Probabilistic prediction of disease outbreaks with application to operational warning systems for infectious diseases in Brazil

A useful intervention and prevention method for mitigating impact from infectious disease is the use of early warning systems, particularly in order to alert for possible epidemic outbreaks. This is currently in place for diseases like dengue, zika, chikungunya and severe respiratory syndrome (SRAG) in Brazil. These warning systems would ideally use counts of the number of people currently infected with the disease in order to determine whether an alert should be sent out. These alerts mean that procedures such as the allocation of resources can be carried out in a way that is suitable to the severity of the disease outbreak that is occurring. However, the actual number of cases of the disease on a given day is almost always not known. This is caused by two main problems: the under-reporting of disease counts and the delayed reporting of disease counts. Delayed reporting is when only a proportion of the disease counts are available immediately after they occur, while the rest of the counts eventually become available but too long (sometimes months) after they have occurred. Hence, the total disease count becomes known eventually but not in time to use in a warning system. Moreover, this total count is usually not the true disease count due to some cases never being reported (under-reporting) which can happen for several reasons such as individuals being incorrectly diagnosed or not seeking medical advice. This project involves research on developing a comprehensive statistical modelling framework for correcting and predicting disease counts for use in early warning systems, with specific application to such systems in Brazil. Current state-of-the-art approaches to correcting delayed reporting are either too inflexible or too computationally intensive to be of optimal use in practice. This work aims to develop a new framework, that brings the "best of both worlds", i.e. modelling flexibility and practical feasibility, while at the same time introduce new elements such as correcting for under-reporting and the incorporation of suitable predictors.