Email updates

Keep up to date with the latest news and content from BMC Medical Informatics and Decision Making and BioMed Central.

Open Access Research article

Modeling emergency department visit patterns for infectious disease complaints: results and application to disease surveillance

Judith C Brillman1*, Tom Burr2, David Forslund3, Edward Joyce4, Rick Picard2 and Edith Umland1

Author Affiliations

1 Department of Emergency Medicine, MSC10 5560, 1 University of New Mexico, Albuquerque NM 87131-0001, USA

2 Mail Stop F600, Los Alamos National Labs, Los Alamos, New Mexico 87545, USA

3 Mail Stop T006, Los Alamos National Labs, Los Alamos, New Mexico 87545, USA

4 Mail Stop F607, Los Alamos National Labs, Los Alamos, New Mexico 87545, USA

For all author emails, please log on.

BMC Medical Informatics and Decision Making 2005, 5:4  doi:10.1186/1472-6947-5-4

Published: 2 March 2005



Concern over bio-terrorism has led to recognition that traditional public health surveillance for specific conditions is unlikely to provide timely indication of some disease outbreaks, either naturally occurring or induced by a bioweapon. In non-traditional surveillance, the use of health care resources are monitored in "near real" time for the first signs of an outbreak, such as increases in emergency department (ED) visits for respiratory, gastrointestinal or neurological chief complaints (CC).


We collected ED CCs from 2/1/94 – 5/31/02 as a training set. A first-order model was developed for each of seven CC categories by accounting for long-term, day-of-week, and seasonal effects. We assessed predictive performance on subsequent data from 6/1/02 – 5/31/03, compared CC counts to predictions and confidence limits, and identified anomalies (simulated and real).


Each CC category exhibited significant day-of-week differences. For most categories, counts peaked on Monday. There were seasonal cycles in both respiratory and undifferentiated infection complaints and the season-to-season variability in peak date was summarized using a hierarchical model. For example, the average peak date for respiratory complaints was January 22, with a season-to-season standard deviation of 12 days. This season-to-season variation makes it challenging to predict respiratory CCs so we focused our effort and discussion on prediction performance for this difficult category. Total ED visits increased over the study period by 4%, but respiratory complaints decreased by roughly 20%, illustrating that long-term averages in the data set need not reflect future behavior in data subsets.


We found that ED CCs provided timely indicators for outbreaks. Our approach led to successful identification of a respiratory outbreak one-to-two weeks in advance of reports from the state-wide sentinel flu surveillance and of a reported increase in positive laboratory test results.