Play Live Radio
Next Up:
0:00
0:00
Available On Air Stations
KPBS Evening Edition

UCSD Researchers Build Database To Predict Future COVID-19 Deaths, Cases

A schematic representation of the Global Epidemic and Mobility (GLEAM) model, created by UCSD researchers, for the United States, October 20, 2020.
UC San Diego
A schematic representation of the Global Epidemic and Mobility (GLEAM) model, created by UCSD researchers, for the United States, October 20, 2020.

Many public health officials across the country, including San Diego, are already using resources like the John Hopkins database to find out in real-time how well their respective cities and towns are flattening the coronavirus curve in real-time.

What if you want to know what those numbers could be weeks from now? San Diego researchers have stepped in to fill the data gap.

UCSD Researchers Build Database To Predict Future COVID-19 Deaths, Cases
Listen to this story by Shalina Chatlani.

UC San Diego computer scientist Yian Ma says there were few resources to tell public health officials how factors like travel and population density could impact a COVID-19 curve in the future.

“People are moving around and that makes this pandemic especially troublesome because you might have a county having an outbreak and then is passing it on to another county,” Ma said.

Ma and his colleague, UC San Diego scientist Rose Yu, developed a new machine learning model that pairs existing COVID-19 data with resources like travel data to predict how many hospital beds a city may need next week.

“We have the travel data, travel statistics on-air network commuting patterns. And we constructed a network for the travel pattern. We also used ground truth observations, about the number of hospitalization, deaths and infection in each different county in the United States,” Ma said.

The model uses data like death certificates, previous movements of COVID-19 victims and information on counties’ opening restrictions. They combined that with machine learning, to build algorithms that could predict how well people adhere to certain coronavirus restrictions in their respective states.

Researchers say right now the model is most accurate one week out and can predict two to four weeks from now with relative success.

“[Public health officials] are making decisions about when to allocate vaccines, how many, ventilators and hospital beds we're going to allocate to different counties,” Yu said. “These are all the important decisions that they need to make every day. And in order to make the decisions, they need to know beforehand, for example, how many deaths or hospitalization or infections a certain county will get in the coming four weeks.”

The scientists say the resource could come in handy for policymakers.

“Gov. Newsom from California was asking us to make predictions about a number of hospital beds that would be used throughout different counties so that we can be better allocated,” Ma said.

The database has been incorporated into the centers for disease control website.

San Diego came close but was not quite pushed into the most restrictive Covid-19 tier due to the county’s high testing rate. Also we’ll review measures B, C and D on this year’s ballot. Plus, San Diego’s Asian Film festival returns.