Emergency Preparedness

Carnegie Mellon Points Flu Forecasting at COVID-19; Outcomes Unknown

• By Dian Schaffhauser
• 03/12/20

Each year researchers at Carnegie Mellon University run a flu forecasting process to provide the Centers for Disease Control and Prevention with data that can help scientists understand timing, peak and intensity of that year's flu season, which typically runs from October through May. This year, however, at the request of the CDC, the researchers will stay on the job into summer since scientists don't yet know whether COVID-19 will follow the same cycle of subsiding as the weather warms up.

Normally, each week the CDC issues an "influenza surveillance report" to provide key indicators, such as hospitalizations and deaths. The primary source is reporting done by doctors' offices and clinics. The flu forecasts, on the other hand, use statistics and the art of estimation to predict flu behavior over coming weeks and months.

Carnegie Mellon runs one of two Influenza Forecasting Centers of Excellence, designated last year by the CDC. The center is led by Roni Rosenfeld, head of Delphi (developing the theory and practice of epidemiological forecasting) and head of the university's Machine Learning Department.

Carnegie Mellon's Delphi research group. Source: Carnegie Mellon University

The team will use the same two methods to forecast the coronavirus illness as they do for the flu. The first is machine learning, which makes predictions based on past patterns; the second uses crowdsourcing to base its predictions on the aggregate judgments of human volunteers who submit weekly estimates.

Although both approaches have proven accurate in the past, Rosenfield emphasized in an article about the project that little is known about COVID-19 and the forecasting may be wrong.

"The wisdom-of-the-crowds method might be useful for this novel coronavirus, given that it's based on the collective wisdom of people, who are good at adapting to previously unseen circumstances," he said. But because the machine learning method requires having large amounts of data from previous seasons, that one may be way off.

"It will be difficult to know whether your forecasting method is reliable," he explained. "Because after this season is over, we will know whether the forecast was accurate or not, but we will not know if we were lucky or unlucky. You could be accurate because of luck, you could be inaccurate because of bad luck. You cannot draw many conclusions from a single season."

Among the participants in the flu forecasting center of excellence are Carnegie Mellon's Department of Statistics and the Department of Engineering and Public Policy, as well as the University of Pittsburgh Graduate School of Public Health and the Harvard School of Public Health.