A model to forecast regional demand for COVID-19 related hospital beds

Johannes Opsahl Ferstad; Angela Jessica Gu; Raymond Ye Lee; Isha Thapa; Andrew Y Shin; Joshua A Salomon; Peter Glynn; Nigam H Shah; Arnold Milstein; Kevin Schulman; David Scheinker

The US classified the coronavirus disease pandemic (COVID-19) as a national emergency on March 14, 2020 . Cumulative US cases surged beyond 122,000 on March 29, 2020. [2, 3] Its rapid spread in China and Italy quickly overwhelmed available hospital beds. [4, 5] County-level forecasts of demand for hospital beds based on data commonly available to hospitals would help guide US hospital efforts to anticipate and mitigate similar bed shortages. [6, 7] In order to plan their response, hospital and public health officials need to understand how many people in their area are likely to require hospitalization for COVID-19; how these numbers compare to the number of available intensive care and acute care beds; and how to project the impact of socialdistancing measures on utilization. Since the majority of people with COVID-19 are asymptomatic and the rates of cases requiring hospitalization differ significantly across age groups, answering these questions requires understanding the epidemic and accounting for the specific vulnerabilities of the local population. [8, 9] The numbers of people in each age group differ significantly across US counties, as do the available hospital resources [10, 11] . Initial analyses of the potential impact of COVID-19 have spurred governmental action at the federal and state level; however, significant differences remain in the policies implemented across states and counties. [9, 12] In order to help facilitate adequate and appropriate local responses, we developed a simple model to project the number of people in each county in the United States who are likely to require hospitalization as a result of COVID-19 given the age distribution of the county per the US Census. The model compares the projected number of individuals needing hospitalization to the publicly known numbers of available intensive and acute care beds and allows users to model the impact of social-distancing or other measures to slow the spread of the virus. The uncertainty surrounding the numbers of people infected and the rates of spread make it difficult to evaluate the accuracy of projections generated by complex epidemiological models. The model presented errs on the side of simplicity and transparency to allow non-specialist policymakers to fully understand the logic and uncertainty associated with the estimates.