This article is more than 5 years old. Information may no longer be current.
New methods for real-time influenza forecasting appeared effective
Click Here to Manage Email Alerts
An influenza forecasting system that combined real-time estimates from Google Flu Trends and data from CDC surveillance programs reliably predicted the timing of the 2012-2013 influenza season up to 9 weeks in advance of its peak.
The new influenza forecasting method is more reliable than traditional approaches that rely on historical data, according to Jeffrey Shaman, PhD, of Columbia University’s Mailman School of Public Health in New York City, and colleagues.
Beginning in late November 2012, the researchers used influenza-like illness observations and a data assimilation technique called the ensemble adjustment Kalman filter (EAKF) to train susceptible-infected-recovered-susceptible models. Using this formula, they were able to generate weekly forecasts of influenza activity in 108 US cities.
Four weeks into the 2012-2013 influenza season, the forecasting system was able to accurately predict influenza activity for 63% of the cities, producing nearly twice as many accurate predictions as the best traditional methods that relied on resampling historical outcomes, the researchers reported in Nature Communications.
“Our method greatly outperformed these alternate schemes,” Shaman said in a press release, adding that overall accuracy continued to improve as the season progressed.
This type of systems approach is analogous to weather forecasting, in that it employs real-time observational data to reduce model forecasts error.
Despite its success, regional differences in the accuracy of the forecasting system were observed. Overall, longer lead times for smaller populations, higher population densities or smaller geographies tended to be more accurate. Optimal spatial scales for this type of influenza forecasting still need to be determined, the researchers acknowledged.
“In a city like New York, we may need to predict at a finer granularity, perhaps at the borough level,” Shaman noted. “In a big sprawling city like Los Angeles, we may need to predict influenza at the level of individual neighborhoods.”
The influenza forecasting system will be put back into action as soon as the 2013-2014 influenza season begins, according to Shaman. Data will be more readily available to the public on a Mailman School of Public Health website expected to launch in the coming weeks.
“Going forward, we must work with public health officials to increase their familiarity with the capabilities and limitations of these forecasts, as well as our own familiarity with the public health intervention and response decision-making process,” the researchers wrote.
In the future, the system could be used to forecast how an influenza pandemic spreads through populations once a local outbreak begins.
Different lead forecasts could have different uses, with short lead predictions aiding public awareness and attention to personal hygiene and longer lead predictions aiding vaccination efforts or school closure decisions, the researchers explained.
Disclosure: Funding was provided by an NIH grant, the NIH Models of Infectious Disease Agent Study program and the Department of Homeland Security Science and Technology Directorate’s RAPIDD program.
Click Here to Manage Email Alerts