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

Johannes O. Ferstad, Angela Gu, Raymond Y. Lee, Isha Thapa, Andrew Y. Shin, Joshua A. Salomon, Peter W. Glynn, Nigam H. Shah, Arnold Milstein, Kevin A. Schulman, David ScheinkermedRxiv 2020 · DOI

Abstract: COVID-19 threatens to overwhelm hospital facilities throughout the United States. We created an interactive, quantitative model that forecasts demand for COVID-19 related hospitalization based on county-level population characteristics, data from the literature on COVID-19, and data from online repositories. Using this information as well as user inputs, the model estimates a time series of demand for intensive care beds and acute care beds as well as the availability of those beds. The online model is designed to be intuitive and interactive so that local leaders with limited technical or epidemiological expertise may make decisions based on a variety of scenarios. This complements high-level models designed for public consumption and technically sophisticated models designed for use by epidemiologists. The model is actively being used by several academic medical centers and policy makers, and we believe that broader access will continue to aid community and hospital leaders in their response to COVID-19. LINK TO ONLINE MODEL https://surf.stanford.edu/covid-19-tools/covid-19/

Introduction

In the early months of 2020, the rapid spread of COVID-19 threatened to overwhelm hospital facilities across the United States. While high-level national models existed, local leaders—many without access to dedicated epidemiological teams—struggled to understand how the virus would impact their specific communities.

The challenge was twofold: first, COVID-19 risks vary significantly by age, meaning a county's specific demographic makeup dictates its hospitalization rate. Second, local leaders needed a way to visualize the "lag" between infections and hospitalizations to make timely policy decisions. This paper addresses the gap between complex, "black-box" epidemiological models and the practical, transparent tools needed for regional operational planning.

Methods and Datasets

The authors built a deterministic, interactive simulation model deployed as a web application. The model calculates county-specific hospitalization rates by mapping US Census age distributions against age-specific morbidity and mortality data. It integrates several key data sources:

  • US Census Bureau (2018/2019): For county-level age distributions.
  • American Hospital Association (2018): For regional counts of acute and ICU beds.
  • New York Times Online Repository: For real-time confirmed case counts.
  • Imperial College COVID-19 Response Team: For age-specific rates of severe symptoms, critical symptoms, and mortality.
Regional variation in hospitalization rates per 100 cases across the United States.

Key components include growth modeling based on "doubling time," capacity mapping against local resources, and intervention simulation, which allows users to adjust doubling times to see the impact of policy changes like social distancing.

Key Insights and Ideas

  • Transparency over Complexity: By making underlying logic and assumptions (like doubling time) transparent, the model enables non-specialist policymakers to trust and manipulate the projections.
  • The "Small County" Vulnerability: Smaller counties with fewer total beds are likely to see their capacity exceeded at much lower fractions of the population being infected compared to large urban centers.
  • Sensitivity to Doubling Time: Hospitalization rates are extremely sensitive to the contagion doubling time. Even modest increases in bed capacity are insufficient if the growth rate isn't slowed through social interventions.
  • Visualizing the Lag: The tool helps leaders overcome the challenge of understanding exponential growth, specifically the 3-to-14-day lag between infection and the need for a hospital bed.

Key Results

The paper provides a sobering look at regional capacity. For instance, if only 1% of the population becomes symptomatic, Santa Cruz County, CA, would face an ICU bed utilization of 936%, and Harford County, MD, would reach 917%.

Simulation of daily hospitalizations showing the impact of intervention on acute and ICU bed demand.

As shown in the simulation above, without intervention, demand for ICU beds in a hypothetical county would exceed capacity in just 25 days. However, by simulating an intervention that increases the doubling time, the model demonstrates how demand can be kept below the critical threshold of hospital capacity.

Limitations and Future Work

The authors acknowledge several limitations inherent to the early stages of the pandemic, including parameter uncertainty regarding the "true" number of cases and doubling times. The model may also miss complex variables like pediatric bed overflow or the impact of sequential interventions. Future work involves continuously updating the model with new age-specific data as it becomes available from different global regions to refine the accuracy of the projections.

Summary updated

Abstract

ABSTRACT COVID-19 threatens to overwhelm hospital facilities throughout the United States. We created an interactive, quantitative model that forecasts demand for COVID-19 related hospitalization based on county-level population characteristics, data from the literature on COVID-19, and data from online repositories. Using this information as well as user inputs, the model estimates a time series of demand for intensive care beds and acute care beds as well as the availability of those beds. The online model is designed to be intuitive and interactive so that local leaders with limited technical or epidemiological expertise may make decisions based on a variety of scenarios. This complements high-level models designed for public consumption and technically sophisticated models designed for use by epidemiologists. The model is actively being used by several academic medical centers and policy makers, and we believe that broader access will continue to aid community and hospital leaders in their response to COVID-19. LINK TO ONLINE MODEL https://surf.stanford.edu/covid-19-tools/covid-19/
BibTeX
@misc{ferstad2020model,
  author = {Johannes O. Ferstad and Angela Gu and Raymond Y. Lee and Isha Thapa and Andrew Y. Shin and Joshua A. Salomon and Peter W. Glynn and Nigam H. Shah and Arnold Milstein and Kevin A. Schulman and David Scheinker},
  title = {{A model to forecast regional demand for COVID-19 related hospital beds}},
  year = {2020},
  doi = {10.1101/2020.03.26.20044842},
  howpublished = {medRxiv},
}
Figure