Over a century and a half later, epidemiologist John Brownstein doesn’t have to knock on a single door. Rather, he tracks the spread of infectious diseases by analyzing the symptoms that people search for on the Internet. Brownstein’s photo on the Children’s Hospital Boston’s website shows a serious-looking man in his mid-30s wearing a button-down shirt and tie and a professional haircut. But on this early fall morning, he’s in jeans, shaggy haired and drinking from a jumbo cup from Subway. His relaxed look and demeanor belie the seriousness of his work—spotting disease outbreaks before they become epidemics.
As a disease travels through a population, the number of people affected looks like a bell curve. Early on, only a few people fall ill each day, but as time passes, the number of new victims climbs. At its peak, the population is well-infiltrated, and there are fewer people around to get sick, so the disease slows down, infecting fewer and fewer people each day until it trails off to pre-outbreak levels. Though there is yet no crystal ball to predict outbreaks, early detection could dramatically flatten and shorten the bell curve, reducing the peak number of infected individuals and the duration of the outbreak. Early detection means early public warning: physicians prepare themselves to treat symptoms, tell the sick to stay indoors, and tutor the healthy in how to avoid contracting the disease including getting vaccinated. These efforts can’t eliminate outbreaks, but can reduce illness, and death.
For the last ten years, the CDC has also been experimenting with the use of informal disease reports to improve disease surveillance, but Brownstein’s team is pushing things one step further. He has just completed his second test-run with of a new tool for the early detection of contagious disease—tracking Google searches. Brownstein and his colleagues partnered with the CDC and Google.org, the company’s philanthropic arm, to analyze disease symptoms typed into Google search and look for patterns that matched the actual symptoms of a particular disease. They began with flu tracking and found that spikes in searches for symptoms like fever, nausea, and body aches, correlated with 90 percent accuracy to confirmed cases of flu from formal reporting sources. Brownstein’s team developed a mathematical algorithm that collects and analyzes search terms as people enter them, allowing potential flu cases to be identified on the spot. And because searches are associated with an IP address, the searcher’s location can be pinpointed down to their zip code. Google strips the search data of all other identifying information in order to protect individuals’ privacy.
When search criteria match flu symptoms, a pinpoint is added to Brownstein’s map. If a large number of cases cluster in one area, it signals the beginning of an outbreak. “By detecting disease early, you can buy yourself a week or two to get ready,” says Taha Kass-Hout, a deputy director with the CDC’s Public Health Surveillance Program Office. “You can catch up with your resources and direct your public policy and communication to warn and treat the public. You can delay the disease’s progression.” In 2008, Google Flu Trends launched its first map based on Brownstein’s work with the search term data. The CDC has incorporated Flu Trends into what Kass-Hout calls its “surveillance mosaic,” using the new technique in combination with other informal and formal reports. He says that the public also benefits by having direct access to up-to-date disease information, empowering them to make important health decisions. The flu tracking tool is so new that the CDC has not yet calculated the number of cases of illness prevented, “but we do see great potential in the new technology,” says Ashley Fowlkes, an epidemiologist with the CDC’s National Center for Immunization and Respiratory Disease, Influenza Division.
Last year Brownstein’s team began developing a similar tool to track Dengue fever. The CDC estimates that one-third of the world’s population is at risk of contracting Dengue, a mosquito-borne illness characterized by sudden high fever, severe headache and pain behind the eyes, rash, muscle and joint pain, and bleeding. Though the disease is tropical and therefore rare in the continental United States, it is on the rise even here as mosquito populations move north due to climate change. In 2010, the CDC reported the first US case of Dengue since 1945--in Florida. Dengue virus mutates as it is carried by mosquitoes from person to person, and this constant transformation of the virus makes vaccine development very challenging. But the sooner researchers can reach an infected population and study the disease, the better the chance of developing an effective vaccine. Earlier this year, Brownstein’s team showed that the algorithm used to track flu also works with Dengue fever, and Google launched Dengue Trends. “These tools do not predict whether or not a disease will hit, but they do detect its progress in real-time,” says Corrie Conrad, a program manager at Google.org who worked on the Flu Trends team.
Though Brownstein’s new tracking techniques can reduce the lag time between onset and detection of an outbreak, all parties agree that they will never replace formal reporting methods. The reason is the informal nature of the data used to make the Google-based early detection. Suspected disease cases are never confirmed by a physician and not all symptom searches necessarily indicate actual illness, so not all of the pinpoints on Brownstein’s maps will be true cases of disease. For example, you never know whether someone is searching out of panic or for the symptoms of a relative in a distant country, says Justin Stoler, a spatial epidemiologist in the geography department at San Diego State University. For this reason, Brownstein’s team monitors their data to minimize erroneous outbreak spikes. False spikes in search term frequency can be distinguished from an actual outbreak spike because the rate of increase of the search term is much faster than the rate at which the disease normally spreads through the population. In a sense, panic is a much faster-moving virus than flu.
The problem is, when you’re monitoring the data in real time, it may take a while to see the difference in rate between an outbreak spike and a panic spike, says Laura White, associate professor of biostatistics at Boston University’s School of Public Health. “You don’t want to be constantly sounding the alarm because people may stop listening.” White also points out that flu symptoms overlap with a lot of other diseases, so it’s often hard to distinguish it from a common cold, especially without a doctor’s diagnosis. “It’s a good monitoring tool for the flu,” she says, “but I haven’t seen a lot of evidence yet that it’s a good outbreak detection tool.”
Another problem is that not everyone has access to the Internet. While 65 percent of Europeans are online, and 79 percent of Americans, only 9.6 percent of Africans are, the International Telecommunication Union reported last year. But access to the Internet will not be a hindrance for long, says Kass-Hout. Advancing mobile phone technologies are the key. Over 90 percent of the world’s population has access to mobile phone networks and nearly three quarters of all subscribers live in the developing world. Of course cell phone subscription does not guarantee access to the Internet. For that, one needs the money to pay for both a smart phone device and the monthly data plan. Plus, not all countries have access to a 3G network, the minimum technology required to support mobile Internet access. But at least 130 of 196 countries in the world do have access and that number continues to grow. Mobile subscriptions in developing countries are one tenth the cost of fixed Internet connections and the infrastructure is far easier to install, so the potential exists for even the poorest individuals in the least-developed countries to get online in the very near future. Just as with fixed Internet searches, location can be linked to mobile Internet searches by tracking the location of the cell towers the phone is using at the time of the search.
Until then, even mobile subscribers without smart phones can benefit from instant access to public health information. During the 2009 H1N1 swine flu pandemic, residents of many Mexican towns received text messages from public health officials asking whether they were experiencing flu symptoms. Many people responded and officials were able to track the pandemic’s progress through Mexico. In return, the ill were sent medical advice on how to treat their symptoms. “I think that these new tools have a lot of promise,” says White, “but there are still a lot of hurdles to overcome since this data can be challenging to work with. That said, I think that innovative approaches, such as this, are the way we need to go in surveillance.”
Kass-Hout agrees that Brownstein’s flu and Dengue tracking techniques are works in progress. But as they evolve and improve, he says other Internet-based sources of data from the internet such as social media sites, could be used to track and target even non-contagious public health problems. Brownstein and the CDC are already experimenting with using these data sources to track depression and mental illness. And Marcel Salathé, a computer scientist and biology professor at The Pennsylvania State University, is interested in behavioral patterns. He recently showed that Twitter updates could be used to track anti-vaccination sentiment in the US during the 2009 flu pandemic. “I don’t want to stretch the analogy too far, but behaviors can sometimes spread as if they were a disease,” he said.