Today, CDC announced that weekly influenza (flu) forecasts for 2016-2017 season are available via CDC’s “FluSight” flu forecasting websiteexternal icon, which has added some updated features this season. The “FluSight” website was launched in January 2016 as a product of a three-year collaboration between CDC’s Influenza Division and external research groups to develop flu forecasting capacity in the United States. CDC also acknowledged the Delphi Research groupexternal icon at Carnegie Mellon University for contributing the three most accurate national-level flu forecasts during the 2015-2016 flu season. Recent flu forecasting data from FluSight suggest that flu activity will likely increase over upcoming weeks, making now a good time to get vaccinated. This week is National Influenza Vaccination Week (NIVW), the time of year when CDC encourages people who haven’t yet gotten their flu vaccine to get vaccinated.
Flu activity forecasting involves predicting in advance when increases in influenza (flu) activity will occur. Unlike CDC’s traditional influenza surveillance systems, which measure influenza activity after it has occurred, flu forecasting offers the possibility to look into the future and plan ahead. This is important because flu places a significant disease burden on the U.S. population each year. The potential benefits of flu forecasting are immense. When experts can accurately predict — similar to a weather forecast — when significant increases in flu activity will occur, the ability to plan ahead and more effectively implement disease mitigation strategies becomes possible. For example, disease forecasting could help determine when best to schedule vaccination clinics or educational campaigns; it could help decide the optimal time to distribute influenza antiviral medications; and it could help doctor’s offices, hospitals, businesses and schools plan for the impact of flu on daily operations.
However, the science of flu forecasting is still in its infancy and has proven to be complicated, as every flu season tends to be different. To address the issue, CDC took a creative approach and enlisted the help of the public. In 2013, CDC launched the “Predict the Influenza Season Challenge“, which was a public contest that encouraged outside researchers from around the world to predict the timing, peak, and intensity of the 2013-2014 flu season using social media data (e.g., Twitter, internet search data, web surveys, etc.) along with data from CDC’s routine flu surveillance systems.
Since the 2013 contest, many of the original contest participants as well as new research teams have continued to advance the science of flu forecasting through their engagement with the Epidemic Prediction Initiative (EPI), which is a joint effort between federal and external researchers.
In 2016, CDC launched FluSight on the Epidemic Prediction Initiative websiteexternal icon, which has become the central hub through which the different research teams participating in the EPI submit weekly flu forecasts. CDC created this central website so that visitors and participants can follow the progress of these predictions and compare them to actual flu activity over the course of the season. This season, 20 different teams are participating in the forecasting initiative and submitting 28 different forecasts to the site.
In the spirit of the original contest, the different research teams participating in the EPI continue to engage in a friendly and collaborative competition via FluSight. Throughout the flu season, each team vies to produce the most accurate flu forecasts each week while working together to identify the best practices for forecasting. Forecasts posted to the site are evaluated against actual flu activity as measured by CDC’s ILINet system, and the participants are able to measure the accuracy and reliability of their forecasts in comparison to one another through their collaboration with CDC.
Last season, the Delphi Research groupexternal icon at Carnegie Mellon University contributed the three most accurate overall national-level flu forecasts to the site. The team, led by Dr. Roni Rosenfeld, used a combination of machine learning and crowd-sourcing to generate its flu forecasts.
The FluSight website is available at https://predict.phiresearchlab.org/external icon. Visitors to the site can view forecasts for flu season onset week, the week of peak activity, peak intensity, and near-term influenza-like-illness (ILI) activity. This season, the visualizations available on the site also incorporate information such as historical averages, so viewers can see how the forecasts align with historical flu data. Throughout the flu season, ILINet data are generally updated every Friday, and forecasts are submitted on Monday evening and are usually published on the website on Tuesdays.
(Disclaimer: Visitors to the FluSight website should note that publication of these models does not signal endorsement by CDC. All of these models will be continually refined and adjusted based on how well the forecasts align with actual flu activity. The various forecasts provided by the website may vary significantly from one another and may be inaccurate.)