Predicting patient admission from the emergency department using administrative and diagnostic data

TBRHSC-DS-Predicting Patient Admission

Thunder Bay Regional Health Sciences Centre (ED)

Emergency department (ED) overcrowding is a growing problem in Canada. Many   interventions have been proposed to increase patient flow. The objective of   this study was to predict patient admission early in the visit with the goal   of reducing waiting time in ED for admitted patients. ED data for a one-year   period from Thunder Bay, Canada was obtained. Initial logistic regression   models were developed using age, sex, mode of arrival, and patient acuity as   explanatory variables and admission yes or no as the outcome. A second stage   prediction was made using the diagnostic tests ordered to further refine the   predictive models. Predictive accuracy of the logistic regression model was   adequate. The AUC was approximately 81%. By summing the probabilities of patients   in the ED, the hourly prediction improved. This study has shown that the  number of hospital beds required on an hourly basis can be predicted using   triage administrative data.

Authors: David Savage, Douglas G. Woolford, Mackenzie Simpson, David Wood, Robert Ohle

David Savage -

Preliminary data gathering / Baseline