Skip to Main Content
COVID-19 ISO Insights

New Models and AI May Improve COVID-19 Case Estimates

March 15, 2021

By: Christopher Sirota, CPCU

COVID-19 tracking websites, such as the one provided by the Centers for Disease Control and Prevention (CDC), typically rely, in part, on counting cases of infection confirmed by testing. But how many infectious people are actually among the population? And where might hotspots be in the next 7 days?

Researchers are reportedly trying to address these questions with new data models and help from some artificial intelligence.

A New Model Suggests Multiplying By Ten

NPR has reported about a new study, not yet peer-reviewed, that seeks to better estimate how many people may be infected but have not been tested.

According to the article, researchers at Columbia University have developed a new mathematical model which indicates that to better estimate the number of infectious people in the U.S. for a single day, one should multiply daily reported cases by four; however, to better estimate the level of infection risk in a community, the article notes that one needs to account for the average three to four days an infected person is shedding virus—perhaps asymptomatically—by multiplying the daily reported cases by ten.

The Modeling Data

The new model reportedly sourced data that included the total number of positive cases since the beginning of the pandemic in 2020 and "anonymized cellphone location data — provided by the company SafeGraph — that told [researchers], for each day, how much people were intermingling by moving outside of their homes […]" Some earlier estimates, using "more rudimentary" techniques and some extrapolation of blood testing data, reportedly also arrived at factors of 8 and 10 times the number of reported cases to better gauge the spread of the infection.

Is the U.S. Near Herd Immunity?

Based on the new model, the researchers estimate that about half of the U.S. population, around 120 million people, have likely been infected, with the percentage varying by state. Such a high percentage might suggest the possibility for herd or community immunity--when enough people have survived the infection,developed immunity as a result, and the virus can no longer effectively spread. Unfortunately, herd immunity in the U.S. likely needs to reach 90%, according to Harvard Medical School; NPR notes a percentage higher than 50% is especially more likely due to the unknown length of immunity gained from infections and how such immunity may or may not protect against the new variants of the COVID-19 virus.

The Next Seven Days: Forecasting COVID-19 in Your Community

Axios has reported on a startup that is combining multiple sources of data and leveraging artificial intelligence (AI) to help predict outbreaks of COVID-19 at the community level about a week into the future.

The startup, Fludemic, reportedly provides a high-level version of its forecast map for the public at its website. According to Axios, the startup repurposed its previously developed influenza model for COVID-19, and considers current case forecasts to be about 92% accurate.

Fludemic's website notes that it is collaborating with the University of Berkeley, California's Smart Pandemic Management initiative.

You will soon be redirected to the 3E website. If the page has not redirected, please visit the 3E site here. Please visit our newsroom to learn more about this agreement: Verisk Announces Sale of 3E Business to New Mountain Capital.