Estimating SARS-CoV-2-positive Americans using deaths-only data
share
We fit a Bayesian model to data on the number of deaths attributable to COVID-19 with the goal of estimating the number of infected individuals. Our model links an underlying Susceptible Infectious Removed (SIR) model of disease dynamics to observed deaths via a time-to-death distribution informed by previous studies. This allows us to actually fit a statistical model to the data, unlike many epidemiological studies in which the SIR model parameters are simply "calibrated" to obtain outputs that look similar to the real data. The main outputs of our model are estimates of the number of infections currently, as well as forecasts of the number of infections and deaths under various scenarios for the effectiveness of social distancing measures. All of our outputs have attached Bayesian credible intervals. An important conclusion is that the confirmed case counts greatly underestimate the total number of infected individuals.
READ FULL TEXT
