Prediction of emergency department revisits using area-level social determinants of health measures and health information exchange information. (September 2019)

• Record Type: Journal Article
• Authors: Vest, Joshua R; Ben-Assuli, Ofir
• Is Part Of: International journal of medical informatics. Volume 129(2019)
• Page Start: 205
• Page End: 210
• Publication Date: 2019-09
• Subjects: Health information exchange -- Machine learning -- Risk prediction -- Emergency department -- Social determinants -- Emergency department revisits; Medical informatics -- Periodicals; Information science -- Periodicals; Computers -- Periodicals; Medical technology -- Periodicals; Medical Informatics -- Periodicals; Technology, Medical -- Periodicals; Computers; Information science; Medical informatics; Medical technology; Electronic journals; Periodicals; Electronic journals; 610.285
• DOI: 10.1016/j.ijmedinf.2019.06.013
• Languages: English
• ISSNs: 1386-5056

Abstract

Abstract: Introduction: Interoperable health information technologies, like electronic health records (EHR) and health information exchange (HIE), provide greater access to patient information from across multiple organizations. Also, an increasing number of public data sources exist to describe social determinant of health factors. These data may help better inform risk prediction models, but the relative importance or value of these data has not been established. This study assessed the performance of different classes of information individually, and in combination, in predicting emergency department (ED) revisits. Methods: In a sample of 279, 611 adult ED encounters. We compared the performance of Two-Class Boosted Decision Trees machine learning algorithm using 5 classes of information: 1) social determinants of health measures only, 2) current visit EHR information only, 3) current and historical EHR information, 4) HIE information only, and 5) all available information combined. Results: The social determinants of health measure only model had the overall worst performance with an area under the curve AUC of 0.61. The model using all information classes together had the best performance (AUC = 0.732). The model using HIE information only performed better than all other single information class models. Conclusions: Broad information sources, which are reflective of patients' reliance on multiple organizations for care, better support risk prediction modeling in the emergency department.