The Pentagon is working on an algorithm to detect Covid early

What if a fitness tracker could predict that the wearer was Covid-positive hours or even days before they start noticing symptoms?

To answer the question, the Pentagon has spent the past two years experimenting with “predictive bio-wearables” technology — in this case, a wristwatch and a ring on the user’s finger. The idea is to wear the combo daily just like a fitness tracker, and once users notice a change, they get tested for Covid-19.

And after some promising early results, the agency is ready to move out of the research stage and expand the number of testers.

The project hit a new phase on Wednesday when the Pentagon’s Defense Innovation Unit awarded a contract to Philips Healthcare, which is leading the health monitoring project. The contract award allows DIU to continue its work to advance the algorithm by tracking another 200 users.

The new award effectively moves the project out of the prototyping phase to where other government agencies can acquire the technology.

DIU awarded the contract to Philips using “other transaction authority,” a process that allows the Pentagon to move outside the traditional acquisition system. Under the authority, DoD and the vendor aren’t required to publicize details of the contract award. A spokesperson for DIU declined to disclose the contract amount. Philips did not respond to a request for comment.

The award follows an earlier $12 million contract in 2020, which was received through the Coronavirus, Aid, Relief and Economic Security Act. DIU, which helps the Pentagon field and grow commercial technology, has worked on this program in partnership with Phillips Healthcare, Texas A&M and the Defense Advanced Research Projects Agency.

The team developed an algorithm that uses data gathered by the watch and ring and predicts whether a user has Covid-19 two days before they are showing symptoms. The technology is akin to a “check engine light” in a car that alerts the user that something might be off, but does not pinpoint what is wrong, said Jeff Schneider, DIU rapid analysis threat exposure program manager.

DIU tracked 11,500 users from June 2020 to September 2021. The final results of the experiment won’t be released until they have been peer-reviewed, Schneider said. However, preliminary results published in August show the algorithm’s projected accuracy is about 82 percent with a false positive rate of 11 percent over a 14-day period.

The test group is using the Garmin Fēnix 6 watch and an Oura ring, which tracks sleep and activity, but officials want participants to eventually use whatever health monitoring device they want, Schneider said.

The watch tracks respiration rate, oxygen saturation and heart rate, while the ring collects heart rate and skin temperature, he said.

“We take those features, and we derive something like 160 features from those, and that is where the secret sauce is with this,” Schneider said. The algorithm inputs the data and calculates a score for the user.

Scores range from 1 through 100 — the higher the number, the higher the likelihood of infection.

“These things that this algorithm is sensing are imperceptible. You wake up the next morning and see your score jump to a 20, but you feel fine,” he said.

In fact, Schneider said he woke up on Jan. 21 with a score of 6 and two days later was showing symptoms of the common cold.

“What does that mean? It means I can isolate, it means I’m probably not going to drink that glass of wine, it means that I’m going to hydrate and get ahead of this and try not to get everyone else sick,” Schneider said.

Units may decide to keep people at home who see a single-digit rise in score to thwart spreading Covid-19, Schneider said. The experiment observed higher scores, up to the 70s, for people who contracted Covid-19. Receiving the vaccine also increases the score.

Now out of the research phase, the program will track 200 users for one year, Schneider said. The number of users has room to grow as the server allows for up to 100,000 subjects.

“We would love to be able to scale this to 11,500 [users], but we don’t have the money for that,” Schneider said. “Right now, we are targeting the high-end frontline supervisors … and also DIU members.”

DIU is using the technology to enforce its office’s Covid guidelines, Schneider added.

During the research phase, DIU offered the technology to different military organizations, such as U.S. Northern Command and North American Defense Aerospace Command. The team told users not to rely on the gear 100 percent and instead use it as a supplemental tool because it was still in the research phase.

The score alerted users — who were otherwise presymptomatic — to get tested. “We were able to prevent the spread. While we’re in research that’s very unique,” Schneider said. During the research experiment, there were 491 Covid-positive cases out of 9,381 people who wore it.

The algorithm detected 73 percent of those cases and identified individuals 2.3 days before diagnostic testing, Schneider said.

Schneider hopes the technology will transition to the Defense Threat Reduction Agency, an office that defends the U.S. against emerging threats.

DIU is not the only organization working on predictive infectious disease technology. The Mount Sinai Health Systemconducted a study using Apple watches to predict Covid-19, and the Yale School of Public Healthdeveloped a clip to alert health care workers that they were exposed to Covid-19.

DIU is in talks with both the Defense Health Agency and the Department of Health and Human Services to gauge interest and eventually gain more users.

By working with the Defense Health Agency, Schneider’s team hopes to attract users from DoD families. If a family member becomes Covid-19 positive, that person can isolate and the service member does not spread the virus to their unit.

DIU did not validate the algorithm until last year, and Schneider admits he’s been hesitant to promote the program because it sounds like “spooky science.”

“We had to prove ourselves that this works even though we had two years of data that show it worked in the clinical world, we still had to prove it to ourselves and the public-at-large,” Schneider said.