Columbia Model Forecasts Flu Season Trends In Real Time
A forecasting system developed by Columbia University scientists successfully predicted the onset of last year’s seasonal flu across the US in real-time, according to a paper published this week in Nature Communications.
The system used techniques similar to those used in weather forecasting to predict when the flu would peak – in some cases, up to nine weeks in advance. Four weeks into the flu season, the system was able to predict – with more than 60 percent accuracy – when the flu would peak in 108 cities across the US.
Each year, the seasonal flu affects between 5 and 20 percent of the population in the US, and more than 200,000 people are hospitalized from seasonal-flu related complications, according to the Center for Diseases Control (CDC).
Although the flu season generally occurs in winter, the duration and onset times vary each year. Last year’s flu season peaked in the southeast in December and the rest of the country by early January 2013.
Knowing when exactly to expect the flu would help health officials take necessary preventative measures, stock up on vaccinations ahead of time and better inform the public.
The system designed by the researchers crunched together numbers from local CDC-verified flu cases as well as data from Google Flu Trends, which estimates numbers based on flu-related search queries.
Predictions varied in accuracy in different cities; overall, it was easier to predict flu onset in smaller cities, the researchers reported.
“Population density may also be important. It suggests that in a city like New York, we may need to predict at a finer granularity, perhaps at the borough level. In a big sprawling city like Los Angeles, we may need to predict influenza at the level of individual neighborhoods,” said Dr. Jeffrey Shaman, Columbia University environmental health scientist and first author, in a statement. The forecasts made by the system were also much more reliable than existing techniques, he added.
In a previous study, the researchers tested the model to predict seasonal peaks in New York City between 2003 and 2008. Encouraged by the results, they refined and expanded the forecasting to all major cities in the US last year. Although there were some setbacks – Google Flu Trends overestimated the number of flu occurrences last year, for instance – the researchers are confident in the model’s accuracy in predicting upcoming flu seasons. Forecasts for the current flu season will be available on the Mailman School of Public Health website soon, they reported.