Results from the centers for disease control and prevention’s predict the 2013–2014 Influenza Season Challenge

BMC Infectious Diseases

Table 1 Characteristics of nine teams that competed in the Predict the 2013–14 Influenza Season Challenge

Team	Digital Data source	Model type	Regional forecast^a	Brief description^d
A	Wikipedia	mechanistic^b	Yes	Susceptible-Exposed-Infected-Recovered (SEIR) model using data assimilation to probabilistically fit models to ILINet data
B	Twitter	mechanistic	Yes	SEIR model initialized with current Twitter and ILINet data
C	Google Flu Trends; Twitter	statistical^c	Yes	Utilized method of analogues, Kalman filtering, Poisson regression, and an ensemble method averaging the results of the three models to forecast ILINet
D	Google Flu Trends	statistical	Yes	Utilized empirical Bayes model and a spatio-temporal likelihood function
E	Google Flu Trends; Twitter	statistical	Yes	Utilized multiplicative time series model
F	Google Flu Trends	mechanistic	Yes	Susceptible-infected-recovered-susceptible (SIRS) model initialized with Google Flu Trends data and data assimilation methods
G	Twitter	statistical	No	Extrapolation of filtered Twitter data
H	Google Flu Trends; HealthMap; Twitter	mechanistic	Yes	Statistical models used to make short term forecasts and agent based models combined with mean field models with non-linear optimization techniques used to output long term forecasts.
I	Twitter	statistical	Yes	Utilized time series model and method of analogues

^aYes denotes forecast for ≥1 region (for all weeks)
^bIncludes models that incorporate compartmental modeling like Susceptible-Exposed-Infected-Recovered [SEIR] models
^cIncludes models like time series analysis and generalized linear models
^dAdditional information on methodology and results for select teams available in references [34–38]