PREDICTING EMERGENCY DEPARTMENT CENSUS USING TIME SERIES REGRESSION

Open Access
Author:
Sheetz, Nathaniel Cloud
Graduate Program:
Industrial Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
None
Committee Members:
  • Deborah Jean Medeiros, Thesis Advisor
Keywords:
  • regression
  • forecasting
  • patient census
  • emergency department
  • time series
Abstract:
Hospital emergency departments (EDs) in the United States face increasing patient demand even as capacity declines, causing crowding and negatively affecting the patient experience. Due to variable patient arrival rates and hospital conditions, particularly the availability of beds for those who require inpatient treatment, ED census—that is, the number of patients in the ED—can vary hour to hour and day to day. If ED administrators had forecasts of the number of patients expected to be in the ED several hours in advance, they could potentially alleviate some of the problems associated with crowding by making better staffing and resource allocation decisions. In this thesis, time series regression models are developed to make short-term forecasts of ED census at the ED of the Hershey Medical Center. Multiple linear regression models were developed using one month of hourly data that includes only factors that are readily available in real time to ED administrators. The models generate predictions of ED census up to four hours in advance. When tested against two months of out-of-sample data, all models had mean absolute error between 2.63 and 2.83 patients per hour.