By Daniel Dunaief

About 40 percent of the counties in the United States are at high risk for COVID-19 and related death rates, according to a new computer program created by Stony Brook University Computer Science Professor Klaus Mueller.

Putting together data from the over 3,000 counties throughout the United States, Mueller used a computer program he created with a start up company he co-founded, called Akai Kaeru LLC, to search for counties that present factors that would put them at greater risk for an increase in COVID-19 deaths.

Analyzing data from 500 factors, the scientists found that death rates increased in communities with a combination of traits that are catalytic for the spread and fatality rate of the virus. These include sparsely populated counties with a poor and aging population; counties with sleep-deprived, low-educated, low-insured residents; and wealthy counties with high home ownership and increasing housing debt, among other factors.

Many of the counties are in the southern United States. In June, Mississippi, Louisiana and Georgia had the highest density of high-risk counties at a coverage of 80 to 90 percent.

Mueller said he considered this approach in late April. When the data from the Centers for Disease Control and Prevention came online, the group did its first test run on May 10th, which ended on June 10th.

When he looked at the June 10th mortality rates throughout the country, he was amazed at how effectively the patterns based on the conditions from the computer algorithm predicted increases.

To be sure, not all of the counties that fit one or more of these sets of conditions had high death rates in May, but others that were similar had. The preconditions existed, but the spark to cause those deaths hadn’t occurred, Mueller said.

“In June, some of these so far untouched counties caught the virus and they flared up like a tinderbox,” Mueller explained in an email. “This phenomenon continued in July for other counties that had escaped it so far but had the critical condition sets.”

In June, some of the counties that had characteristics that made them vulnerable caught the virus, Mueller explained.

Mueller anticipates a rapid increase in August in counties in Florida and Texas, in which the virus has spread and the conditions for increased mortality are prevalent.

“There are counties in these states that from the socio-economic perspective look a lot like those that already experienced great tragedy,” he wrote.

Mueller explained that people in many counties think they’re not at risk even if their neighbors are. The danger, however, comes from a spark, such as a visit by someone carrying the virus, that increases the infection, hospital and mortality rates.

Indeed, in wealthy counties where residents are stretched thin by the costs needed to maintain their homes, the incidence of illness and death is also higher. In part, that reflects how some of the people in these communities cut corners in terms of health insurance.

Mueller said Akai Kaeru, which means “red frog” in Japanese, is working on a dashboard that will be accessible from a web browser where users can click on a map of counties and see the risk and the patterns that define it. The staff at Akai Kaeru, which includes three principals and four interns, have virtual team meetings each weekday at 11 am. The dashboard they create can help residents see the other counties that share similar characteristics. Users can also compare the death rate in these counties to the average death rate in the United States.

While the observations of trends linking characteristics of a county with COVID-related health challenges could be useful for county and state planners, Mueller acknowledged that these observations are “just a start. Now, you know where to look, which is way better than before.”

The data could be useful for policy and law makers as well as for actuaries at life insurance companies.

Mueller believes this artificial intelligence tool acts like a magnet that pulls out the proverbial needle from the data haystack.  Local leaders can use the dashboard to see the critical conditions for their counties. They can try to find solutions to remove those conditions.

Demonstrating how the health care system in similar areas became overwhelmed can increase compliance with social distancing and mask-wearing guidelines.

Mueller added that the predictions from the model are only as good as the data he used to analyze trends across the country. He and his team aren’t making these observations or collecting this information themselves.

He said some counties have a lower likelihood than the average of developing a wider contagion. While the entire state doesn’t have the same lower probability of the disease spreading, areas like Montana and Indiana have fewer of the variables that typically combine to create conditions that favor the spread of the virus.

Mueller suggests that the risks from COVID-19 are tied to compliance with policies that reduce the spread of the disease and to the development of a vaccine.

Despite the high infection rate through April and May and the deaths during those unprecedented months, Suffolk County isn’t at the same level of risk as some regions in the south. “Suffolk is much better than those counties in the South and even Westchester, Rockland and adjacent counties in Connecticut and New Jersey,” Mueller said. “But it is not without risk.”

Prior to developing a program to analyze epidemiological trends with COVID, Mueller worked with medical visualization, which included the three-dimensional data of human parts that were generated through computed tomography, or CT.

In his work, the Computer Science professor seeks to find ways to communicate high-dimensional data to the lay population. He has routinely worked on clustering and has partnered with Pacific Northwest, Brookhaven National Laboratory, and health care companies.

Mueller has been at Stony Brook University since 1999. He earned his PhD from Ohio State University. Originally from Germany, he has done considerable work online, including teaching.

He and his wife Akiko, who works on marketing for his company, have an eight-year-old daughter named Nico.

Readers interested in learning more about his research with COVID can find information at the following link: https://akaikaeru.com/covid-19-1.