Using Digital Surveillance Tools for Near Real-Time Mapping of the Risk of International Infectious Disease Spread: Ebola as a Case Study

Sangeeta Bhatia; Britta Lassmann; Emily Cohn; Malwina Carrion; Moritz U.G. Kraemer; Mark Herringer; John Brownstein; Larry Madoff; Anne Cori; Pierre Nouvellet

is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19011940 doi: medRxiv preprint Predicted weekly presence of cases for each country in weeks following a week with no observed cases. The left panel shows the True and False alert rates using different thresholds for classification for alerts raised 1 (violet), 2 (light violet), 3 (dark pink) and 4 (light pink) weeks ahead. The black curve depicts the overall True and False alert rates. On each curve, the dot shows the True and False Alert rates at 93% threshold. For a given threshold (x th percentile of the forecast interval), we defined a True alert for a week where the x th percentile of the forecast interval and the observed incidence for a country were both greater than 0; false alert for a week where the threshold for a country was greater than 0 but the observed incidence for that country was 0; and missed alert for a week where the threshold for a country was 0 but the observed incidence for that country was greater than 0. True alert rate is the ratio of correctly classified true alerts to the total number of true and missed alerts (i.e., (true alerts)/(true alerts + missed alerts)). False alert rate is similarly the ratio of false alerts to the total number of false alerts and weeks of no alert (where the observed and the threshold incidence are both 0). The right panel shows the True (green), False (orange) and Missed (red) 1 week ahead alerts using the 93 rd percentile of the forecast interval as threshold. The figure only shows countries on the African continent for which either the 93 rd percentile of the predicted incidence or the observed incidence was greater than 0 at least once. The first alert in each country is shown using larger symbols (square or triangle). Alerts in a country in a week where there were no observed cases in the previous week are shown using hollow triangles. In each case, weeks for which all observed points were imputed are shown in lighter shades. Country codes, shown on the y-axis, are as follows: BFA -Burkina Faso, CIV -Côte d'Ivoire, GHA -Ghana, GIN -Guinea, LBR -Liberia, MLI -Mali, NGA -Nigeria, SEN -Senegal, SLE -Sierra Leone. The alerts are based on forecasts using the ProMED data, a 2-week calibration window and a 4 week forecast horizon.