Grippenet, a citizen science flu-tracking project, uses the internet to allow people to contribute to public health data. But our activities on the web, on our own time, can also affect the work of epidemiologists—sometimes providing information, but also complicating matters by influencing our behavior. Analysis of search engine queries and communication via social networks could become standard tools of epidemiology.
This article is the second of two on the connections between citizen science, social networks and epidemiology.
1 - What works in citizen science? Grippenet reveals strengths and challenges
Vittoria Colizza spent months following the travels of cows. The epidemiologist with INSERM (France’s National Institute for Health and Medical Research) was tracking the movement of livestock throughout Italy and the corresponding spread of infectious disease. The conclusions drawn from this data could lead to better animal health surveillance systems. When the subjects of such a study are human, though, the situation is clearly more complicated. In the case of the flu, for instance, not only do people move about at will, carrying their germs with them, but there is also a behavioral component at play in their decisions and actions. This effect is difficult to quantify and may be amplified by today’s array of social networking opportunities.
During the 2009 H1N1 pandemic, in the wake of poor communication and scandal surrounding vaccine production, many people developed negative feelings toward vaccines, Dr. Colizza explains. “Nobody expected the uptake of vaccines to be so low, and that changes the effectiveness of the intervention. This was a case of a relatively mild pandemic, where we have models that do quite a good job (predicting outcomes), but it’s hard to incorporate into a data-driven model this change of feeling in people. We don’t have data on this stuff, how that sentiment spreads.”
One way it might spread is via social networking sites. Several studies have already examined the H1N1 pandemic through the lens of Twitter (for example, here or here.) But, in some ways, the explosion of online communication has made the epidemiologist’s job harder: tweets contain information, yes, but ever-more data does not necessarily make it easier to extract meaningful conclusions. At the same time, Vittoria Colizza is convinced that social networks hold information that is different from other sources and will prove useful. “Before, people just went to the doctor and we used that mechanism to get information. Now people are using other ways to communicate that they’re sick. We also need to use those tools, not to replace the current systems, but because it’s a different aspect of this reality with multiple faces.”
Understanding your tools: context & scale
To diversify the array of epidemiological tools, it is important to understand the validity of each—in what context and at what scale its information can be trusted. Google Flu Trends, for instance, tracks internet searches for flu-related terms, as an indicator of infection rate. Google itself performed a study evaluating the performance of this tool in the context of the 2009 H1N1 pandemic. The results showed that Google Flu Trends’ methods are valid in a normal situation of seasonal flu. “But in a big pandemic, panic at the start could alter your results, because lots of people are googling without having symptoms,” explains Dr. Colizza. The number of doctor’s visits also increases in this scenario, “but Google is even more susceptible, because you don’t have a high barrier to overcome.” Punching a few words into the search bar is a lot less trouble than making an appointment with a doctor.
Google’s effort seems useful, then, on a large scale, to collect a lot of data without recruiting a soul, but to get at factors with a behavioral element, contact with the individual is still necessary. An advantage of platforms like Grippenet, Dr. Colizza’s online project to survey citizens for the flu, is also in their flexibility. If something new arises—a particularly severe symptom, maybe—Vittoria’s team will be able to respond quickly with new questions for their users, tracking their reactions and the evolution of the pandemic. One tool is not better than the next, for Dr. Colizza; what’s important is a panel of different information sources.
Flu tracking in the UK during the 2009 flu pandemic made this last point clear. Traditionally, data comes from a system of “sentinel” doctors, reporting on flu rates among their patients. From these numbers, the rate of infection in the general population is extrapolated. In this case, though, it turned out that the sentinel system data was misleading. At the beginning of the pandemic, due to peer pressure or communications, more people were scared into visiting their doctor. “By winter, people understood more, they knew about misinformation, vaccines had come, so they were less afraid.” All told, the winter pandemic wave produced more flu cases, but fewer doctors’ visits. “If you only look at these visits, you can get the wrong number for the rate in the general population.”
It’s clear that complementary ways of obtaining information on people’s health, their beliefs and fears, and reasons for their actions would be extremely helpful for accurately modeling infectious disease. “We don’t have a clear idea of how this works, how sentiment spreads. We think it involves social networks, but we don’t have the data. I want to look into this over the next three years, with Grippenet. This is really to provide additional information. With these different tools, we need to learn from one, and adapt it to another.”
Find out more :
Vittoria Colizza, Inserm, "Epidemiology, Information Systems, Modeling" group
"More Diseases Tracked by Using Google Trends", Emerging Infectious Disease, August 2009
Grippenet, the online, citizen science, flu-tracking project
Adam Sadilek, Researcher in "computational analysis, modeling, and prediction of complex phenomena, namely human behavior exhibited both off-line and on-line."
"AI predicts when you're about to get sick", New Scientist