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

For each US county, the model accepts as an input the number of COVID-19 hospitalizations and the associated doubling time, if these are available. If these are not available, the model imports the latest number of confirmed cases from the New York Times online repository and accepts user-entered parameters of the ratio of total cases to confirmed cases (e.g., 5:1) [8, 23] and the COVID-19 populationlevel doubling time (e.g., 7 days) [13] . The effects of interventions that mitigate the spread of infection (such as social distancing) are simulated with user-entered parameters in the form of a greater doubling time and a start date for that new doubling time. County-specific hospitalization rates are derived from combining age-distributions derived from the US census [11] and age-group specific estimates of the case rates of severe symptoms, critical symptoms, and mortality (together morbidity) derived from Imperial College COVID-19 Response Team [9] . The default assumptions are that: people are admitted to the hospital on the day they test positive (the assumptions will change when testing begins for nonsymptomatic people); those with severe and critical symptoms spend, respectively, 12 days in acute care and 7 days in intensive care; and 50% of each type of bed is available for COVID-19+ patients. [14, 15] The numbers of patients requiring each type of bed are compared to the numbers of relevant beds derived from data from the American Hospital Association. [10] The detailed technical description of the model is available in our Technical Supplement.