Use of large electronic health record databases for environmental epidemiology studies.
Citation:
Ward-Caviness, C., W. Cascio, R. Devlin, L. Neas, AND D. Diaz-Sanchez. Use of large electronic health record databases for environmental epidemiology studies. The 29th Annual Scientific Conference International Society of Environmental Epidemiology, Sydney, New South Wales, AUSTRALIA, September 24 - 28, 2017.
Impact/Purpose:
This abstract examines how to use a large electronic health record database for environmental epidemiology studies. As such databases become more common they will become a valuable tool for epidemiologists, however the inherent risks, limitations, and potential biases must be carefully considered.
Description:
Background: Electronic health records (EHRs) are a ubiquitous component of the United States healthcare system and capture nearly all data collected in a clinic or hospital setting. EHR databases are attractive for secondary data analysis as they may contain detailed clinical records on millions of individuals, including many with rare conditions, with some individuals observed repeatedly over many years. However, the limitations of such databases when applied to environmental epidemiology research must be carefully considered. Method: The Carolina Data Warehouse (CDW) is a large database of EHRs from individuals who visited a University of North Carolina affiliated hospital from January 1st, 2004 onward. Since inception the number of hospitals submitting data to the CDW has steadily increased and the number of distinct patients seen in a year has grown from 125,574 in 2004 to 752,621 in 2016. Since 2004 the CDW has recorded 18,211,428 hospital visits with an average of 4.2 visits per individual per year. We will use multivariate Cox proportional hazards models to analyze the relationship between residential air pollution exposure and mortality, effectively treating the CDW as a prospective open enrollment cohort. Despite the large sample size, limitations to consider include the lack of records for out-of-state hospital visits and deaths as well as potential biases within the population with respect to insurance status and underlying disease/reason for visit. Given the size of the CDW, potential confounding and biases may be addressed via both statistical methodologies and carefully considered subgroup analyses. Conclusion: As EHRs become more prevalent and as standardized access for research purposes increases it will be important for researchers to utilize these novel resources in a carefully considered manner to understand relationships between air pollution exposure and health outcomes. This abstract does not necessarily reflect EPA policy.