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

Abstract

Background: 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).

Methods: 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).

Results: 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.

Conclusion: 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.

Figure 1

Predicted Respiratory Complaints. The predicted respiratory complaints and corresponding upper confidence limits illustrates seasonality and day-of-week effects.

Figure 2

Day-of-week effects for each CC category. The average day-of-week effect with corresponding error bars for six CC categories.

Figure 3

Annual cycles in Respiratory Complaints. Annual cycles in respiratory complaints (by week) for the past three flu seasons, from 2000–1 through 2002–3.

Figure 4

Average daily number of respiratory complaints by calendar year. The average daily number of respiratory CCs decreases over the training data.

Figure 5

Results on validation data for Respiratory Complaints. Prospective data (June 1, 2002 through May 31, 2003 (a) Daily, predicted, and upper control limit for respiratory counts ; (b) Scaled forecast errors for respiratory counts; (c) Page's statistic applied to the same forecast errors. A control value of 3.3 bfor Page's statistic results in an approximate theoretical 2.5% false alarm rate when forecast errors are Gaussian.