Brief description of the flusurvey

Further details about the UK flusurvey can be found elsewhere [9]. In brief, the UK flusurvey (http://www.flusurvey.org.uk) is an internet-based survey that was launched on 16^th July 2009. The survey was publicised via press releases, radio and TV appearances and newspaper articles. There were no restrictions on who could register and volunteers could register throughout the season. The study was approved by the Research Ethics Committee at the London School of Hygiene and Tropical Medicine and conformed to the Helsinki Declaration on ethics of Medical Research. Participants gave their informed consent to take part and were told that anonymised data would be analysed for ILI trends by members of the study team.

The survey consisted of a background questionnaire completed upon enrolment and a weekly symptoms questionnaire (see additional file 1 and [9]).

Volunteers received a weekly email newsletter, which encouraged them to complete the symptoms questionnaire. The symptoms questionnaire asked participants to record their recent ILI and respiratory symptoms, if any. Individuals reporting any symptoms were asked date of symptom onset and whether they had phoned or visited a GP or other medical professional.

Estimated healthcare usage from the flusurvey

As national ILI consultation numbers reflect the number of GP consultations by phone, in person and via NPFS, we included all contact with medical services in the analysis. In practice, as few users reported contact with health services other than their GP and the NPFS, we used the proportion that sought no medical attention to avoid double counting for multiple contacts with medical services. Symptom reports were grouped by date and participant gender, age and "at-risk" classification. The mean proportion (and 95% confidence intervals) of ILI cases seeking medical attention, stratified by date, age, gender and risk group, were calculated by bootstrapping. Differences between groups were tested for significance using a non-paired t-test.

To estimate the healthcare-seeking behaviour of a typical ILI case during the epidemic, we weighted age- and gender-estimates of healthcare-seeking behaviour by the age distribution of cases.

Adjusting real-time numbers and estimates of severity

To estimate the number of cases of symptomatic ILI, the number of ILI consultations reported via RCGP and NPFS, stratified by week and age and adjusted by virological positivity rate, was divided by the fraction of flusurvey participants with ILI who sought medical attention, also stratified by age and week (see equation 1).

(1)

l _w,a is the number of ILI consultations in week w for age group a estimated from RCGP and NPFS data; ρ _w,a is the proportion of ILI consultations that were lab confirmed as H1N1 (in week w for age group a) estimated through virological surveillance; is proportion of individuals aged a with HPA case definition ILI that sought NO medical attention during weeks [w - 1, w + 1], estimated using the flusurvey.

Because of variability in the data, we used a three-week rolling average of healthcare usage. We compared the flusurvey-adjusted case numbers to the current best HPA estimates and real-time estimates reported by Donaldson in 2009 [2]. CFR by age was re-calculated from the ratio of H1N1v deaths to flusurvey-adjusted case numbers for the period up to 8^th November 2009 to allow comparison with previous estimates [2].

Flusurvey participation

There were 5,738 registered flusurvey participants, approximately two thirds of whom were female; 54% were between 25 and 44 years of age [9]. 21.6% of respondents self-identified as being at higher risk for influenza-related complications (chronic heart disease, diabetes, asthma, chronic lung disease, pregnant, immuno-compromised or other chronic disease).

Volunteers were more likely to report ILI symptoms during their first survey than in subsequent surveys, however retention in the study was over 80% after a participant had completed the survey three times (additional file 2).

Between July and December 2009, the symptoms questionnaire was completed 20,901 times with representation from all UK regions and broadly across all age groups. Individuals reported no symptoms on approximately 60% of occasions. Of the 40% of reports with at least one symptom, 1,522 reports matched the HPA ILI case definition [11].

Age-associated patterns in healthcare usage

We observed age-dependent healthcare usage during the autumn wave (September to December) of the 2009 H1N1v pandemic (Figure 1a). The greatest difference was between 45 to 64 year-olds, among whom 19% [12%, 25%] of respondents with ILI sought medical attention, and 0 to 24 year-olds, among whom 39% [29%, 49%] of respondents with ILI sought medical attention (p < 0.001). 25 to 44 year-olds showed intermediate behaviour with 32% [26%, 37%] of respondents with ILI seeking medical attention during the autumn wave. The small number of respondents aged over 65 with ILI meant that we were unable to provide detailed estimates of healthcare usage by month, and we found no significant differences between July-August and September-December for this age group.

Overall, we observed that females with ILI were slightly more likely to seek medical attention than males, but the difference was not significant. An exception was males under the age of 25 who were more likely to seek medical attention than females under 25 (Figure 1a). Although there was a significant difference in general healthcare usage between individuals with a self-identified risk factor and those without, we found no significant difference when we compared the healthcare usage of users with ILI.

Changes in healthcare usage during the epidemic

Healthcare usage changed significantly from month to month during the epidemic (Figures 1b, 1c and 1d). Individuals over 25 years old with ILI decreased their healthcare usage significantly from August onwards: in July 43% [41%, 45%] of ILI cases sought medical attention, whereas between August and December this decreased to 25% [13%, 37%] (p < 0.0001, Figure 1b). In contrast, healthcare usage for individuals with ILI under 25 peaked in October and this is reflected in the overall pattern of healthcare usage. At the peak of the summer wave (week beginning 20^th July 2009), 43.1% of ILI cases sought medical attention, whereas at the peak of the autumn wave (week beginning 19^th October 2009), 34.5% of ILI cases sought medical attention (Figure 1c). The changes in healthcare usage during the autumn wave were not observed in users with other symptoms not matching the ILI case definition (labelled 1+ symptom in figure 1c), in whom medical attention was sought on an average of 7% [6.4%, 7.5%] of occasions. We observed that in general, the propensity to seek medical attention for ILI increased during the peak of epidemic and decreased during holiday periods.

Re-estimating case numbers and disease severity

Using the healthcare-seeking behaviour estimated from the flusurvey, we estimate that there were 1.1 million symptomatic cases in the UK in 2009 (95% CI = [860,000, 1,600,000]) (Figure 2a). We estimate that the epidemic peaked in the week beginning 19 October 2009 and that 72% of the cases occurred from September 2009 onwards.

Adjusting for the difference in healthcare usage between adults and children shifts the burden of infection towards older age groups. We estimate that the average age of symptomatic ILI cases was 29.3 years, as opposed to an average age of 23.2 years for H1N1v cases seeking medical attention (see additional file 3).

Our increased estimates of the number of ILI cases, compared to previously published figures, result in a 35% overall reduction in CFR to 17 deaths per 100,000 cases. The decrease in CFR was most notable in adults, who were less likely to seek medical attention (Figure 2b). We estimate a CFR of 27 deaths per 100,000 cases in adults aged 45 to 64 years and from 490 deaths per 100,000 cases in adults aged 65 and over.

Strengths and limitations of the surveillance method

A strength of the internet-based approach described here is its ability to assess sentinel surveillance from an alternative perspective; we were able to access people who did not necessarily report their symptoms through any other means.

Telephone-based reporting provides an alterative to internet-based surveillance for community surveillance [14, 15]. While telephone-based reporting can control the level participation, a major advantage of web-based implementation is its ability to run throughout the pandemic and to expand to a large number of participants quickly and with little extra cost. In addition to the questionnaire, the flusurvey site provided general information and guidance about influenza and the pandemic to anyone visiting the site. Engagement with the survey was good with a cohort of regular participants, among whom retention was above 80%. We observed that volunteers were more likely to report ILI symptoms during their first survey than in subsequent surveys, suggesting that ILI might have been a stimulus for registering. It is also possible that first-time users could have reported on ILI episodes that had occurred many weeks before they registered, whereas existing users were only asked to report symptoms that had occurred during the past week. In assessing healthcare usage, it is possible that we missed participants who reported their symptoms soon after onset, as they would have had less opportunity to seek healthcare before completing the survey.

Non-probability sampling meant that we recruited several thousand volunteers within a couple of days, although it led to a bias in the age and gender distribution of users. Measured biases in recruitment could be adjusted for using the background questionnaire completed during enrolment. However, as with other active surveillance, it is difficult to quantify how the non-probability sample used differed from the general population in terms of susceptibility to influenza infection and health-care seeking behaviour. The flusurvey, as has been observed with NHS direct [16] and online ILI surveys in other countries [7, 17], had an underrepresentation of older people. It is likely that the flusurvey also suffered from the same access issues as NHS direct for individuals with low incomes, from minority ethnic groups or born outside the UK.