This past week, an update released by the Centers for Disease Control and Prevention reported that influenza-like illness activity has reached 7.7 percent. To put this in perspective, the 2009 H1N1 pandemic peaked at exactly 7.7 percent.
Spencer Fox, a Ph.D. student at UT-Austin, was an undergraduate at the time of the H1N1 pandemic, and that was what inspired him to pursue influenza as a topic of research.
Fox’s research aims to predict the behavior of seasonal flu patterns, focusing on when the flu season will peak, when it will start decreasing and what the magnitude of the season will be.
“It all started with a flu prediction challenge the CDC put on in 2013,” Fox said. “I led our team in that competition, and since then, things have evolved so that there are a bunch of groups working on this.”
Fox said that other groups from that competition, including the one behind FluSight Network, are still continuing to work closely with the CDC. Although Fox isn’t currently working directly with the CDC, his data and research are still important resources for public health officials.
“One way to think about (our) forecasts and predictions is to see them as another line of evidence that public health officials can use to make decisions,” Fox said.
Fox draws information from slightly unorthodox sources, such as Twitter, to develop statistical models from which he can make these predictions. The information that he is able to draw from this Twitter data comes from hashtags and tweets that might indicate someone has the flu, or that their sibling or relative has the flu, Fox said.
Although this data might not be the most accurate, since not everyone who has the flu tweets about it and because Twitter trends might have been influenced by media coverage regarding the flu, the fact that this data indicates what is happening in real time is still incredibly useful, Fox said.
“In the traditional epidemiological surveillance system the CDC has in place, physicians are submitting data to the CDC and the CDC compiles all the data and releases it in a report,” Fox said. “That compilation of data and the waiting for the physicians to send in that data takes a couple of weeks. So, having real-time digital data can really improve predictions for seasonal flu, because you’re seeing it in real-time without that two-week lag.”
In addition to Twitter, Fox’s group also uses Wikipedia data because it is publicly available, and data for an article is easily accessible. Other groups have shown that Wikipedia data can be just as useful as Twitter data, Fox added.
Fox once said in an interview with KXAN that there’s almost a paradox with epidemiological predictions — that when they make predictions, those predictions never come to fruition if they’ve done their job right.
“That’s the large issue with all public health prediction goals,” Fox said. “If I predict that there will be a large flu epidemic in Austin, and then work with public health officials to enact intervention — such as increasing advertising for vaccines, stocking up on antivirals or making ad campaigns about washing your hands — then you probably won’t have a flu epidemic in Austin.”
The quote is nice in principle, but in practice, the group is not there yet, Fox said.
“That’s a great goal to have: to come up with these forecasts and predictions and work closely with public health officials to prove the forecast wrong,” Fox said. “I don’t think our predictions are consistently performing well enough to have the full weight of public health officials behind them right now, but hopefully in the future things will become much more accurate.”
