· 6 min read
Project investigating fever-related data as early indicator of COVID-19 outbreaks
Putting a smart thermometer to the ear could mean putting an ear to the ground for future COVID-19 outbreaks and the consequences of relaxing social distancing, says a University of Nebraska–Lincoln engineer.COVID-19 hotspots in Nebraska up to weeks before new outbreaks are officially reported. With a boost from that data and machine learning, the researchers are also busy constructing a model that might better predict how the spread of the novel coronavirus will respond to the relaxation of social distancing guidelines.
Since late 2014, Kinsa has sold or donated more than a million thermometers that, with a user’s approval, can anonymously and wirelessly transmit temperature data to the cloud. Because its thermometers transmit the ZIP codes associated with high-temperature readings, Kinsa has spent several years tracking the prevalence, timing and geography of U.S. fevers down to the county level. And given that fevers often emerge as a response to influenza viruses, the company has shown that its data can help reasonably predict the number and seasonality of flu cases in a typical year.
That predictability — and the fact that 2020 is very much atypical — has also yielded an opportunity to track and even predict outbreaks of the novel coronavirus. Though the majority of people infected with the coronavirus do not exhibit symptoms, up to 90% of those who do will get a fever, according to the World Health Organization. But the relatively long incubation period of the novel coronavirus, combined with still-sparse levels of testing in some areas, has created a notable lag between outbreaks and confirmations of COVID-19 cases.
By comparing the five-year average number of fevers at a given place and time with their corresponding incidence in 2020, then identifying the areas with substantial spikes in fevers, Kinsa has reported promising efforts to forecast coronavirus outbreaks much further in advance. A non-peer-reviewed study, posted to the preprint server medRxiv in April, reported that one anomalous fever case might correspond to as many as 14 future confirmed cases of the novel coronavirus.
When Alsaleem compared the historical fever data of Nebraska with the emergence of fevers in mid-March, he likewise saw a substantial spike — one that predated the outbreak of officially reported coronavirus cases by about a month. The disparity in fevers between 2020 and prior years closely aligned with the number of coronavirus cases reported in Nebraska from mid-April to mid-May, further suggesting that the coronavirus was responsible for most of the spike.
“It’s a big thing if we can know that we have this virus almost a month before it is reported from testing,” said Alsaleem, assistant professor of architectural engineering and construction. “One quick way we could potentially use this is to forecast a new outbreak.”
With assistance from Kinsa and the Office of Research and Economic Development’s COVID-19 Rapid Response Grant Program, Alsaleem hopes to drill down into the data by factoring in the number of Kinsa thermometers sold in each state and the respective demographics of its users. Better integrating that contextual information, he believes, could help bolster the predictive power of the fever data and determine the benefits of adding more data points in the form of more thermometers. He’s also examining the state-specific lags between fever spikes and coronavirus confirmations — longer in Nebraska than New York, for instance — which Alsaleem hypothesizes are dictated mostly by the availability and forms of testing in each state.
While reviewing Nebraska’s fever data, Alsaleem had another realization. Data had been streaming in both before social distancing, when the novel coronavirus barely registered in the consciousness of many Nebraskans but may have already begun infecting them, and after, when personal space expanded to 6 feet and quarantines became routine. As he expected, the incidence of fevers in Nebraska began sharply declining when state officials announced social distancing guidelines, schools shifted to remote instruction, and some employers began allowing employees to work from home.
Alsaleem said the trajectory of that decline offers a much-needed empirical perspective on the effectiveness of social distancing — and could help preview the outcomes of relaxing such guidelines. In tandem with Basheer Qolomany, who researches machine learning and big data at UNK, and Alison Freifeld, professor of contagious diseases at UNMC, Alsaleem is incorporating that data into a model aimed at projecting how infection rates will respond in Nebraska and elsewhere.
“There are a lot of models out there now trying to predict the impact of removing social distancing,” said Alsaleem, who is also seeking grant support from the National Institutes of Health. “Many of them are not based in much data. But this one will be, because we have data on (fever) cases with social distancing and without.
“This data can be used … to predict the impact of social distancing, which can then be used as a guideline for how much to relax and when we get to relax or have to go back to distancing.”
Alsaleem and Qolomany are even looking into whether Twitter mentions of the word “fever,” which appeared to spike with roughly the same magnitude and advance warning as the fever data itself, could further refine the model. Integrating the data on bike-riding frequency and out-of-state riders collected during two recent Nebraska Department of Transportation studies — data that also appears responsive to the social distancing guidelines — might prove useful, too.
“Thermometer data will never give you 100% accuracy,” Alsaleem said. “Twitter, by itself, will never give you 100% accuracy. But the more you bring these leading indicators together, the stronger your signal.”