Using AI to Forecast the Growth of Covid-19

Model predicts more than 4.8 million new cases nationwide by January 1, 2021.

News

Covid-19

The Facebook AI Research Model predicts 4.87 million new cases nationwide by Jan. 1, 2021.
The Facebook AI Research Model predicts 4.87 million new cases nationwide by Jan. 1, 2021.

With Covid-19 vaccinations underway, countries continue to log a record number of coronavirus cases and deaths.

“This increase is being driven by rapid change in a number of different countries – with the runaway pandemic in the United States really leading the pack,” explained Andrew Schroeder, the head of research and analysis at Direct Relief.

On Wednesday, the U.S. reported 274,000 new cases and recorded 3,656 additional deaths, according to data collected by John Hopkins University.  With more than 17 million active cases, the United States has the highest number of per capita cases in the world. More than 300,000 Americans have died.

Globally, cases surpassed 74 million this week and deaths exceeded 1.6 million.

Direct Relief’s research and analysis team have been using Facebook-provided data, data from other sources, and AI to predict, visualize, and analyze the spread of Covid-19 in the United States. The model provides county-level predictions about the projected growth of Covid-19 and identifies which counties are most at risk. Using past predictions as a comparison, the average variance between actual cases versus model-based predictions is less than 1%.

The model predicts more than 4.8 million new cases nationwide by January 1, 2021.

In addition, the team is using data to track hospital capacity at the county level, including the number of staffed beds and critical care facilities in a particular community.

These insights are used by Direct Relief staff to guide allocation decisions about where resources are needed most, allowing the organization to provide targeted assistance. The model is also used being used by public health officials to determine how best to deploy resources, set policies, and communicate with the public on how to mitigate transmission of the virus.

U.S. Counties at Highest Risk

Based on current trends, U.S. counties most at risk include Los Angeles (CA), Cook (IL), Maricopa (AZ), Miami-Dade (FL), and San Bernardino (CA). “Really almost all of SoCal is in a very bad position, with forecast growth at some of the highest rates nationally in Riverside, San Bernardino, Orange, and San Diego,” said Schroeder. “Collectively, this may be the single greatest area of concern nationwide.”

California is expected to see an increase of nearly 700,000 cases by Jan. 1, 2021, according to the Facebook AI Research Mode.

Schroeder also flagged rural communities as areas of concern due to their high growth rates relative to population and lack of large-scale hospital resources. For example, in Sevier County, Tennessee, which has a population of 90,000, cases are expected to increase 35% over the next three weeks. The county has one hospital with 58 staffed beds and 8 staffed ICU beds, both of which are near or at capacity. “Bed occupancy at that hospital is averaging about 85% with 100% ICU utilization. About half of those totals are all Covid-19 patients,” noted Schroeder.

Sevier County, Tennessee is predicted to see a 35% increase in cases over the next three weeks.

Tennessee and Ohio have the highest case rates per capita in the country, with more than 1,000 cases per million. If either of these states were a country, they would be leading the world in case growth relative to population size. “Each one is growing rapidly – almost straight line upwards – with no sign of a peak in sight,” said Schroeder.

Ohio has one of the highest case growth rates relative to population in the country.

Direct Relief will continue to monitor the spread of Covid-19 and update predictions as data becomes available. More information on Direct Relief’s response to the Covid-19 pandemic can be found here.

Giving is Good Medicine

You don't have to donate. That's why it's so extraordinary if you do.