Locally Informed Modeling to Predict Hospital and Intensive Care Unit Capacity During the COVID-19 Epidemic

Background: In the early phases of the 2019 novel coronavirus (COVID-19) pandemic, health system leaders faced the urgent task of translating the unknown into forecasting models for hospital capacity. Our study objective was to demonstrate the application of a practical, locally informed model to estimate the hospital capacity needed even though the community COVID-19 caseload was unknown. Methods: We developed a susceptible-infected-recovered (SIR) model that was adopted from the University of Pennsylvania COVID-19 Hospital Impact Model for Epidemics and employed at 8 hospitals within Ochsner Health, the largest integrated delivery system in Louisiana, between March 16 and April 15, 2020. Intensive care unit (ICU) admissions of cases in the New Orleans area were used to estimate the community case load when testing was delayed. Results: Initially, the observed ICU census trended near R(0)=2.0, whereas the ventilator census trended between R(0)=2.0 and 3.0. After implementing social distancing, both the ICU and ventilator capacity trended toward R(0)=1.3, while non-ICU medical/surgical beds trended toward R(0)=1.5. The model accurately predicted peak ICU (n=250) and hospital bed (n=487) usage by April 6, 2020. In response to model trends, Ochsner added 130 ICU beds across its hospitals by opening a new ICU and converting operating rooms and parts of emergency departments to ICU beds. Conclusion: When disease testing is limited or results are delayed, ICU admissions data can inform SIR models of the rate of spread of COVID-19 in a community. Our model used various R(0) plots to demonstrate an array of scenarios to guide planning for hospital and political leaders.