Title: Inferring True Historical Evolution of COVID-19 from Death Incidence

Abstract:

Countries around the world have instituted social distancing interventions at a level thought unimaginable prior to this outbreak of SARS CoV-2. However, determining when to lift these interventions is dependent on accurate estimates of the impact of the disease over time as a result of these interventions. These estimates are complicated by several factors such as: data quality issues due to high numbers of asymptomatic patients who may transmit the disease yet are difficult to detect, lack of testing resources, failure of recovered patients to be counted, delays in reporting hospitalizations and deaths, and the co-morbidity of other life-threatening illnesses. Other ideas have been proposed to infer the true infection counts from deaths using Monte Carlo estimates but this still leaves us without an accurate estimate of exposed individuals. Social distancing interventions directly affect the rate of exposure to the virus. Without an accurate estimate of this figure to seed a compartment model, it is difficult to measure the impact of such an intervention on the spread of the disease. Here we use a recently published linear noise approximation with Markov Chain Monte Carlo (MCMC) sampling technique for uncovering the hidden disease states using reported death incidence as the observed variable. We use literature-derived clinical estimates of Infection Fatality Ratios and transition probabilities between disease states. We further demonstrate the validity of this approach by comparing against the serology reports for New York state. Using initial conditions derived from this approach we simulate the future evolution of disease with a Markov SEIRD model while implementing various social distancing and mobility restrictions.