Identifying COVID-19 peaks using early warning signals

Abstract The SARS-CoV-2 (COVID-19) pandemic has had catastrophic effects on public health and economies. Around the world, many countries employed modelling efforts to help guide pharmaceutical and non-pharmaceutical measures designed to reduce the spread of the virus. Modelling efforts for future pandemics could use the theory of early warning signals (EWS), which aims to predict ‘critical transitions’ in complex dynamical systems. In infectious disease systems, such transitions correspond to (re-)emergence, peaks and troughs in infections which can be indirectly observed through the reported case data. There is increasing evidence that including EWS in modelling can help improve responses to upcoming increases or decreases in case reporting. Here, we present both theoretical and data-driven analyses of the suitability of EWS to predict epidemic transitions in reported case data. We derive analytical statistics for a variety of infectious disease models and show, through stochastic simulations of different modelling scenarios, the applicability of EWS in such contexts. Using the COVID-19 reported case dataset from the United Kingdom, we demonstrate the performance of a range of temporal and spatial statistics to anticipate transitions in the case data. Finally, we also investigate the applicability of using EWS analysis of hospitalisation data to anticipate transitions in the corresponding case data. Together, our findings indicate that EWS analysis could be a vital addition to future modelling analysis for real-world infection data. Author summary The application of early warning signals (EWS), which uses statistics to predict changes in real-world data, has increasingly shown promise in successfully anticipating disease (re-)emergence or elimination of infectious diseases. We extend the existing literature through a theoretical study of EWS indicators under a variety of modelling scenarios that could have occurred during the COVID-19 pandemic (and in future pandemics). We show that EWS calculated on hospitalisation data can also accurately anticipate changes in the cases of related regions whilst also showing that spatio-temporal trends in EWS can be used to identify regions of concern.