RTI uses cookies to offer you the best experience online. By clicking “accept” on this website, you opt in and you agree to the use of cookies. If you would like to know more about how RTI uses cookies and how to manage them please view our Privacy Policy here. You can “opt out” or change your mind by visiting: http://optout.aboutads.info/. Click “accept” to agree.


Improving Resolution of Hospital Modeling Across the United States


Hospital-flux simulator aims to signal surges 30 days in advance

In December 2020, the U.S. Department of Health and Human Services (HHS) began sharing detailed data on Coronavirus Disease 2019 (COVID-19) hospitalizations and hospital capacity, including the average number of patients in hospital and intensive care units (ICU) beds each week. As the past year has shown, a high proportion of reported COVID-19 cases are characterized as severe, requiring hospitalization in acute treatment units or ICUs. While our understanding about why some individuals progress towards severe disease and others do not is still evolving, it is clear that risk factors for severe disease include age and underlying health conditions. Based on the U.S. Centers for Disease Control and Prevention (CDC) COVID-19-Associated Hospitalization Surveillance Network (COVID-NET) data, that describes trends in clinical outcomes, “15.9%, or about 1 in 6 hospitalized COVID-19 patients, were admitted to the ICU.”

Since the pandemic began, the proportion of severe cases has remained consistent; a growing concern focuses on the threat of new variants – whether they are more transmissible and pathogenic, resulting in increased transmission, greater incidence of severe disease, and possibly death.  While the promise of enough vaccines to reach herd immunity in the United States is heartening, the weeks and months until this stage is achieved will—especially with increasing COVID-19 pandemic fatigue—continue to challenge health systems and hospitals.

Combating COVID-19 with Computing

The COVID-19 pandemic has been an accelerator for the automation of data modeling and computing. Recently, Dr. Donal Bisanzio and Dr. Rainer Hilscher led a team of data scientists and epidemiologists to develop an innovative, high-resolution hospital-flux simulator (at the facility level) to describe COVID-19 case flow across the United States and identify those locations with a higher risk of increased hospital and ICU admissions. The simulator was added to RTI’s COVID-19 Data Insights Tool. Figure 1 overlays the average of ICU bed occupancy rates projected by the model for a series of select states, displaying these averages over the projected national average for the past month and 30-days into the future.

Figure 1. Hospital-flux Simulator Offers 30-day Estimates Comparing select state averages to the national average of projected ICU bed occupancy rates from March 25 to May 25, 2021, based on model output

Similar to other hospital modeling efforts, RTI leverages state-level COVID-19 hospitalization estimates provided weekly by the Institute of Health Metrics and Evaluation (IHME) and county-level case data daily from the Johns Hopkins Coronavirus Resource Center (CRC). RTI's Merge™, an AI-driven research platform, actively monitors and refreshes the simulation as hospital-level acute and ICU occupancy rates are updated.

But unlike other model platforms, the hospital-flux simulator combines these rates with community-level factors—daily case reporting, high-resolution population demography (using RTI SynthPop™), and IHME hospitalization rates—to project hospital bed occupancy for a given hospital, hospital system or the entire United State 30 days into the future. The model simulates the percent of total beds and ICU beds projected to be in use for 2,619 individual acute care hospitals across the country.

By leveraging RTI’s recent advancements in scalable computing and harnessing the power of the RTI Merge platform, Dr. Bisanzio and Dr. Hilscher's hospital-flux simulator has the ability to downscale state hospitalization estimates to the facility-level and project them 30 days into the future to provide early signals of potential surges, and when and where facilities may reach full capacity. For example, at the end of March 2021, the model projected—based on prevailing COVID-19 trends—20% of hospitals in Michigan could reach the capacity of their ICU beds (>75% bed occupancy) by the end of April; as of April 19, 2021, hospitals around Detroit were already filling up, reaching up to 95% bed capacity. In Figure 2, while the current risk remains high across Michigan counties, the hospital-flux simulator projects ICU occupancy rates should decline over the next 30 days.

Figure 2. Overlaying current county-level COVID-19 risk with future ICU bed use for hospitals in Michigan Model projects decline in ICU bed use for communities at great risk, showing projections for May 25, 2021 from RTI COVID-19 Data Insights Tool

Continuing to Support Community Response

The recent uptick in localized case counts and increases in positivity rates raises concerns about possible surges in cases for the weeks ahead. Amplified by vaccine hesitancy and pandemic fatigue, it is unclear how long and how severe cases may be over the summer months, reiterating the critical role of hospitals in this response. Using RTI's COVID-19 Data Insights Tool, users can view hospitals in their community layered with community-level social and epidemiological data to identify areas that could be of greater risk and vulnerability. Decision-makers can also feel more empowered to make data-driven decisions regarding their localities and public health interventions. The more informed we all are, the better we can anticipate, prepare, and respond to this and any future public health crises.

If you are interested in adopting this model or learning more about its data sources and creation, contact our team. 


The authors would like to extend their appreciation to the RTI Catapult team, including Trevor Downey, Anne Marie Miller, and Sam Fenimore, along with Dr. Pia MacDonald, and the RTI Infectious Disease Response Team for their support and resources.

Note on Modeling

As generally accepted, model predictions are prone to error/imprecision and rely on the availability and quality of data, despite best efforts to date around the concerted efforts of HHS and CDC to boost data availability. The hospital-flux simulator uses an individual base model simulating the flux of severe COVID-19 cases living in hospitals’ catchment areas and allocating cases requiring hospitalization to hospitals by taking into account case rates across the community and all available hospitals’ capacity. The model does not account for local alternate medical sites (or ”Field Medical Stations“) set up to support hospitals and local surge in patients; as an example, the model excludes the two alternative medical sites previously opened in Massachusetts'. Occupancy data is consistent with other sites.

Disclaimer: This piece was written by Donal Bisanzio (Senior Epidemiologist), Rainer Hilscher (Senior Research Data Scientist), Richard Reithinger (Distinguished Fellow, Global Health), Christine Bevc (Research Public Health Analyst), and James Rineer (Director, Geospatial Science and Technology) to share perspectives on a topic of interest. Expression of opinions within are those of the author or authors.