Correcting notification delay and forecasting of COVID-19 data

Since the first official case of COVID-19 was reported, many researchers around the world have spent their time trying to understand the dynamics of the virus by modeling and predicting the number of infected and deaths. The rapid spread and highly contagiousness motivate the necessity of monitoring cases in real-time, aiming to keep control of the epidemic. As pointed out by [3], some pitfalls like limited infrastructure, laboratory confirmation and logistical problems may cause reporting delay, leading to distortions of the real dynamics of the confirmed cases and deaths. The aim of this study is to propose a suitable statistical methodology for modeling and forecasting daily deaths and reported cases of COVID-19, considering key features as overdispersion of data and correction of notification delay. Both, reporting delays and forecasting consider a Bayesian approach in which the daily deaths and the confirmed cases are modelled using the negative binomial (NB) distribution in order to accommodate the population heterogeneity. For the correction of notification delay, the mean number of occurrences regarding time t notified at time [Formula: see text] (mean delayed notifications) is associated to the temporal and the delay lag evolution of the notification process through a log link. With regard to daily forecasting, the functional form adopted for the number of deaths and reported cases of COVID-19 is related to the sigmoid growth equation. A variable regarding week days or days off was considered in order to account for possible reduction of the records due to the lower offer of tests on days off. To illustrate the methodology, we analyze data of deaths and infected cases of COVID-19 in Espírito Santo, Brazil. We also obtain long-term predictions.