How Twitter Tracks the Flu

New re­search sug­gests Twit­ter — the so­cial-me­dia site best known for spot news and cat GIFs — can also be used to pin­point flu out­breaks.

A team of re­search­ers at Johns Hop­kins Uni­versity and George Wash­ing­ton Uni­versity found that tweets are an ac­cur­ate pre­dict­or of where cases of the flu are highly con­cen­trated.

Be­cause Twit­ter pro­duces in­form­a­tion in real time, loc­al pub­lic health of­fi­cials can use the ag­greg­ate tweet data to act early to make re­sources avail­able to com­bat the spread of the flu. Flu re­ports from the Cen­ters for Dis­ease Con­trol and Pre­ven­tion — which are cur­rently used to make pub­lic health de­cisions — have about a two-week delay from the time the in­form­a­tion is col­lec­ted to the time it is re­leased, said lead au­thor Dav­id Bro­ni­atowski.

“I don’t see this as re­pla­cing what [the CDC] does,” Bro­ni­atowski said, “but it might be used as an early warn­ing sys­tem, in par­al­lel with the hos­pit­al data as it’s col­lec­ted.”

Armed with the early in­form­a­tion, pub­lic health of­fi­cials can in­crease vac­cin­a­tions and ready hos­pit­als for an in­flux of pa­tients.

Tweets are tied to a loc­a­tion, and the re­search­ers were able to ac­cur­ately gauge the spread of the flu at both a na­tion­al and loc­al level by check­ing their data against re­ports from the CDC and the New York City De­part­ment of Health and Men­tal Hy­giene.

“This has ma­jor pub­lic health im­plic­a­tions, be­cause if you are try­ing to fig­ure out what the like­li­hood is that you need to pre­pare for a flu out­break and you’re look­ing at na­tion­al data, it doesn’t help you if they’re con­cen­trated in Ore­gon and you’re in At­lanta,” Bro­ni­atowski said.

In ad­di­tion to pre­par­a­tion, pub­lic health of­fi­cials can use the Twit­ter data to un­der­stand the so­cial as­pect of how the flu spreads, Bro­ni­atowski said, be­cause they are able to sep­ar­ate tweets that re­veal aware­ness of the flu in the com­munity from the tweets that in­dic­ate the user is in­fec­ted with the flu.

Fu­ture re­search could ex­plore Twit­ter’s ef­fect­ive­ness in track­ing oth­er dis­ease out­breaks. One of the lim­it­a­tions of Twit­ter is its demo­graph­ics, however, which tend to un­der-rep­res­ent chil­dren and older pop­u­la­tions. In rep­lic­at­ing the res­ults with com­munit­ies smal­ler than New York City, a smal­ler num­ber of Twit­ter users could also de­crease the re­li­ab­il­ity of the tweets in track­ing the flu trends, the re­search­ers said.

The team eval­u­ated Twit­ter’s ac­cur­acy dur­ing the last full flu sea­son, Sept. 30, 2012, through May 31, 2013. One of the re­search­ers on the team was sup­por­ted in part by an award from the Na­tion­al In­sti­tutes of Health and an­oth­er re­ceived a grant from the Na­tion­al Sci­ence Found­a­tion. They con­duc­ted their study in­de­pend­ent of any out­side or­gan­iz­a­tions, and the find­ings were pub­lished in the peer-re­viewed journ­al PLOS ONE.