FLU OUT­BREAKS?

Popular Science - - THE BIGS - BY JES­SICA WAP­NER

com­plex al­go­rithms. These for­mu­las re­veal pat­terns that in­ves­ti­ga­tors can then com­pare with what­ever the CDC or other health agen­cies re­port about the sick­ness. If a com­puter-gen­er­ated pre­dic­tion matches re­al­ity, we know the ex­perts are onto some­thing.

Search queries aren’t the only vein of data that re­searchers mine for flu clues. Svit­lana Volkova, a data sci­en­tist at the Pa­cific North­west Na­tional Lab­o­ra­tory, looks for gems of in­for­ma­tion on Twit­ter. She re­cently ver­i­fied a new deep-learn­ing method that probes tweets for signs of the flu. In an anal­y­sis of more than 170 mil­lion tweets posted over three years, Volkova and her col­leagues found their model could ac­cu­rately pro­duce three-day forecasts of flu-like ill­nesses at a lo­cal level. That’s much quicker than waiting for flu reports from the CDC, which lag up to two weeks be­hind what’s hap­pen­ing in the world. (Face­book says it’s not in the flu-pre­dict­ing busi­ness, so for now, your sick emoji doesn’t serve a greater good.)

So­cial me­dia adds more data for re­searchers to work with, but it still has lim­i­ta­tions. An­noy­ingly, the im­age we present on­line doesn’t al­ways match the mu­cus-plagued per­son we are at home. Michael Paul, an in­for­ma­tion sci­en­tist at the Uni­ver­sity of Colorado at Boul­der, re­cently found that peo­ple rarely tweet about their flu-like symp­toms. In fact, the re­searchers found that peo­ple tweet less when they’re ill. So the next time your fa­vorite Twit­ter per­son­al­ity seems oddly quiet, it could be be­cause they’re sick of Twit­ter—but it might just be that they’re sick. Paul also in­ves­ti­gated In­sta­gram and found that acute ill­ness is the least-com­mon health topic for photo post­ing. Not sur­pris­ingly, flu-rid­den peo­ple don’t love tak­ing self­ies.

Dis­ease de­tec­tives, in­clud­ing Si­mon­sen, hope that elec­tronic health records could aug­ment data from our tweets and posts. In­sur­ance-claim forms, which list ail­ments and how they were treated, are par­tic­u­larly cru­cial. But peo­ple are typ­i­cally re­luc­tant to share pri­vate health data with re­searchers.

Epi­demi­ol­o­gists would like to calm those pri­vacy worries. They want only the num­bers, never the names. But the fi­nal call ul­ti­mately lies with in­di­vid­u­als. The pub­lic, Si­mon­sen says, must weigh the bal­ances: “Pri­vacy on one side and the need to know more on the other.” That de­lib­er­a­tion is even more perti­nent since the EU im­ple­mented the Gen­eral Data Pro­tec­tion Reg­u­la­tion this year—giv­ing peo­ple more say in how their in­for­ma­tion is used.

Adding in­for­ma­tion from an app used to log health sta­tus—just as we do with fit­ness track­ers or diet pro­grams—could make big data-based flu forecasts even more ac­cu­rate, Si­mon­sen says. And pri­vate com­pa­nies might come around: UNICEF is work­ing with sev­eral, in­clud­ing IBM, to gather data in or­der to im­prove re­sponses to global ill­nesses.

Ul­ti­mately, the po­ten­tial for big data to pre­dict the next flu pan­demic might de­pend on peo­ple around the globe all over­shar­ing our ill­nesses. The more we tweet about our #flu symp­toms, the more data we gen­er­ate. The more we al­low com­pa­nies to share that data with re­searchers, the more ac­cu­rate they can make their pre­dic­tions. And all that shar­ing, Volkova says, “will help the world.”

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.