Table 2.

Summary of study characteristics

Author	Publication year	Date type	Data scale	Data range	Location	Method	Predicted measure	Measure of accuracy
Longini et al.15	1986	ILI	Weekly	1968–1969	52 cities	Mathematical model defined on a continuous state space in discrete time	ILI across 425 days and peak period	Deviation from ILI estimated based on WHO reports
Aguirre & Gonzalez9	1992	ILI	Daily	1988	Havana, Cuba	Mathematical model defined on a continuous state space in discrete time	Daily ILI, peak, and duration	Correlation and statistical tests
Viboud et al.10	2003	ILI	Weekly	1984–2002	France & Administrative Districts	Method of analogs	Weekly ILI	Correlation and RMSE
Hall et al.3	2006	ILI and deaths attributable to influenza	Weekly	1968–1970, 1918–1919 & 1957–1958	United Kingdom	Deterministic mass action model	Timing and amplitude of peak, duration, and magnitude	Error and time difference
Polgreen et al.13	2007	Influenza activity	Weekly	2004–2005	Iowa, USA	Prediction markets	Weekly activity based on CDC's color coded system	Proportion predicting correct color code
Andersson et al.20	2008	LCI cases	Weekly	1999–2006	Sweden	Regression model and prediction rules	Peak timing and height	Error and time difference
Jiang et al.11	2009	ILI and deaths attributable to influenza	Daily	2006	USA	Bayesian network	Epidemic curve	Correlation and error
Towers & Feng16	2009	Influenza case count data	Weekly	2009	USA	SIR model	Peak time and attack rate	Confidence intervals
Soebiyanto et al.12	2010	LCI cases	Weekly	2005–2008	Hong Kong & Maricopa county, AZ, USA	ARIMA model	Weekly case counts	RMSE
Ong et al.4	2010	ILI	Weekly	2009	Singapore	SEIR model with particle filtering	Weekly case counts, peak timing, and duration	Error
Chao et al.2	2010	CDC influenza case estimates and estimates of vaccine availability and distribution	None	2009–2010	USA & LA County, USA	Epidemic simulation model based on a synthetic population	Peak timing and magnitude	Predicted range
Nishiura14	2011	Influenza cases	Weekly	2009–2010	Japan	Discrete time stochastic model	Weekly case counts	Prediction intervals
Shaman & Karspeck18	2012	Google Flu Trends	Weekly	2003–2008	New York City, USA	SIRS model with ensemble adjustment Kalman filter	Peak timing	Posterior estimates and deviation
Tizzoni et al.5	2012	ILI, ARI incidence, LCI	Weekly	2009	48 countries	Metapopulation stochastic epidemic model	Peak timing and attack rate	Confidence intervals and time difference
Hyder et al.21	2013	LCI	Weekly	1998–2006	Montreal, QC, Canada	Individual-based model	Peak timing, peak intensity, and epidemic duration	Error and time difference
Nsoesie et al.19	2013	Google Flu Trends	Weekly	2004–2005, 2007–2008 & 2012–2013	Seattle, WA, USA	Individual-based model	Peak timing	Confidence intervals and deviation

LCI, laboratory confirmed influenza; ILI, influenza-like illness; ARI, acute respiratory infection; ARIMA, autoregressive integrated moving average; RMSE, root-mean-squared-error.