Anticipating epidemic transitions in metapopulations with multivariate spectral similarity

Abstract

Prediction and control of emerging pathogens is a fundamental challenge for public health. To meet this challenge, new analytic tools are needed to characterize the underlying dynamics of the geographical spread of pathogens, identify predictable changes in their dynamics, and support strategic planning for disease elimination and control. Nonparametric and model-independent tools are particularly needed. Here, we propose a multivariate method that uses similarity in cross-spectral density between measured spatial time series of disease prevalence as a feature measuring the proximity of a tipping point, i.e., emergence or elimination. In particular, we show that the increase in the average value of spectral similarity in measured epidemiological time series contains crucial information about the underlying dynamics and proximity to critical points in infectious disease systems. Theoretical analysis of a standard metapopulation SIR model and empirical analysis of case reports of pertussis in the continental USA demonstrate that this increase is observed when the disease approaches elimination. Therefore, this nonparametric indicator provides insight into the fundamental underlying state of the epidemiological system, which is key in developing appropriate strategies to more quickly achieve elimination goals.