Flu 2017: When The Season Will Peak and When It Will Come to an End

Updated | Predicting stuff—elections, baseball seasons, the weather—is useful. But predicting infectious diseases like the flu could actually save lives. Scientists across the country are working on just that. But if you thought accurately predicting the results of the 2016 presidential election was hard, you should talk to a flu researcher.

"Flu is extremely unpredictable," said Roni Rosenfeld, a flu forecaster and researcher at Carnegie Mellon University. "At the Centers for Disease Control and Prevention, they say, 'If you've seen one flu season, you've seen one flu season.'" Each year can be remarkably different.

Last year, Carnegie Mellon's models, one of which used machine learning, were the two most accurate of the flu forecasts available, according to the CDC. But Rosenfeld is leading just one group—more than a dozen others are participating in the CDC's flu forecasting challenge, too.

Forecasting the flu could be important for one major reason: so people can prepared. A hospital administrator might have extra staff on call and stock more antiviral medication than usual, said Michael Jackson, a researcher at Kaiser Permanente Washington Health Research Institute who is working on a flu forecasting model. "In a really severe flu season, the number of patients coming in with influenza can really start to strain hospital resources," he said, citing published reports.

Forecasting could also help people time their flu shots properly, Rosenfeld said. People who are older tend to have weaker immune systems, so getting a flu vaccine too early or too late might mean a person is unprotected for the very early or late days of a flu season.

Rosenfeld's model indicates at the moment that flu may peak earlier than usual in most parts of the U.S., he said—and it will probably be higher than usual, too. That may mean that flu season might wind down sooner than usual, too, but that's not a sure thing.

However, flu modeling is still very rudimentary. "To be honest, I think we're still very far from where we want to be," he said. "One of the problems, this is a personal belief, is we won't be able to make really good forecast until we're able to do them at very small geographical areas, like cities."

Of course, all these forecasts rely on data—so making better and finer-grained models will necessarily require better information. The CDC tracks reports of flu-like illnesses and, in some areas, positive flu tests and the strains that are detected. However, that data isn't reported on a city-wide level. Additionally, some people don't go to the doctor if they have the flu; not everyone who does go to their doctor gets a flu test.

But in an ideal world, once flu forecasts have matured, they could help people plan their lives in the same way a weather forecast does. Especially for people who are very old, very young or particularly weak, flu can be deadly. An accurate forecast would allow vulnerable people to avoid the virus in the same way they might avoid a blizzard.

"All of us look to see what the weather is going to be like," said Rosenfeld. "For older people, the flu coming is a major snowstorm—an invisible one. And in some senses, it's more deadly."

The practical benefits are obvious. "If you have a 90-year-old grandmother, and she wants to visit her sister in Cincinnati but you know that flu is going to spike there next week or two weeks from now, you may talk her into delaying her trip."

When asked to forecast the future of forecasting, Rosenfeld hesitated. "I'd be a fool to try and do that," he said. But he did note that the development of weather forecasting began after the Civil War and took decades to become useful. Important technology and networks had to be developed to get to even the (still imperfect) 10-day forecasts that we now expect.

It may be some time before you can check the flu forecast on your phone with the same ease you check the weather. Still, the last few years have brought some momentum. The number of teams participating in the CDC's forecasting challenge has grown in the last few years, from about 7 to more than 20, Rosenfeld said. "The community is growing."

Correction: A previous version of this article referred to Roni Rosenfeld as Roni Rosenberg.