Weekly sequential Bayesian updating improves prediction of deaths at an early epidemic stage

TL;DR

This paper introduces weekly sequential Bayesian updating methods to improve early-stage epidemic death predictions, making models less reactive and more adaptable using under-reporting adjustments and mobility data.

Contribution

The study proposes novel Bayesian updating variations that incorporate under-reporting and reporting delays, enhancing early epidemic death forecasting accuracy.

Findings

Models become less reactive and adapt faster after peaks.

Under-reporting assumptions improve model stability.

Case data with overestimation parameters aid long-range predictions.

Abstract

Background: Following the outbreak of the coronavirus epidemic in early 2020, municipalities, regional governments and policymakers worldwide had to plan their Non-Pharmaceutical Interventions (NPIs) amidst a scenario of great uncertainty. At this early stage of an epidemic, where no vaccine or medical treatment is in sight, algorithmic prediction can become a powerful tool to inform local policymaking. However, when we replicated one prominent epidemiological model to inform health authorities in a region in the south of Brazil, we found that this model relied too heavily on manually predetermined covariates and was too reactive to changes in data trends. Methods: Our four proposed variations of the original method allow accessing data of daily reported infections and take into account the 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. That workflow allowed us to run a lighter version of the model after the first calibration week. Google Mobility data, weekly updated, were used as covariates to the model 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 reported cases were 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 on 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.