BIDMC appointed a research group within the Center for Healthcare Delivery Science to apply epidemiology, machine learning and causal inference to forecast where COVID-19 could surge next.

Becker’s Hospital Review – July 17, 2020

The predictive analytics tools hospitals are using to forecast COVID-19 case surges

Hospitals nationwide have been utilizing predictive analytics tools to project the location and severity of future COVID-19 outbreaks.

Though COVID-19 forecasting tools vary on the data they use to make predictions, they all seek to help healthcare providers make informed decisions about care and how to best utilize their resources.

Boston-based Beth Israel Deaconess Medical Center appointed a research group within the Center for Healthcare Delivery Science to apply epidemiology, machine learning and causal inference to forecast where COVID-19 would surge next. The 670-bed academic medical center began to formulate its response in February, relying on national models to predict how the virus would spread. However, national models didn’t consider socioeconomic factors or local hospital and community decision-making. The research team created a hyper-local alert system, integrating a preliminary Susceptible, Infected and Recovered people model into the hospital’s incident command structure. The researchers also relied on machine learning to take real-time data from the EHR to examine disease characteristics like incubation time, infection period and transmissibility, which allowed them to predict peaks and declines five days before the national models.

Children’s Hospital of Philadelphia developed a tool to track COVID-19 cases in June that is the first of its kinds to utilize historic weather data. The tool uses temperature and humidity data from 389 U.S. counties experiencing some level of COVID-19 activity to predict the severity of future surges. COVID-19 transmission often seems to rise as temperatures do, which researchers say is likely attributed to people socializing more outside and the virus staying on surfaces longer during warmer months. The model also uses GPS data to track each county’s increase in visits to nonessential locations and takes into account demographic data, such as population density, poverty levels, number of people older than 65 and number of people with preexisting conditions.

Researchers at Cleveland Clinic developed a prediction model to forecast the likelihood of patients testing positive for COVID-19 and the potential disease outcomes in June. It makes projections based on age, race, gender, socioeconomic status, vaccination history and current medications. Cleveland Clinic researchers developed the tool, a nomogram, with data from about 12,000 patients enrolled in its COVID-19 registry. The tool is available freely online as a risk calculator, which can be found here.

In June, Chicago-based CommonSpirit Health created a COVID-19 forecasting tool using deidentified cellphone, public health and health system data. The health system took into consideration fixed data, including population and availability of healthcare providers, as well as variable data, including social-distancing relaxation and new cases. The collection of cellphone data used to show how much people travel outside their communities is HIPAA-compliant and can’t be associated with anyone. CommonSpirit can generate a predictive outlook for about 75 percent of its markets, including Texas, California, Arizona and the Pacific Northwest.

Dallas-based Parkland Center for Clinical Innovation created a COVID-19 vulnerability calculator in June to measure communities’ vulnerability to COVID-19 by tracking and analyzing their comorbidity rates, population over age 65, social factors and ability to observe stay-at-home measures. The model determined social deprivations, such as inadequate access to food, medicine, employment and transportation, as the largest contributor to higher COVID-19 mortality rates among Black and Latinx communities. Local healthcare providers can use the vulnerability index as a tool to better tailor their COVID-19 response to the neighborhoods that need it most, deploy more testing and education in at-risk areas, and plan culturally sensitive initiatives to address infection disparities in Black and Latinx communities.

A team of researchers working at Ames-based Iowa State University, led by Lily Wang, PhD, developed a COVID-19 forecasting dashboard in June. The dashboard provides users with a seven-day rolling forecast and four-month predictions of COVID-19 infections and deaths, which can be viewed at both state and county levels. It analyzes data from Baltimore-based Johns Hopkins University, The New York Times, The Atlantic, the World Health Organization, the CDC and USAFacts. To produce county-specific predictions, it also uses data from state and county public health departments, censuses, local databases, mobility tracking and government news releases. Yuan Gu, a graduate student at Pittsburgh-based Carnegie Mellon University, is currently developing a mobile version of the dashboard.