Keywords: COVID-19, NPIs, hierarchical modelling, R_t, Bayesian modelling
TL;DR: In this paper we analysed the impact of non-pharmaceutical interventions in the early stage of the COVID-19 pandemic in Brazil.
Abstract: Since the beginning of the pandemic, Brazil has reported the second highest number of COVID-19 deaths in the world. Here we characterise the early transmission that seeded the country-wide spread of the disease, and assess attempts to attenuate the spread through implementing non-pharmaceutical interventions (NPIs) at subnational level. The analysis presented uses a Bayesian hierarchical approach to model transmission based on mortality data. The statistical model encodes a causal inferential bias for generic infectious disease transmission --- deaths are generated by infections which arise from earlier infections. As transmission is heterogeneous at subnational level, from differences such as the timing of seeding and hospital capacities, this is modelled by partially pooling parameters across geographic regions, using state-level mobility covariates for the reproduction number ($R_t$), and through inference of region-specific epidemiological parameters. We report extensive heterogeneity in the initial epidemic trajectory across Brazil underscoring the importance of sub-national analyses in understanding asynchronous state-level epidemics underlying the national spread and burden of COVID-19.