Table 1 A comparison of the different ecological forecasts.

From: Consensus and conflict among ecological forecasts of Zika virus outbreaks in the United States

  Carlson Messina Samy Mordecai (97.5% min) Mordecai (2.5% max)
n points 242 323 168 NA NA
n predictors 15 6 15 NA NA
AUC 0.970 0.829 NA NA NA
Counties Predicted 13 465 1616 1937 3108
Accuracy 99.6% 85.2% 48.2% 37.8% 0.2%
County Population at Risk 19,653,445 95,359,408 270,249,781 218,444,263 320,957,062
Mean Outbreak Size 12,871,005 63,622,367 181,290,371 37,598,099 198,910,979
Median Outbreak Size 14,552,250 64,038,273 181,732,629 37,312,233 197,731,918
  1. Four different methods, each performing well based on sufficient data and predictors, produce highly contrasting results. Out of a total of 3108 counties in the continental U.S., only five have experienced outbreaks (Cameron County, TX with 6 cases of local transmission in 2016; Miami-Dade, FL with 241; Palm Beach, FL with 8; Broward County, FL with 5; and Pinellas County, FL with 1)39,69. Accuracy values were calculated from the confusion matrix of observed outbreaks against predicted suitability. The Carlson model comes closest to predicting the geography of those outbreaks most accurately; but all epidemiological models “overpredict” the number of suitable counties based on the current extent of outbreaks. (Mordecai results are split for the highest bound with minimum temperatures, and the lowest bound for maximum temperatures, to give the full range of predictions. Self reported AUC values are shown not as a comparative measure of accuracy, but simply as the self-reported accuracy of the studies. Samy et al. used the Partial ROC in place of the AUC but did not report values. NA = Not Applicable; NR = Not Reported).