Transmission matrix parameter estimation of COVID-19 evolution with age compartments using ensemble-based data assimilation

Abstract

The COVID-19 pandemic, with its multiple outbreaks, has posed significant challenges for governments worldwide. Much of the epidemiological modeling relied onpre-pandemic contact information of the population to model the virus transmissionbetween population age groups. However, said interactions underwent drastic changesdue to governmental health measures, referred to as non-pharmaceutical interventions.These interventions, from social distancing to complete lockdowns, aimed to reducetransmission of the virus. This work proposes taking into account the impact of nonpharmaceutical measures upon social interactions among different age groups by estimating the time dependence of these interactions in real time based on epidemiological data. This is achieved by using a time-dependent transmission matrix of the diseasebetween different population age groups. This transmission matrix is estimated usingan ensemble-based data assimilation system applied to a meta-population model andtime series data of age-dependent accumulated cases and deaths. We conducted aset of idealized twin experiments to explore the performance of different ways in whichsocial interactions can be parametrized through the transmission matrix of the metapopulation model. These experiments show that, in an age-compartmental model, allthe independent parameters of the transmission matrix cannot be unequivocally estimated, i.e., they are not all identifiable. Nevertheless, the time-dependent transmissionmatrix can be estimated under certain parameterizations. These estimated parameterslead to an increase in forecast accuracy within age-group compartments compared to asingle-compartmental model assimilating observations of age-dependent accumulatedcases and deaths in Argentina. Furthermore, they give reliable estimations of the effective reproduction number. The age-dependent data assimilation and forecasting of virustransmission are crucial for an accurate prediction and diagnosis of healthcare demand.