Potential for early warning of viral influenza activity in the community by monitoring clinical diagnoses of influenza in hospital emergency departments

For the five-year period 2001 to 2005, time series were assembled of ED visits assigned a provisional ED diagnosis of influenza and of laboratory-confirmed influenza cases in New South Wales (NSW), Australia. Poisson regression models were fitted to both time series to minimise the confounding effects of trend and autocorrelation and to control for other calendar influences. To assess the relative timeliness of the two series, cross-correlation analysis was performed on the model residuals. Modelling and cross-correlation analysis were repeated for each individual year.Using the full five-year time series, short-term changes in the ED time series were estimated to precede changes in the laboratory series by three days. For individual years, the estimate was between three and 18 days. The time advantage estimated for the individual years 2003–2005 was consistently between three and four days.Monitoring time series of ED visits clinically diagnosed with influenza could potentially provide three days early warning compared with surveillance of laboratory-confirmed influenza. When current laboratory processing and reporting delays are taken into account this time advantage is even greater.Early detection is crucial for achieving effective control of outbreaks and epidemics of influenza. Monitoring of clinical and other health data streams which are available electronically in real-time or near real-time is now recognised internationally, nationally and locally as an essential complement to established mechanisms for public health surveillance [1-4]. Several studies have explored the potential utility of surveillance using various data streams for influenza monitoring [5-12]. Syndromic surveillance using emergency department (ED, or 'emergency room') data streams is becoming a popular method of monitoring disease activity [13-19].While some studies have evaluated syndromic surveillance against continuous laboratory series [20,21], none have assessed the timeliness of ED-