Online Flutracking Survey of Influenza-like Illness during Pandemic (H1N1) 2009, Australia

We compared the accuracy of online data obtained from the Flutracking surveillance system during pandemic (H1N1) 2009 in Australia with data from other influenza surveillance systems. Flutracking accurately identified peak influenza activity timing and community influenza-like illness activity and was significantly less biased by treatment-seeking behavior and laboratory testing protocols than other systems.

We compared the accuracy of online data obtained from the Flutracking surveillance system during pandemic (H1N1) 2009 in Australia with data from other infl uenza surveillance systems. Flutracking accurately identifi ed peak infl uenza activity timing and community infl uenza-like illness activity and was signifi cantly less biased by treatment-seeking behavior and laboratory testing protocols than other systems.
A variety of surveillance methods were used to monitor the incidence and severity of infl uenza A pandemic (H1N1) 2009 in Australia. Severity of illness was measured by number of hospitalizations, intensive care unit (ICU) admissions, and deaths. Infl uenza disease incidence was monitored through laboratory-confi rmed cases, general practitioner sentinel surveillance of infl uenza-like illness (ILI), emergency department visits for ILI, absenteeism data from large employers, and the Flutracking surveillance system (1).
Flutracking is a national weekly online survey of ILI (completed by >8,000 participating community members each week in 2009); it is the only ILI surveillance system that provides comparable data across Australia's states and territories. Flutracking integrates participants' ILI symptom information with their infl uenza vaccination status (2). Flutracking surveillance has correlated well with other Australian infl uenza surveillance systems in describing the timing and scale of the 2007 and 2008 seasonal infl uenza epidemics (3,4). We compared Flutracking data with data from other routine infl uenza surveillance systems during the 2009 pandemic wave in New South Wales (NSW), Australia's most populous state.

The Study
From May 4, 2009, through October 31, 2010, participants received an automatically generated weekly email link to the online questionnaire, which asked whether they had experienced fever or cough and how many days they had been absent from work or normal duties because of these signs (recruitment details in 2,3; location of participants at www.fl utracking.net). Each individual response usually took <15 seconds. Participants who had previously reported not receiving seasonal infl uenza vaccine were asked whether they had received infl uenza vaccination in the prior week during each weekly survey. If they answered yes, the question was automatically deleted from their subsequent weekly surveys. Participants were permitted to enroll at any time during the surveillance period. Participants could respond on behalf of household members, and children >12 years of age could complete their own survey. During online enrollment, participants provided the following information: their month and year of birth; whether they had received a seasonal infl uenza vaccine in the preceding year; whether they worked face to face with patients in hospitals, nursing homes, doctors' clinics, or as community health workers; and their residential postal code.
The weekly proportion of participants with ILI signs or symptoms was calculated as the proportion of participants for that week who reported both fever and cough within the previous 7 days. These proportions were compared with infl uenza activity recorded in 2009 by other established New South Wales infl uenza surveillance systems, i.e., number of patients who visited emergency departments with ILI symptoms (5), laboratory-confi rmed infl uenza A antigen tests (PCR and direct immunofl uorescence) (5), Google Flu Trends data (aggregated Google search data used to estimate current infl uenza in Australia) (6), workplace absenteeism data (5), and Australian Sentinel Practice Research Network (ASPREN) general practice ILI data (7).
Surveillance data were compared with data from 2007 and 2008. NSW was selected because no other states had suffi cient Flutracking participants in 2007 and 2008 to allow year-to-year comparisons. The number of NSW participants who completed >1 survey in the 2009 Flutracking surveillance system was 3,447.
The concordance across NSW infl uenza surveillance systems was high for ILI peak weeks during the past 3 years. During 2009, Flutracking, laboratory infl uenza notifi cations, and Google Flu Trends peaked 1 week before emergency department ILI, workplace absenteeism, and ASPREN ILI surveillance (Table).
A comparison of the weekly scale of NSW Flutracking fever and cough symptom rates during 2007, 2008, and 2009 showed that the peak attack rate of 6.8% in 2009 was signifi cantly lower than that of 9.4% in 2007 and  (Table).
When the surveillance systems were compared, laboratory notifi cations and emergency department surveillance appeared to be more affected by health-seeking behavior and changes in physician's testing protocols and may not have refl ected true community ILI rates, in contrast to Flutracking, Google Flu Trends, workplace absenteeism, and ASPREN. Potential biases in laboratory notifi cations and emergency department surveillance may vary, depending on the pandemic phase. For example, during the protect phase of the pandemic, testing for infl uenza was recommended only for those admitted to the hospital for ILI or when test results could alter clinical care of a patient. Before the protect phase (during the contain phase), testing for pandemic (H1N1) 2009 virus was conducted to confi rm diagnosis for anyone with ILI.
Flutracking's fi nding of a 2009 peak ILI rate similar to those of previous years was also consistent with NSW mortality data. The number of NSW deaths attributed to infl uenza or pneumonia suggested that the 2009 infl uenza season did not result in excess overall deaths but rather a redistribution of deaths with a relative increase of deaths in younger age groups (8). The low ILI rate found by Flutracking was initially viewed with suspicion because other near real-time surveillance (laboratory notifi cations and emergency department surveillance) suggested a high pandemic (H1N1) 2009 attack rate compared with rates for previous years. However, Flutracking results were consistent with other pandemic infl uenza attack rate estimates in NSW and other countries (9)(10)(11)(12).
Because Flutracking does not rely on the health sector for ILI or laboratory reporting, it is not biased by changes in testing, treatment seeking, jurisdictional protocols, or resource constraints. Flutracking, Gripenet, and other similar Internet-based surveillance could potentially facilitate near real-time comparison of ILI activity between regional jurisdictions and among countries to assist with monitoring the global spread of infl uenza (13).

Conclusions
During the initial pandemic (H1N1) 2009 outbreak, Flutracking demonstrated its ability to accurately identify peak infl uenza activity timing and the relative magnitude of community infl uenza activity when compared with infl u-  enza tracking efforts in previous years. Its results were also less affected by treatment-seeking behavior and by laboratory testing protocols during different pandemic phases than was health system-based surveillance.