Our results show that online ILI surveillance can be an effective tool for measuring ILI incidence, and that volunteers are willing to take part in ILI surveillance. The analyses shown in this paper highlight the advantages of community surveillance of ILI compared to traditional GP surveillance techniques. Six percent of all symptom reports were positive for ILI in the 2012-2013 flu season based on a sample of 4523 participants. When standardised by age to the UK population, this corresponds to an incidence of 6.3 per 100 person weeks. As in previous years, we showed that incidence was lowest in participants over 65. This is in part because individuals over 65 were more likely to be vaccinated. There were no significant regional differences. Our results show that people with underlying health conditions were more likely to report ILI, a similar result to that found using telephone surveillance in the United States .
In our study, females were more likely to report ILI. This pattern was still evident when adjusting for living with children, or for daily contact with groups of children (which would include teachers and nursery workers). This same pattern was observed in a study of healthy adults in Australia . We speculate that females may be more likely to take care of symptomatic individuals, and therefore be exposed to illness or that the different influenza rates in females may be due to physiological differences .
The highest rates of ILI were in the youngest age category, in agreement with other community-based findings . The decline in incidence is coincident with Christmas holidays, and the resurgence of cases in this age group in the new year, after schools had reopened, is similar to other data [7, 19–21]. The observation that incidence in the 19-45 year olds lagged behind the youngest age group, suggests that under 18s may have brought the illness back to their families.
Our results also show that taking public transport does not increase your risk of reporting an ILI regardless of definition used. Analyses from other European countries and over different seasons confirm this finding [Van Noort S, Codeço C, Kopperschaar C, Van Ranst M, Gomes M: Influenzanet: ILI trends, behaviour and risk factors in cohorts of internet volunteers, submitted, , though others have found the opposite .
Our results suggest that influenza vaccination gives some protection against ILI. With any definition of ILI considered, not being vaccinated was the greatest predictor of reporting an ILI. This result is consistent with results from previous years . Some ILI was still reported in participants who reported vaccination. This can in part be attributed to the fact that the vaccine is only effective against influenza and we are measuring ILI some of which will not be due to influenza. Restricting the analysis of our data to the largest peak of virologically confirmed influenza (9th of December through 30th of December), the OR for being unvaccianted increased to 2.14 (95% CI 1.68, 2.74).
Strengths and weaknesses of the study
The use of internet based surveillance has a number of limitations.
All of our data are based on self-reports of symptoms. As such we are unable to comment on influenza, only on ILI. However, most GP surveillance is also based on ILIs since most cases are not virologically confirmed. There is little information on the specificity and sensitivity of ILI during the influenza season. As shown by Figure 1, our measured incidence of ILI corresponded well to virological surveillance recorded by PHE. Ideally, weekly swabbing of Flusurvey participants would be used to assess the proportion of ILI cases that are confirmed influenza. Currently the costs of swabbing have not allowed this. However we would like to attempt this in the future.
Our sample does not represent a random sample; by nature of its design it is biased towards internet users, and people who are willing to fill in surveys. Additionally Flusurvey participants are more likely to be from London, be female, have risk factors and be vaccinated than the UK general population. Other methods such as telephone surveillance can be used to overcome these limitations [15, 24], however these surveillance techniques are more expensive, time absorbing, have representativeness problems of their own, and still have the problem of self-reports of ILI.
Our decision to exclude first reports in incidence estimates may have caused us to underestimate the overall amount of ILI in our sample. When including first reports, 7.8% (95% CI 7.6, 8.0) of total symptom reports were positive. When removing first reports 6.0% (95% CI 5.7, 6.2) were positive. However, the survey asks participants if they had had any symptoms since the last time they logged in, so participants’ first reports may reflect symptoms that occurred outside the current flu season.
Our risk factor analysis is restricted to the questions that are asked at the beginning of the survey. Our multivariate analysis suggests that being under the age of 18 is not a risk factor for reporting ILI. However, being in contact with groups of children is. Similarly, living with children was not seen to be a risk factor, but this is colinear with being a child. As a result, due to the wording of this question, it is difficult to understand the implications of these results. If we had asked the question in another way, for example, are you an adult who works with children, we may have been better able to separate the effects of working with children versus being a child. It is plausible that living with children is a risk factor, but the colinearity with being a child masks this relationship. Sensitivity analysis showed no substantial difference in results when children were removed from the analysis.
Flusurvey does have many strengths. In effect the survey consists of an online cohort. We were able to attribute illness directly (although anonymously) to individuals, thus understanding their individual risk factors. This is in comparison to other online influenza tracking sights such as Flu Near You (https://flunearyou.org/) which is run in the United States. They do not ask background questions about individual risk factors, and can therefore only report on influenza prevalence. Flutracking (https://www.flutracking.net/), based in Australia, only asks about gender, age, and working with patients. Only Flusurvey and the other members of Influenzanet ask in-depth questions about risk factors. The data are available in real-time allowing rapid examination of these risk factors and the effectiveness of control programmes. Finally, the size of the cohort can be expanded at very little additional cost.