Abstract
There is substantial interest in estimating and forecasting influenza incidence. Surveillance of influenza is challenging as one needs to demarcate influenza from other respiratory viruses, and due to asymptomatic infections. To circumvent these challenges, surveillance data often targets influenza-like-illness, or uses context-specific normalisations such as test positivity or per-consultation rates. Specifically, influenza incidence itself is not reported. We propose a framework to estimate population-level influenza incidence, and its associated uncertainty, using surveillance data and hierarchical observation processes. This new framework, and forecasting and forecast assessment methods, are demonstrated for three Australian states over 2016 and 2017. The framework allows for comparison within and between seasons in which surveillance effort has varied. Implementing this framework would improve influenza surveillance and forecasting globally, and could be applied to other diseases for which surveillance is difficult.
