TY - JOUR
T1 - Finding the signal in the noise
T2 - Could social media be utilized for early hospital notification of multiple casualty events?
AU - Callcut, Rachael A.
AU - Moore, Sara
AU - Wakam, Glenn
AU - Hubbard, Alan E.
AU - Cohen, Mitchell J.
PY - 2017/10
Y1 - 2017/10
N2 - Introduction: Delayed notification and lack of early information hinder timely hospital based activations in large scale multiple casualty events. We hypothesized that Twitter real-time data would produce a unique and reproducible signal within minutes of multiple casualty events and we investigated the timing of the signal compared with other hospital disaster notification mechanisms. Methods: Using disaster specific search terms, all relevant tweets from the event to 7 days post-event were analyzed for 5 recent US based multiple casualty events (Boston Bombing [BB], SF Plane Crash [SF], Napa Earthquake [NE], Sandy Hook [SH], and Marysville Shooting [MV]). Quantitative and qualitative analysis of tweet utilization were compared across events. Results: Over 3.8 million tweets were analyzed (SH 1.8 m, BB 1.1m, SF 430k, MV 250k, NE 205k). Peak tweets per min ranged from 209–3326. The mean followers per tweeter ranged from 3382–9992 across events. Retweets were tweeted a mean of 82–564 times per event. Tweets occurred very rapidly for all events (<2 mins) and represented 1% of the total event specific tweets in a median of 13 minutes of the first 911 calls. A 200 tweets/min threshold was reached fastest with NE (2 min), BB (7 min), and SF (18 mins). If this threshold was utilized as a signaling mechanism to place local hospitals on standby for possible large scale events, in all case studies, this signal would have preceded patient arrival. Importantly, this threshold for signaling would also have preceded traditional disaster notification mechanisms in SF, NE, and simultaneous with BB and MV. Conclusions: Social media data has demonstrated that this mechanism is a powerful, predictable, and potentially important resource for optimizing disaster response. Further investigated is warranted to assess the utility of prospective signally thresholds for hospital based activation.
AB - Introduction: Delayed notification and lack of early information hinder timely hospital based activations in large scale multiple casualty events. We hypothesized that Twitter real-time data would produce a unique and reproducible signal within minutes of multiple casualty events and we investigated the timing of the signal compared with other hospital disaster notification mechanisms. Methods: Using disaster specific search terms, all relevant tweets from the event to 7 days post-event were analyzed for 5 recent US based multiple casualty events (Boston Bombing [BB], SF Plane Crash [SF], Napa Earthquake [NE], Sandy Hook [SH], and Marysville Shooting [MV]). Quantitative and qualitative analysis of tweet utilization were compared across events. Results: Over 3.8 million tweets were analyzed (SH 1.8 m, BB 1.1m, SF 430k, MV 250k, NE 205k). Peak tweets per min ranged from 209–3326. The mean followers per tweeter ranged from 3382–9992 across events. Retweets were tweeted a mean of 82–564 times per event. Tweets occurred very rapidly for all events (<2 mins) and represented 1% of the total event specific tweets in a median of 13 minutes of the first 911 calls. A 200 tweets/min threshold was reached fastest with NE (2 min), BB (7 min), and SF (18 mins). If this threshold was utilized as a signaling mechanism to place local hospitals on standby for possible large scale events, in all case studies, this signal would have preceded patient arrival. Importantly, this threshold for signaling would also have preceded traditional disaster notification mechanisms in SF, NE, and simultaneous with BB and MV. Conclusions: Social media data has demonstrated that this mechanism is a powerful, predictable, and potentially important resource for optimizing disaster response. Further investigated is warranted to assess the utility of prospective signally thresholds for hospital based activation.
UR - http://www.scopus.com/inward/record.url?scp=85030661102&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85030661102&partnerID=8YFLogxK
U2 - 10.1371/journal.pone.0186118
DO - 10.1371/journal.pone.0186118
M3 - Article
C2 - 28982201
AN - SCOPUS:85030661102
VL - 12
JO - PLoS One
JF - PLoS One
SN - 1932-6203
IS - 10
M1 - e0186118
ER -