Bayesian sequential data assimilation for COVID-19 forecasting

TL;DR

This paper presents a Bayesian sequential data assimilation approach for COVID-19 forecasting that updates predictions in real-time using epidemic data and dynamical systems models, improving accuracy over time.

Contribution

It introduces a novel Bayesian method combining MCMC sampling with dynamical systems for real-time COVID-19 forecasting, incorporating auto-regressive models for key parameters.

Findings

Effective in forecasting COVID-19 in Mexican localities

Sequential updates improve prediction accuracy

Integrates prior knowledge with real-time data

Abstract

We introduce a Bayesian sequential data assimilation method for COVID-19 forecasting. It is assumed that suitable transmission, epidemic and observation models are available and previously validated and the transmission and epidemic models are coded into a dynamical system. The observation model depends on the dynamical system state variables and parameters, and is cast as a likelihood function. We elicit prior distributions of the effective population size, the dynamical system initial conditions and infectious contact rate, and use Markov Chain Monte Carlo sampling to make inference and prediction of quantities of interest (QoI) at the onset of the epidemic outbreak. The forecast is sequentially updated over a sliding window of epidemic records as new data becomes available. Prior distributions for the state variables at the new forecasting time are assembled using the dynamical system, calibrated for the previous forecast. Moreover, changes in the contact rate and effective population size are naturally introduced through auto-regressive models on the corresponding parameters. We show our forecasting method's performance using a SEIR type model and COVID-19 data from several Mexican localities.