A potential measure of flu severity? Google Flu Trends

[Vibrant62]

During the outbreak of pandemic H1N1, the hardest things to assess with any degree of accuracy was flu severity and the extent of underlying infection in the general community.

The 'hard' measures available in any flu season are the numbers of ILI referals to GPs, the numbers of hospitalisations, and the numbers of fatalities - however there is no accepted measure of the potential numbers of mild infections, for which individuals may self treat and not refer for further medical assistance. It is this figure that is required for any relative measure of flu severity.

I have long studied the Google Flu Trends tool, and as imperfect as it is, believe that this is probably the closest measure we have of underling rates of ILI in general populations. This data can be downloaded to provide numbers of queries by geographical region.

Therefore I would suggest that an approximate monitor of flu severity could be gained by determining an ongoing and comparative relative ratio of flu related searches via Google Flu trends divided by the numbers of consultations; the numbers of hospitalisations and the numbers of deaths in any given period or season.

What do others at FT think of this, in principle?

What I find striking at present in Australia is reports of high numbers of consultations and hospitalisations relative to an overal low incidence of flu as measured by Flu trends. This would tend to indicate viral infections of increased severity in Australia at the moment.

e.g From Cidrap
Aug 16, 2011

Australia sees continued high flu levels
Flu activity in Australia continues to increase, with rising case numbers not confined to Queensland, New South Wales, and South Australia states, which have been the hot spots over the past several weeks, according to a report today from the country's Department of Health and Aging. Its latest surveillance report covers the week ending Aug 5. Most virus detections have been the 2009 H1N1 strain, but influenza B is circulating and has been dominant in some areas, such as Tasmania. Flu activity in four areas?Australian Capital Territory, New South Wales, Queensland, and Tasmania?has surpassed 2010 peaks. Nationally, the weekly number of confirmed cases is above annual peaks for previous years, except for 2009. The health department said the reason for the unusually high activity isn't clear, but the numbers don't appear to be only a reflection of increased testing. The flu season in the Southern Hemisphere typically runs from May through October.
Aug 16 Australian flu surveillance report
http://www.cidrap.umn.edu/cidrap/content/influenza/swineflu/news/aug1611fluscan.html

VERSUS

http://www.google.org/flutrends/au/#AU

Comments welcome: could this be a useful measure before I start number crunching??

 

[Laidback Al]

Re: A potential measure of flu severity?

Re: A potential measure of flu severity?

The Utility of Google Flu Trends

I think that comparing ?hard? data with soft data (word searches on Google) requires so many assumptions that any results would be suspect. Certainly, number of deaths, high numbers of consultation, and numbers and length of hospitalizations relative to the total population will provide a good measure of severity.

The problem with Google Flu Trends is that it is only measuring interest, not measuring flu severity. As a statistic it is only measuring the relative increase or decrease in online searches. Sure, as the numbers increase there are probably more symptomatic people, but how do you control for and separate out searches based on actual cases of influenza versus cases of common rhinitis and other sicknesses confused with influenza. Influenza and rhinitis are seasonal so there needs to be control for seasonality on the numbers.

Google posts some optimistic expectations about Flu Trends in the quotes below:

These graphs show historical query-based flu estimates for different countries and regions compared against official influenza surveillance data. As you can see, estimates based on Google search queries about flu are very closely matched to traditional flu activity indicators. Of course, past performance is no guarantee of future results. Our system is still new, but we hope to see similar correlations in the coming years.

It is not surprising that as influenza sweeps across a population that people search for information about care and protection. A generation ago, people consulted libraries or their physicians, now they search on line. So a correlation between increases in influenza cases and increases in information searching is not new, it is just more publicly apparent with Google Flu Trends.

So why bother with estimates from aggregated search queries? Traditional flu surveillance is very important, but most health agencies focus on a single country or region and only update their estimates once per week. Google Flu Trends is currently available for a number of countries around the world and is updated every day, providing a complement to these existing systems.

Influenza outbreaks ebb and flow within cities, regions, and countries. Local and regional public health officials need to be concerned about the situation in their area. Aggregate information about influenza searches from countries on the other side of the world do not assist public health officials in doing a better job in their local community.

For epidemiologists, this is an exciting development, because early detection of a disease outbreak can reduce the number of people affected. If a new strain of influenza virus emerges under certain conditions, a pandemic could ensue with the potential to cause millions of deaths (as happened, for example, in 1918). Our up-to-date influenza estimates may enable public health officials and health professionals to better respond to seasonal epidemics and pandemics.

Lofty goals, indeed. But there is no indication that Google Flu Trends is a?leading? indicator in the sense that it provides enough information for public health officials to take concrete action. In fact, it is a "trailing" indicator. People only start searching once the disease is spreading locally, too late to stop the spread. Also, Google Flu Trends can not provide any advance information about the nature of the infectious agent, its virulence, its target population, its rate of transmission, etc. These are the critical bits of information that public health officials need to combat an epidemic or a pandemic.

Despite Google?s claims to the contrary, I am unconvinced that Google Flu Trends is a serious public health tool.

 

[sharon sanders]

Re: A potential measure of flu severity? Google Flu Trends

I think that Google trends is only measuring interest and is a lagging indicator.

I think if we had access to the Google search data relating to the recent democratic revolutions in the Middle East we might see interest from all over the world and not just in the countries where these protests are occurring.

 

[Vibrant62]

Re: A potential measure of flu severity? Google Flu Trends

Accepted on Google Flu Trends - but in theory at least, it should reflect an underlying level of concern or distribution of a flu outbreak, and inconsistencies could be assumed to be relatively constant over time i.e years. The key issue here was whether the RATIO between searches and hospitalisations/ consultations/ deaths could provide a real time measure that might have some utility?

At the moment the only 'hard' measure available for underlying incidence rates is a sero-conversion study - but these take too long to conduct (aside from issues of expense) to give real-time results.

I know that there have been a number of recently published studies involving twitter monitoring to provide such a measure - do you think that would be a better tool for measuring real time flu concern/ incidence levels? A measure of a regional severity might be too late to do much to assist the geography where the outbreak was occurring but could be sufficient to alert other regions to prepare.

If not any suggestions for how we could get some approximation of the real time incidence of flu in a region?

 

[sharon sanders]

Re: A potential measure of flu severity? Google Flu Trends

Further on Google trends, we have no idea what effect the "FluTrackers factor" has on their numbers. There are many who are searching for this information - worldwide: Us, Flu Wiki, ProMED, Georgetown University, others.... How much do our efforts on the internet skew the flu indicators on Google trends?

As for twitter, I think it is wholly unreliable as an indicator of anything other than general interest. We have posted over 21,000 tweets in the last 2.5 years and it is my observation that twitter is a valuable social tool but it would be impossible to make any reliable calculation of disease severity as a result of the number or content of tweets.

Groups like FT are the most reliable measurement of disease spread and severity because we act as a clearinghouse for all available data relating to disease outbreaks. It is our collection of this data and resultant analysis that is the most reliable indicator. For instance, recently the WHO reduced the number of confirmed H5N1 cases in Egypt on their cumulative H5N1 chart. We knew immediately that this must be a typo and we posted this comment on FT, twitter, and facebook. The WHO realized their mistake and corrected it.

I wish there was a reliable "shorthand" method available but it seems this is not the case.