A Bayesian susceptible-infectious-hospitalized-ventilated-recovered model to predict demand for COVID-19 inpatient care in a large healthcare system

Abstract

The COVID-19 pandemic has strained healthcare systems in many parts of the United States. During the early months of the pandemic, there was substantial uncertainty about whether the large number of COVID-19 patients requiring hospitalization would exceed healthcare system capacity. This uncertainty created an urgent need to accurately predict the number of COVID-19 patients that would require inpatient and ventilator care at the local level. As the pandemic progressed, many healthcare systems relied on such predictions to prepare for COVID-19 surges and to make decisions regarding staffing, the discontinuation of elective procedures, and the amount of personal protective equipment (PPE) to purchase. In this work, we develop a Bayesian Susceptible-Infectious-Hospitalized-Ventilated-Recovered (SIHVR) model to predict the burden of COVID-19 at the healthcare system level. The Bayesian SIHVR model provides daily estimates of the number of new COVID-19 patients admitted to inpatient care, the total number of non-ventilated COVID-19 inpatients, and the total number of ventilated COVID-19 patients at the healthcare system level. The model also incorporates county-level data on the number of reported COVID-19 cases, and county-level social distancing metrics, making it locally customizable. The uncertainty in model predictions is quantified with 95% credible intervals. The Bayesian SIHVR model is validated with an extensive simulation study, and then applied to data from two regional healthcare systems in South Carolina. This model can be adapted for other healthcare systems to estimate local resource needs.