In recent years progress has been made towards the creation of early-warning signals. Early-warning signals are analytical techniques intended to provide some forewarning about future transitions in the dynamics of a system. Taking insight from the theory of dynamical systems, various indicators which are observable in time-series data have been developed and tested.
The key change in the epidemiological behavior of an emerging disease is found at the invasion threshold. Above this threshold the disease prevalence becomes sustained by transmission within the population, rather than due to spillover from external sources. A goal when analyzing emerging pathogens is to have access to model free methods by which this transition can be anticipated.
On a more technical level, the study of early-warning signals is focused around the concept of critical slowing down, in which the persistence of small perturbations to the system increases as a dynamical transition is approached.
Early-warning signals have had success in anticipating transitions in other ecological systems, in particular various experimental models of ecosystem collapse. Their potential for emerging diseases has been largely unexplored.
Results/Conclusions
We develop a set of candidate early-warning signals, some from the theory of stochastic processes (e.g. the autocorrelation) and others from information theory (e.g. the Kolmogorov complexity). We demonstrate how these summary statistic measures successfully anticipate an emerging diseases' approach to the invasion threshold under theoretically ideal conditions. The early-warning signals are fully ready to be applied to empirical data from real-world emerging diseases.
The candidate early-warning signals are tested against simulated data generated using models which possess a range of possible confounding factors found in real-world disease systems, such a finite population size and immune individuals. We find the signals are successful and independent of the model details provided the pool of susceptible individuals is not significantly depleted prior to the transition.
The speed at which the transition is approached has an impact on the reliability of the early-warning signals, with faster approaches reducing reliability. Certain candidate signals still perform well even when emergence occurs over an epidemiologically reasonable period of 20 years.
Preliminary results using weekly incidence data for notifiable diseases in the USA is supportive.