From model to forecasting: a multicenter study in emergency departments

Acad Emerg Med. 2010 Sep;17(9):970-8. doi: 10.1111/j.1553-2712.2010.00847.x.


Objectives: This study investigated whether mathematical models using calendar variables could identify the determinants of emergency department (ED) census over time in geographically close EDs and assessed the performance of long-term forecasts.

Methods: Daily visits in four EDs at academic hospitals in the Paris area were collected from 2004 to 2007. First, a general linear model (GLM) based on calendar variables was used to assess two consecutive periods of 2 years each to create and test the mathematical models. Second, 2007 ED attendance was forecasted, based on a training set of data from 2004 to 2006. These analyses were performed on data sets from each individual ED and in a virtual mega ED, grouping all of the visits. Models and forecast accuracy were evaluated by mean absolute percentage error (MAPE).

Results: The authors recorded 299,743 and 322,510 ED visits for the two periods, 2004-2005 and 2006-2007, respectively. The models accounted for up to 50% of the variations with a MAPE less than 10%. Visit patterns according to weekdays and holidays were different from one hospital to another, without seasonality. Influential factors changed over time within one ED, reducing the accuracy of forecasts. Forecasts led to a MAPE of 5.3% for the four EDs together and from 8.1% to 17.0% for each hospital.

Conclusions: Unexpectedly, in geographically close EDs over short periods of time, calendar determinants of attendance were different. In our setting, models and forecasts are more valuable to predict the combined ED attendance of several hospitals. In similar settings where resources are shared between facilities, these mathematical models could be a valuable tool to anticipate staff needs and site allocation.

Publication types

  • Multicenter Study

MeSH terms

  • Emergency Service, Hospital / statistics & numerical data*
  • Emergency Service, Hospital / trends
  • Forecasting
  • Health Services Needs and Demand
  • Humans
  • Linear Models
  • Models, Organizational
  • Paris
  • Risk Factors
  • Seasons