TODAY'S PAPER
85° Good Afternoon
85° Good Afternoon
BusinessCoronavirus

LI artificial intelligence startup predicts where COVID-19 will spike

Klaus Mueller, CEO and co-founder of Akai Kaeru

Klaus Mueller, CEO and co-founder of Akai Kaeru LLC, says the Stony Brook firm's model can predict counties that are likely to see higher COVID-19 death rates. Credit: Akai Kaeru LLC

A Long Island artificial intelligence startup has built software aimed at pinpointing U.S. counties where the COVID-19 outbreak is likely to be most deadly.

In a June report, the data-mining company, Akai Kaeru LLC, forecast spiking COVID-19 mortality with the heaviest concentrations in counties of the Southeast, including Mississippi, Georgia and Louisiana, said co-founder and chief executive Klaus Mueller.

Nationwide, the software found 985 out of all 3,007 U.S. counties are at risk.

"These patterns identify groups of counties that have a steeper increase in the death-rate trajectory," he said.

Closer to home, the software found Nassau and Suffolk counties are likely to be relatively stable, but Westchester and Rockland counties are potential tinderboxes that could tip into crisis, said Mueller, a computer science professor at Stony Brook University.

The factors making Westchester and Rockland more vulnerable to a spike in mortality include areas with more crowding and fewer residents with access to cars, he said.

"They need to be very careful with reopening," Mueller said of the northern suburbs. "It just takes a spark for there to be a second wave."

At the same time, he said, Long Island "is not out of the woods" and abandoning policies like social distancing could lead to a new surge.

The software analyzes more than 500 attributes related to demographics, economics, infrastructure, race and ethnicity as well as deaths and other health data directly related to COVID-19.

Dr. Isaac Weisfuse, an epidemiologist and adjunct professor at Cornell University's public health program, said that data-mining software is used by public health departments.

If the software provides sufficient warning, he said, preventive measures like screening and mask-wearing policies can be instituted.

"It's more valuable if it's accurate two months before, but it's still valuable two weeks before," he said.

The Centers for Disease Control and Prevention aggregates mortality forecasts from about two dozen software programs and expects 140,000 to 160,000 total reported COVID-19 deaths in the United States by July 25.

While many COVID-19 models provide specific fatality forecasts at the state level, the Akai Kaeru software is one of the few that assesses risk at the county level.

Mueller said that based on the one-month snapshots, the software is finding that counties at the highest risk have a death rate that grew two- to three times more than the United States, overall.

In June, the fatality rate for U.S. COVID-19 related deaths was 24.1 per 100,000 population, he said.

Aside from finding geographies in jeopardy, the software is able to unearth specific — and sometimes surprising — combinations of factors that appear to be connected to counties with higher death rates.   

For instance, counties with low poverty levels, high homeownership rates, but high levels of housing debt were found to be at high risk.

"The more housing debt you have, the more death you have," Mueller said.

Other counties at risk had a combination of residents who were sleep-deprived (according to data from the CDC) and had low levels of education and low rates of health insurance coverage.

Another group of counties had few Asian residents but high overall minority populations, including impoverished Black children. 

Rural counties with high poverty rates and an aging population also were deemed at risk.

"One of the defining characteristics is we focus on explainability," said Eric Papenhausen, chief technology officer and co-founder of the company. "You can create a narrative around it," which can lead to changes in public policy.

Akai Kaeru is based at the Center of Excellence in Wireless and Information Technology on the Stony Brook University campus.

The 4-year-old company, whose name is Japanese for red frog, has raised $1 million in funding from the National Science Foundation's Small Business Innovation Research program and about $200,000 through the New York State Strategic Partnership for Industrial Resurgence program and the New York State Center for Advanced Technology.

The COVID-19 software is a demonstration project for the company, whose data-mining software can be applied to a variety of tasks, including assessing mortgage risk, speeding drug discovery and investment analysis.

Another startup, Manhattan-based Dataminr, is seeking to use social media posts as a leading indicator of COVID-19 infections at the county level.

Artificial intelligence refers to the ability of software programs to learn and perform actions previously reserved for humans.

Mueller said his company's "explainable AI" is not a black box and can provide insight into how the software reached its conclusions.

A note to our community:

As a public service, this article is available for all. Newsday readers support our strong local journalism by subscribing.  Please show you value this important work by becoming a subscriber now.

SUBSCRIBE

Cancel anytime

More news