Jeffrey Shaman thinks that one day, your weekly weather forecast will predict much more than the chance of rain. In the not-so-distant future, he says, the odds of the flu sweeping through your city could be displayed right next to those cartoon clouds and raindrops.
Seasonal flu kills between 3,000 and 49,000 Americans each year, according to the Centers for Disease Control and Prevention. Outbreaks happen every winter, typically peaking in January and February. But the ability to predict the exact timing, magnitude and duration of an outbreak would help public health officials take preventative measures, such as closing schools or distributing vaccines to high-risk areas.
“The way they do it now is reactive — they don’t have any idea of how an outbreak will progress,” says Shaman, who studies meteorology and infectious disease transmission at Columbia University. Shaman and his colleagues have just created the most accurate flu forecaster yet, and it may become a useful tool for predicting and mitigating the spread of influenza through urban areas.
The research is based on a model that Shaman’s team used to retroactively predict flu outbreaks in New York City between 2003 and 2008. The new study, published in Nature Communications, used a similar model to predict the 2012-2013 outbreaks in real time as they occurred across 108 cities.
Starting with their original model, the researchers updated it every week of the 2012 flu season with new data from the Centers for Disease Control and Prevention and Google Flu Trends, which estimates flu activity in real-time based on the use of flu-related search terms. Refining the model based on real-time data made it more accurate than any previous model. By mid-December 2012, it accurately predicted the peak timing of 63 percent of outbreaks, and in some cases gave up to nine weeks’ advance notice.
The structure and timing of an outbreak can vary from city to city, and the paper notes that some cities were easier to forecast than others. Smaller cities like Birmingham and Buffalo could be accurately forecasted throughout the flu season, whereas more sprawling cities such as Chicago and New Orleans gave the model more trouble. In other cities — San Diego, Atlanta, Boston — the predictions became more accurate as the season progressed.
The researchers will now assess whether the model did a good job of predicting the magnitude and duration of the 2012-2013 outbreaks. Future versions will explore how different demographic structures and strains of flu affect outbreak predictability, and will determine what level of granularity is best. Should the model look at outbreaks at the level of counties, cities or neighborhoods? The model may also need to be modified to forecast outbreaks in less densely populated suburban and rural areas.
As the new flu season revs up, the team plans on making 2013-2014 forecasts for approximately 100 cities. It will make those predictions available online within the next couple of weeks.
Flu forecasting is still in its infancy, but Shaman hopes it will eventually help people take precautions to protect themselves from infection. “When I hear there’s an 80 percent chance of rain, I bring an umbrella,” he says. Similarly, if people learn there’s an 80 percent chance of a flu outbreak in their city next week, they might be more likely to get a vaccine, wash their hands, and avoid others who are sick.
It could take some time before people gain enough confidence in the predictions to take those behavioral adjustments. Modern weather forecasting, for instance, has been around for about 50 years and has improved dramatically in accuracy since it first debuted.
How many years will it take until flu forecasts are accurate enough to show up alongside the local weather and pollen reports? “It could happen in the next few years, but only time will tell,” Shaman says. “I may be in forecasting, but I don’t have a crystal ball.”
The Science of Cities column is made possible with the support of the John D. and Catherine T. MacArthur Foundation.