Muscatello, David J., Churches, T., Kaldor, J., Zheng, W., Chiu, C., Correll, P., and Mannes, T.
Introduction: The Department of Health in New South Wales, Australia, expanded public health surveillance for the 2003 Rugby World Cup and for its ongoing counterterrorism response. Cup games were played in and around the Sydney region. Objectives: This study explains the development of an automated, near real-time, syndromic surveillance system in which data are used that are already being collected in emergency department (ED) databases for routine patient administration. Methods: Demographic, triage, physician-assigned provisional diagnosis, and disposition information from 12 of 49 public hospital EDs in the greater Sydney metropolitan area was frequently and automatically transmitted, analyzed, and reported in daily statistical summaries on Intranet websites beginning October 10, 2003. Diagnoses were categorized by syndrome, disease, and injury. Presenting problem and nursing-assessment free text routinely entered by nurses during patient triage were automatically classified into >30 syndrome categories by using automated preprocessing techniques and naïve Bayesian automatic text classification methods. The diagnosis-based categories were used to train the automatic classifier to associate words in the free text with syndrome categories. An adjusted cumulative sum (CUSUM) accumulating day-of-week differences in daily counts was used to assess the statistical significance of disease and injury trends. Public health personnel monitored the reports daily and notified the Rugby World Cup Public Health Committee of unusual trends. Results: During the tournament, October 10-November 22, health trends identified by the system were not sufficient to cause concern among public health personnel but did provide reassurance that the health of the population was not adversely affected. Data collection did not add to the work load of clinical staff in EDs, and surveillance downtime was negligible. Since the games, this now ongoing surveillance system has rapidly identified a community-based epidemic of gastrointestinal illness, an increase in recreational drug misuse, the annual influenza epidemic, and an increase in episodes of acute asthma. Variation by syndrome occurred in the degree of correlation between daily visit counts from automatically classified nurse text and the equivalent diagnosis groups used to train the classifier. Conclusion: During the Rugby World Cup, the surveillance system complemented traditional public health surveillance to provide a comprehensive assessment of health trends in the population. A substantial advantage of the system has been its ongoing sustainability. Physician-assigned diagnoses are more specific than free text for some syndrome and diagnostic categories. However, diagnoses are not available for analysis until at least the end of a patient's ED visit, and they are sometimes incomplete. Free text from patient triage is available for analysis early in an ED visit, and depending on the scope of the text description, multiple text-based syndromes can be automatically assigned to a visit. [ABSTRACT FROM AUTHOR]