Validation for flu prediction

(Medical Xpress)—In 2009, the H1N1 virus slipped into the blood­streams of more than 40 mil­lion people around the world. In just four months, it killed more than 14,000 indi­vid­uals as it trav­eled from Mexico to India on its most favored vehicle: humans. As trav­elers moved about the planet via air­planes and cars, the pathogen fol­lowed, cre­ating an epi­demic the likes of which had not been seen since the 1970s.

At the time, Alessandro Vespig­nani was at the Uni­ver­sity of Indiana, where he began tracking the dis­ease with as much atten­tion as the Cen­ters for Dis­ease Con­trol. Vespignani—now the Stern­berg Family Dis­tin­guished Uni­ver­sity Pro­fessor of physics, com­puter sci­ence, and health sci­ences at North­eastern University—and his research team built a com­pu­ta­tional model called GLEAM, or Global Epi­demic and Mobility Model, which they used to pre­dict the out­breaks as they sur­faced around the globe.

In the last three years, the team has been tire­lessly working to val­i­date its pre­dic­tions. To that end, its recently pub­lished article in the journal BMC Med­i­cine offers defin­i­tive proof of a strong agree­ment between the pre­dic­tions and the real-​​life sur­veil­lance data col­lected in 2009.

"Although we knew the pre­dic­tion of the model were in pretty good agree­ment in sev­eral places of the world," said Vespig­nani, "here we pro­duce a very exten­sive val­i­da­tion on more than 45 countries."

To model dis­ease spreading, GLEAM inte­grates three data "layers." The first uses a pop­u­la­tion data­base, which was devel­oped by a team at Columbia Uni­ver­sity and pro­vides a high-​​resolution pop­u­la­tion den­sity map of the entire planet. The second uses local com­muting flows and air­line trans­porta­tion data­bases to esti­mate human mobility within and between coun­tries, respec­tively. Finally, an epi­demic layer accounts for the behavior of the dis­ease itself, including infor­ma­tion such has incu­ba­tion and trans­mis­sion times.

Oper­ating from within the prover­bial eye of the storm in 2009, the team used the model to fore­cast the week of the epidemic's peak in 48 coun­tries in the Northern Hemi­sphere. In 42 of these coun­tries, the fore­casts were directly on target; in the other five, the team's pre­dic­tions were off by only one to two weeks.

Nor­mally, flu season peaks months after H1N1 did, making even the two-​​week vari­a­tion a con­sid­er­ably good result. "This is the first large-​​scale val­i­da­tion of a com­pu­ta­tional model that pulled out pre­dic­tions in real time," said Vespig­nani. "It shows that com­pu­ta­tional models have acquired the matu­rity to pro­vide useful infor­ma­tion and at the same time points out the way on how to improve and develop better models and tools."