Prospects for detecting early warning signals in discrete event sequence data: application to epidemiological incidence data

Emma Southall; Michael J. Tildesley; Louise Dyson

Simulation-based studies exploring incidence-type data have suggested that the 65 potential for emergence of an infectious disease can be informed by statistical 66 signatures [3, 4] . These studies represent the first attempts to understand the robustness 67 of some indicators when used with disease emergence incidence data, subject to 68 underreporting and time aggregation. Both studies find that EWS do precede disease 69 emergence even when reporting is low. When the numerical performance of 10 EWS are 70 compared, Brett et al. find that the mean and variance perform well unless incidence is 71 subject to a highly overdispersed reporting error and they compare these results with 72 previously studied prevalence results. Theoretical predictions are given for prevalence 73 data, however the analytical behaviour of incidence is not explored.