Image courtesy of Men's Health

In 1900, the infamous Galveston hurricane struck out of the blue, catching the population unaware and killing between 6,000 and 12,000 people, still the largest natural disaster in US history. A hundred and 12 years later, forecasters at the National Hurricane Center tracked and predicted--with astonishing accuracy--the route that hurricane Sandy took en route to slamming into the Jersey Shore. Now, in a pioneering study just published in the Proceedings of the National Academy of Science, researchers at Columbia University and the National Center for Atmospheric Research have reported successfully adapting the same techniques used by meteorologists to predict an even more lethal force of nature: influenza epidemics.

Worldwide, the flu sickens 10 million people each year, killing between one-quarter to one-half million victims. Just as reliable weather forecasting saves lives by giving a heads-up to those in harm's way, so do the researchers hope that accurate, localized flu forecasts will help minimize the toll taken by future epidemics. (Feel a sneeze coming on? Try these 10 Ways to Stop a Cold in Its Tracks.)

"Public health and government agencies clearly would benefit from having objective forecasts that provide them with a larger lead time for preparedness," says Jeffrey Shaman, Ph.D., an assistant professor at Columbia. "It would give them a leg up on vaccinations, antivirals, and possibly even school closures. On the individual level, a flu forecast could help motivate people to take the steps they need to avoid the flu, such as getting vaccinated 2 weeks or more in advance."

Start with a model

Complex systems like the weather and flu transmission change through time in ways scientists call "nonlinear" or even "chaotic." In order to predict the course they're likely to take, researchers begin by constructing a mathematical model that attempts to factor in known influences on the system's behavior. Westerly winds over the Atlantic Ocean, for example, tend to cause hurricanes to hook right and head to the North. Similarly, reductions in absolute humidity each winter enhance flu transmissibility, in part, by allowing virus particles to survive longer.

But even the best mathematical models can't hope to account for the myriad complexities of the real world. "It's like the butterfly effect, the whole chaos thing," explains Shaman. "Little tiny perturbations will cause an imperfect model to fly off and predict that something will happen when, in fact, something radically different actually happens." (Another cure for the common cold? Eat! Here are 5 Fruits that Fight the Flu.)

Add real-time observations

In order to minimize a model's tendency to drift off course, forecasters add a second key ingredient: real-time observations about how the weather system and/or epidemic are progressing in the here and now.

As far as hurricanes go, comprehensive real-time data didn't become available till the late 1970s, when there were finally enough weather satellites to provide detailed pictures of a storm in progress.

With the flu, the Centers for Disease Control and Prevention has for several decades tracked disease statistics based on case reports from doctors and virology labs around the country. Such data, though helpful, suffers from two notable limitations. For one, it's not well-targeted geographically. Moreover, there is a 1- to 2-week lag time between when people get sick and the CDC publishes its cases in its weekly FluView report. (Steer clear of colds and flu by avoiding the 6 Germiest Places You Touch Every Day.)

In 2009, however, a new tool emerged that is fast proving as important for flu tracking as satellites are for storm monitoring. This tool comes courtesy of a company most of us use every day: Google, developer of the world's most popular Internet search engine.

Researchers at Google Inc., working collaboratively with scientists at the CDC, identified 45 different search terms--from "influenza complications" to "antiviral medications"--that they showed strongly correlate with the actual spread of influenza. Not all those searching such topics actually have the flu, the researchers acknowledged. Still, as they wrote in the journal Nature, by harnessing "the collective intelligence of millions of users," the combined search engine results provide accurate, highly detailed data down to the municipal level, and on the same day it's happening.

Statistical filters: the final ingredient

Over the past several decades, researchers have developed a third key to accurate forecasting, one that tends to receive short shrift in the popular press largely because it's so difficult to understand. This third ingredient goes by a variety of names, says Shaman, including data assimilation methods, sequential Monte Carlo techniques, or in his case, "the ensemble adjustment Kalman filter." "What these statistical techniques do," he explains, "is take the real-world observations and use them to inform the mathematical model, adjusting and nudging it so that it better represents actual current conditions." (Steal the teacher's secrets: Click here to Solve Any Life Problem Using Math.)

Not only does such adjusting and nudging constrain a model's tendency to otherwise drift off course, it also has a secondary benefit: constant auto-correction "trains" the model so that it's more likely to fly on a truer trajectory in the future. It's akin to lengthening the barrel of a rifle so that when the bullet finally exits, it's more likely to find the right target.

After refining their new approach, Shaman and his colleague Alicia Karspeck went back to see how well it could have forecast the flu outbreaks in New York City in the years 2003 to 2008. As they conclude in their PNAS paper, "Real-time skillful predictions can be made at times more than 7 weeks in advance of the actual peak." Such a lead time, if it holds up in future research, represents a potential godsend to anybody hoping to prepare for the flu and curb its miserable impact.

"Flu forecasting," says Shaman, "does have a lot of promise. But the bottom line is there is still a lot of work to be done." Might he foresee a day when FluTracker 9000 will join the nightly forecast, adding "flu fronts" to the standard warm and cold varieties on the weather map?

Not immediately, but check back in a few years. "It's not unreasonable," says Shaman, "that that will happen."

Learn what you have--and the quickest way to fix it--with our Ultimate Cold and Flu Symptom Solver.