Weekly Bayesian modelling strategy to predict deaths by COVID-19: a model and case study for the state of Santa Catarina, Brazil

Background: The novel coronavirus pandemic has affected Brazil's Santa Catarina State (SC) severely. At the time of writing (24 March 2021), over 764,000 cases and over 9,800 deaths by COVID-19 have been confirmed, hospitals were fully occupied with local news reporting at least 397 people in the waiting list for an ICU bed. In an attempt to better inform local policy making, we applied an existing Bayesian algorithm to model the spread of the pandemic in the seven geographic macro-regions of the state. Here we propose changes to extend the model and improve its forecasting capabilities. Methods: Our four proposed variations of the original method allow accessing data of daily reported infections and take into account under-reporting of cases more explicitly. Two of the proposed versions also attempt to model the delay in test reporting. We simulated weekly forecasting of deaths from the period from 31/05/2020 until 31/01/2021. First week data were used as a cold-start to the algorithm, after which weekly calibrations of the model were able to converge in fewer iterations. Google Mobility data were used as covariates to the model, as well as to estimate of the susceptible population at each simulated run. Findings: The changes made the model significantly less reactive and more rapid in adapting to scenarios after a peak in deaths is observed. Assuming that the cases are under-reported greatly benefited the model in its stability, and modelling retroactively-added data (due to the "hot" nature of the data used) had a negligible impact in performance. Interpretation: Although not as reliable as death statistics, case statistics, when modelled in conjunction with an overestimate parameter, provide a good alternative for improving the forecasting of models, especially in long-range predictions and after the peak of an infection wave.