Skip to main content
. 2018 Sep 4;14(9):e1006236. doi: 10.1371/journal.pcbi.1006236

Table 1. Performance of baseline and optimized surveillance systems.

Candidate sources Selected sources RMSE (%ILI) PWE (weeks) PME (%ILI)
Mean [Min, Max] Mean [Min, Max] Mean [Min, Max]
ILINet

  1. ILINet US

 0.66 [0.26,1.10] 2.43 [0, 6] 0.24 [0.008,0.71]
ILINet & WHO

  1. ILINet US

  2. WHO US

 0.63 [0.26,0.98] 2.31 [0, 6] 0.24 [0.009,0.66]
All

  1. ILINet US

  2. WHO US

  3. Athena FluResultVisit IL

  4. Athena FluResultVisit GA

  5. Athena PositiveResult% HHS 08

 0.56 [0.21,0.96] 1.75 [0, 6] 0.19 [0.04, 0.39]
All national (US)

  1. ILINet US

  2. WHO US

  3. Athena ILIVisit US

  4. Athena ILI% US

  5. WordPress Flu

 0.60 [0.26,0.98] 2.12 [0, 6] 0.21 [0.02, 0.40]
All national without ILINet

  1. WHO US

  2. Athena FluResultVisit US

  3. Athena FluRXVisit US

  4. Athena FluVisit US

  5. Athena ILIVisit US

 0.64 [0.18, 1.30] 2.37 [0, 8] 0.21 [0.002,0.45]
All national without ILINet & WHO

  1. Athena ILI% US

  2. Wiki Flu

  3. HM Athena

  4. WordPress Flu

  5. Athena FluResultVisit US

 0.87 [0.29,1.75] 10.81 [0, 19] 0.44 [0.15,0.71]

Data sources were selected based on accuracy (RMSE) of 16 out-of-sample retrospective flu season forecasts (1997-2014, excluding the 2009-2010 H1N1 pandemic), and listed in order of selection. All and National (US) includes 453 and 13 candidate data sources, respectively. Mean, minimum, and maximum values are calculated over the 16 seasons.