How Coronavirus Trackers Are Using AI Tools

In the absence of good data on Covid-19, which is difficult to gather as the crisis unfolds, AI is being used as a proxy to identify combinations of factors that might point to individual vulnerabilities as well as broader outbreaks. JL


Daniela Hernandez reports in the Wall Street Journal:

Software can flag disease risk and outbreak threats based on personal data, such as medical history, real-time body-temperature readings and current symptom reports, as well as demographics. Clinical AI uses a combination of lifestyle, hospitalization, emergency-room use, mortality and outcomes data. It also incorporates social determinants of health, including socioeconomic status, education and zip code.(But) current efforts can suffer from the same limitation as previous attempts to use AI in health care: incomplete information.

Tech companies, health insurers and governments are turning to artificial intelligence to predict potential coronavirus outbreaks and help guide policy decisions about social-distancing.

The software, they say, can learn to flag disease risk and outbreak threats based on personal data, such as medical history, real-time body-temperature readings and current symptom reports, as well as demographics.

The Trump administration said Thursday that states will establish their own time lines for reopening their economies.

Central to such efforts is broad and accurate testing for Covid-19, the disease caused by the new coronavirus. Yet testing lags. As many as a third of test results could be inaccurate, health experts have said. And states such as Florida and Michigan have tested only about 1% of their populations, according to state-health department and U.S. Census data.

“Until we get true, widespread testing, we have to rely on any and all signals,” said Nirav Shah, a senior scholar at Stanford University and former health commissioner for New York state who also advises health tech companies.

In early April, outreach nurses at CareATC Inc., a Tulsa, Okla.-based operator of employer-sponsored clinics, started calling some employees of clients to advise them on proper hygiene and basic social-distancing strategies, as well as what to do if they experience flu-like symptoms.

For most, the nurses didn’t have individuals’ health records; rather an artificial-intelligence system had identified those at high risk of getting really sick if they were to become coronavirus-infected, according to Kimberly Hutton, CareATC’s chief medical officer. The software flagged 255 people from a group of 4,800 associated with an employer in Utah that contracts with the health-services provider. CareATC said the system was subsequently extended to an employer in Georgia and one in Arizona.

The system, developed by clinical AI company Jvion, uses a combination of lifestyle, hospitalization, emergency-room use, mortality and outcomes data. It also incorporates social determinants of health, including socioeconomic status, education and zip code, according to John Frownfelter, the Atlanta-based company’s chief medical information officer.

“This is about prevention,” he said. “Where do we need to put resources to make sure they do shelter in place?”

Some scientists caution that the ability of AI to predict outbreaks or identify at-risk populations hasn’t been validated. Accurate testing data and large-scale contact tracing are still the gold standard for controlling outbreaks, they say.

Current tech-enabled efforts can suffer from the same limitation as previous attempts to use AI in health care: incomplete information. And, in the present crisis, unknowns abound because the virus is so new.

Paula Cannon, a virologist at the University of Southern California’s Keck School of Medicine, said the data points for understanding the spread of Covid-19 are still shaking out. “Our ability to currently use these surrogates of coronavirus infection is still being worked out in real time,” she said.

Decision makers are desperate for information to guide difficult policy choices that must balance public health and economic well-being. Some Asian countries harnessed tech-enabled surveillance programs, but their strategies were bolstered by extensive testing, especially in South Korea, Taiwan and Singapore.

Developers say there are early signs that AI can work.

In Israel, AI researchers at the Weizmann Institute of Science are collaborating with local governments and health insurers to conduct surveys that ask people if they are experiencing Covid-19 symptoms, have tested positive for the new coronavirus or came in contact with someone who has. So far, they have received more than one million responses, including several thousand from patients with confirmed Covid-19 diagnoses, said Eran Segal, the Weizmann computational biologist leading the project.

On April 5, the system, which was described in a peer-reviewed paper in the journal Nature Medicine, flagged the northern city of Umm El-Fahm as a potential coronavirus hot spot. The survey data showed a roughly 50% increase in symptoms relative to the rest of the country, Dr. Segal said.

Based on those findings, plus other available data, officials started doing more testing and found an outbreak, he said. The researchers are collaborating with public-health researchers in the U.S. to develop a similar system.Facebook Inc., in partnership with researchers at Carnegie Mellon University, has announced a similar initiative among the social network’s users in the U.S. The company plans to roll out symptom surveys internationally if early results are promising, according to a post by Chief Executive Mark Zuckerberg earlier this month.

Alphabet Inc.’s Google and Apple Inc. are planning to launch a contact-tracing app that would let health officials reverse-engineer ill patients’ whereabouts using Bluetooth signals. Participation by smartphone users would be voluntary.

Next week, St. Augustine, Fla., which enacted social-distancing rules well before the rest of the state, plans to start distributing hundreds of smart thermometers to its residents. The initiative will include West Augustine, a neighboring, largely African-American community with a lack of access to health care as well as high rates of pre-existing conditions such as diabetes, respiratory illness and high blood pressure, according to Robert Nimmons, chairman of West Augustine’s development board. Such health conditions predispose Covid-19 patients to life-threatening complications.

The pilot study, in partnership with San Francisco-based smart-thermometer maker Kinsa Inc., will monitor body-temperature trends that might point to an imminent outbreak, according to city officials and company representatives. Kinsa says random thermometer readings in other parts of the U.S. signaled atypical spikes in body temperatures days before outbreaks were reported.

Mr. Nimmons said West Augustine residents work in nursing homes, hospital cafeterias and other front-line jobs outside the community, so “it could affect everyone in the entire county if an outbreak happens.”

The pilot project will prioritize distributing thermometers to large households and those with front-line workers, as well as to neighborhoods in St. Augustine and West Augustine considered high risk in the pandemic, officials said.

St. Johns County, where St. Augustine and West Augustine are located, has roughly 190 confirmed cases so far, according to Florida’s department of health.

St. Augustine has a population of about 14,000, but as a center of tourism typically draws thousands of visitors, making it vulnerable once social restrictions are lifted.

“Making decisions in the absence of data makes it difficult,” said St. Augustine City Manager John Regan, who is leading the temperature-monitoring project. “To me, it’s a race against the clock to help improve our public-health monitoring because there’s going to be a push to reopen the economy.”