TemPredict study demonstrates distributed, passive fever detection, and generalized sickness detection potential with live trial
New Bioengineering professor leads effort to detect COVID with wearable sensors
“I still don’t have an office on campus” Dr. Smarr laughs. “True story: I started at UC San Diego in the first week of March, 2020. I showed up and I said ‘where do I put my stuff.’ They told me they still needed to clean out the office I would use, so come back next week. Next week, everything shut down, so I still don’t have an office.” Benjamin Smarr was coming from Berkeley, where he’d been an NIH K99 fellow studying circadian rhythms and reproductive neuroendocrinology in rodent models. Luckily, as it would turn out, he had been transitioning from wet lab work to data science, using implanted sensors to capture physiological time series from his experimental subjects, and developing methods for extracting patterns in these time series that could predict things like pregnancy outcomes or hormonal disruptions. When COVID-19 hit, and he couldn’t find space on campus, it pushed him to complete his conversion to digital science.
Smarr had consulted to numerous wearables and sleep start-ups, looking for ways to move his analyses from rodents into the real world. So when the pandemic started to take off in the US, the CEO or Oura Ring Inc. reached out to Smarr and a clinical collaborator at UCSD, Dr. Ashley Mason, and said “we don’t want to ride this out. We want to help.” The result was TemPredict, a collaboration started in March 2020 as an attempt to rapidly develop COVID-19 detection capabilities for real-world use against the pandemic, with Dr. Smarr playing the role of technical lead and PI at UCSD.
Collaborations to the rescue
“I was very excited. Here I’m starting a new lab, and I get to pivot hard to human-focused work with a real urgency to it. As people started sharing their data, it was clear we could see the illness, so I was optimistic.” Says Smarr. “At the same time, people are dying. I just started. It was very daunting.” TemPredict quickly expanded beyond initial expectations, with tens of thousands of Oura users joining over the spring and summer or 2020, and Smarr reached out to long-time colleague Dr. Ilkay Altintas at the San Diego Supercomputer Center for help. TemPredict eventually received ~$7 million in DoD funding to support the effort of algorithm development and a live trial. UCSD used some of these funds to build new cyberinfrastructure making it possible to quickly explore the tidal wave of data coming in. Using labels from daily and monthly surveys paired to the wearable device data, Smarr and Altintas also collaborated with colleagues at the MIT Lincoln Lab, who brought experience with illness detection.
Their first paper came out in the fall of 2020, showing that fevers could be detected, and often predicted, using Smarr’s time series analysis methods. The algorithm went into live trial at the start of 2021, and data are still being analyzed. Smarr remains optimistic: “the people that have come together to help in this effort are amazing. And we have so much more to learn from these data.” For instance, he points out that several thousand participants got vaccinated during the trial, opening the potential to analyze physiological responses and correlations to eventual antibody titers.
Just the beginning
TemPredict also brought Smarr into contact with organizations across the world interested in how this kind of digital effort might inform future public health efforts. Several similar efforts point at the same conclusion – public health could make much greater use of wearable sensor technology than it currently does. Faster detection of illness, especially across distributed populations, could have a had a dramatic effect in controlling COVID-19’s spread. Smarr remains optimistic following his roller-coaster first year at UCSD: “I think the community is starting to see that we really could build real-time illness detection maps akin to ‘weather radar’ for sickness. There’s a lot of hard problems between here and there though, issues of privacy, access and bias in analyses, and social and clinical acceptance. But we have to try to tackle these issues. We’re not just waiting for the next pandemic. People have everyday healthcare needs. We can’t let complexity discourage these larger efforts.”
Dr. Smarr now has an office to continue his work at UC San Diego.