Optimising Renewal Models for Real-Time Epidemic Prediction and Estimation

KV Parag; CA Donnelly

Optimal Window Selection 180 We apply our APE metric to select k * for various epidemic examples, which 181 examine reproduction number profiles for large outbreaks (Fig. 2a-Fig. 2c ), 182 small ones (which are more difficult to estimate: Fig. 2d-Fig. 2e) , and a random 183 one (Fig. 2f) . We simulated epidemics for t = 200 days using a generation time 184 distribution analogous to that used for Ebola virus disease predictions in [16] . 185 We investigated a window search space of 2 ≤ k ≤ t /2 and computed the APE at 186 each k, over the epidemic duration (1 ≤ s ≤ t). Reproduction number estimates, 187R τ (s) , and incidence predictions, x | I s s−k+1 , are presented on the left and right 188 panels in Fig. 2 . Predictions over the first s < k times use all s data points. 189 We find that the APE metric balancesR τ (s) estimate accuracy against one- When k is unjustifiably large not only do we observe systematic prediction 201 and estimation errors, but alarmingly, we tend to be overconfident in them.