Variation in Cohorts Derived from EHR Data in Four Care Delivery Settings

EHR data are desirable for secondary use in health research but are known to have inconsistencies. The context of their origination, or provenance, may affect the comparability of EHR data for secondary usage. We investigated suspected differences in demographic and comorbidity data and availability of information among four health care settings: ambulatory office visits, hospital inpatients, emergency visits, and visits for tests and other diagnostic and treatment procedures only. Descriptive comparative results suggest that cases accrued to a diabetes cohort in these settings differed on demographics, morbidity profiles and completeness of EHR data. The distribution of cases among the four settings also differed by provider geographic regions. These differences may reflect real differences in the health care services and documentation practices as well as differential patient access and utilization among the settings. This study demonstrates generalizable methods of classifying encounters in order to profile, compare and inform the use of secondary data across organizations.