Registration for the UK Flu Survey began in mid-July. We now have over 5500 registered users from all over the UK. The locations of recent participants are shown in the map on the right. Those whose symptoms are consistent with Influenza are shown in red and those without respiratory symptoms are shown in green. Because users only record the first half of their postcode, points on the map do not indicate user's households. Click here to see an interactive map of the UK results so far (may be slow to load). Click here to plot your own results

Joining the dots

Since the summer, we've been asking flusurvey users about their symptoms and their social contacts. The aim was to use information about social mixing patterns to help predict and explain epidemic behaviour. Below, we've plotted a measure of how the epidemic grew (i.e. how much it changed from week to week) against the average number of social contacts each person reported in that week.

When using adults' contact behaviour there is no clear trend, but contacts reported by children (aged 18 and under) give a much clearer picture - when children made few contacts the epidemic declined, but when they made many contacts it grew. Children reported many fewer contacts during school holidays, which explains why patterns of school terms had such a strong influence on the pandemic.

What the graphs show: each red dot shows one week's data. The blue line is the "best fit" to these dots; the grey region shows the uncertainty surrounding this best fit line - on the left, we can't be certain that the line slopes downwards, but on the right we can be fairly sure that there is a positive connection.

Measuring influenza-like-illness

Using a list of symptoms to measure flu is not simple - there are plenty of infections around that can look like flu, and plenty of people with flu who do not display the "classical" symptoms (high temperature, sore throat, runny/blocked nose, aches and pains, tiredness...). Some definitions of influenza-like-illness (ILI) are very strict - requiring lots of symptoms before deciding that it's ILI - and others less so. You can see in the figure above what happens when we apply these different measures to information provided by flusurvey users.

Without laboratory testing, it's hard to tell which measure is best, but what we can do is compare the different options to the best estimate of the number of cases (red lines), and see which gives the best match. We know we won't get the numbers spot on (because flusurvey users aren't a random sample of the population), but we'd like to be able to match the shape. As you can see, there's still work to do.

One problem is that the definition really ought to change from week to week, depending on what other infections are around (in the summer, most things that looked a bit like flu probably were flu, whereas in the winter there are lots of other infections cropping up too) - this might be too much of a challenge, but it'll keep us busy during the closed season.

Adjusting for sample characteristics

An internet-based survey will always struggle to obtain samples from all parts of society. Flusurvey is no exception: as the figure below shows, adult women are over-represented in the sample, and we have very few children (proportionately). 

This means that the incidence that we report may be biased. The comparative lack of children in our sample is particularly problematic, as they seem to be at higher risk of acquiring flu. There are two ways to account for this. First, to encourage more children to fill out the survey (or parents to fill it out on their behalf). This is clearly the best approach. An alternative method is to adjust for the sample characteristics, by weighting our sample appropriately. The figure below shows the impact of this - by adjusting for the sample characteristics the incidence estimate is increased, as reports from children now have more weight.

Holidays and contact patterns

Since July 2009, we've asked flusurvey users to report their social contacts - this should help us link social mixing behaviour to risk of infection. We've seen in the pattern of estimated swine flu cases that school holidays have had a big impact on transmission, and we can look to see whether this can be explained by changes in contact patterns.

Some flusurvey users have been especially keen, and have recorded their social contacts almost every week. Below are shown how these change over time for a handful of these people (each colour being a different person).

There are clearly differences between people, and the variation makes it hard to see any trends over time. However, when we combine the reports from all users who have completed the contact survey on 12 or more occasions, much clearer patterns emerge:

The red line shows the "middle" person for each week, while the green and blue lines show where the upper and lower 25% fit in. We can see very clear changes at the end of the summer holiday, during school half term, and during the Christmas holiday period - at these times flusurvey users are away from work so make fewer social contacts.

All this ties in extremely nicely with the epidemic data, and, encouragingly, shows the usefulness of our simple contact surveys. We are interested in seeing how patterns will change now that people are back to work (and now that the snow has melted), so we'll keep and eye on this in the weeks to come.

More flusurvey results 

The answer is that we can't know for sure - symptoms vary too much from person to person. The tool that we use to determine whether someone has flu-like-symptoms has been developed over several years by our collaborators in Holland, Belgium, Portugal and Italy. It correlates closely with physician-diagnosed seasonal flu (in technical terms it is quite specific to flu). It is possible that some people who recieve a "respiratory symptoms" diagnosis actually have swine flu (technically, the sensitivity of the diagnosis may not be ideal). People with swine flu may display a different range of symptoms to those that have normal seasonal flu, which is what our diagnosis is based on. As we collect more data, we may revise our definitions.

(If you feel ill and are worried that you might have contracted swine flu, then you should stay at home and contact the health services: click here for health service contact details).