 diagnosis codes or categories, such as psychoses (4x higher), drug withdrawal (15x), alcohol withdrawal (13x), falls (4x), fractures (2x), homelessness (47x), and “procedure not carried out because of patient's decision” (6x). Conclusions: A high volume of records in ≥3 institutions versus in 1 were associated with psychosis, trauma, and homelessness.