The sudden emergence of infectious diseases pose threats to societies worldwide and it is notably difficult to detect. In the past few years, several early warning signals (EWS) were introduced, to alert to impending critical transitions and extend the set of indicators for risk assessment. While they were originally thought to be generic, recent works demonstrated their sensitivity to some dynamical characteristics such as system noise and rates of approach to critical parameter values. Moreover, testing on empirical data is so far limited. Hence, validating their performance remains a challenge. In this study, we analyse the performance of common EWS such as increasing variance and autocorrelation in detecting the emergence of COVID-19 outbreaks in various countries, based on prevalence data. We show that EWS are successful in detecting disease emergence provided that some basic assumptions are satisfied: a slow forcing through the transitions and not fat-tailed noise. We also show cases where EWS fail, thus providing a verification analysis of their potential and limitations. Overall, this suggests that EWS can be useful for active monitoring of epidemic dynamics, but that their performance is sensitive to surveillance procedures. Our results thus represent a further step towards the application of EWS indicators for informing public health policies.